diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-21 11:54:28 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-21 11:54:28 +0000 |
commit | e6918187568dbd01842d8d1d2c808ce16a894239 (patch) | |
tree | 64f88b554b444a49f656b6c656111a145cbbaa28 /src/rocksdb/db_stress_tool | |
parent | Initial commit. (diff) | |
download | ceph-b26c4052f3542036551aa9dec9caa4226e456195.tar.xz ceph-b26c4052f3542036551aa9dec9caa4226e456195.zip |
Adding upstream version 18.2.2.upstream/18.2.2
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
26 files changed, 13729 insertions, 0 deletions
diff --git a/src/rocksdb/db_stress_tool/CMakeLists.txt b/src/rocksdb/db_stress_tool/CMakeLists.txt new file mode 100644 index 000000000..96d70dd0e --- /dev/null +++ b/src/rocksdb/db_stress_tool/CMakeLists.txt @@ -0,0 +1,17 @@ +add_executable(db_stress${ARTIFACT_SUFFIX} + batched_ops_stress.cc + cf_consistency_stress.cc + db_stress.cc + db_stress_common.cc + db_stress_driver.cc + db_stress_gflags.cc + db_stress_listener.cc + db_stress_shared_state.cc + db_stress_stat.cc + db_stress_test_base.cc + db_stress_tool.cc + expected_state.cc + multi_ops_txns_stress.cc + no_batched_ops_stress.cc) +target_link_libraries(db_stress${ARTIFACT_SUFFIX} ${ROCKSDB_LIB} ${THIRDPARTY_LIBS}) +list(APPEND tool_deps db_stress) diff --git a/src/rocksdb/db_stress_tool/batched_ops_stress.cc b/src/rocksdb/db_stress_tool/batched_ops_stress.cc new file mode 100644 index 000000000..3f3446076 --- /dev/null +++ b/src/rocksdb/db_stress_tool/batched_ops_stress.cc @@ -0,0 +1,399 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" + +namespace ROCKSDB_NAMESPACE { +class BatchedOpsStressTest : public StressTest { + public: + BatchedOpsStressTest() {} + virtual ~BatchedOpsStressTest() {} + + bool IsStateTracked() const override { return false; } + + // Given a key K and value V, this puts ("0"+K, V+"0"), ("1"+K, V+"1"), ..., + // ("9"+K, V+"9") in DB atomically i.e in a single batch. + // Also refer BatchedOpsStressTest::TestGet + Status TestPut(ThreadState* thread, WriteOptions& write_opts, + const ReadOptions& /* read_opts */, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys, + char (&value)[100]) override { + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + const std::string key_body = Key(rand_keys[0]); + + const uint32_t value_base = + thread->rand.Next() % thread->shared->UNKNOWN_SENTINEL; + const size_t sz = GenerateValue(value_base, value, sizeof(value)); + const std::string value_body = Slice(value, sz).ToString(); + + WriteBatch batch(0 /* reserved_bytes */, 0 /* max_bytes */, + FLAGS_batch_protection_bytes_per_key, + FLAGS_user_timestamp_size); + + ColumnFamilyHandle* const cfh = column_families_[rand_column_families[0]]; + assert(cfh); + + for (int i = 9; i >= 0; --i) { + const std::string num = std::to_string(i); + + // Note: the digit in num is prepended to the key; however, it is appended + // to the value because we want the "value base" to be encoded uniformly + // at the beginning of the value for all types of stress tests (e.g. + // batched, non-batched, CF consistency). + const std::string k = num + key_body; + const std::string v = value_body + num; + + if (FLAGS_use_merge) { + batch.Merge(cfh, k, v); + } else if (FLAGS_use_put_entity_one_in > 0 && + (value_base % FLAGS_use_put_entity_one_in) == 0) { + batch.PutEntity(cfh, k, GenerateWideColumns(value_base, v)); + } else { + batch.Put(cfh, k, v); + } + } + + const Status s = db_->Write(write_opts, &batch); + + if (!s.ok()) { + fprintf(stderr, "multiput error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } else { + // we did 10 writes each of size sz + 1 + thread->stats.AddBytesForWrites(10, (sz + 1) * 10); + } + + return s; + } + + // Given a key K, this deletes ("0"+K), ("1"+K), ..., ("9"+K) + // in DB atomically i.e in a single batch. Also refer MultiGet. + Status TestDelete(ThreadState* thread, WriteOptions& writeoptions, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + std::string keys[10] = {"9", "7", "5", "3", "1", "8", "6", "4", "2", "0"}; + + WriteBatch batch(0 /* reserved_bytes */, 0 /* max_bytes */, + FLAGS_batch_protection_bytes_per_key, + FLAGS_user_timestamp_size); + Status s; + auto cfh = column_families_[rand_column_families[0]]; + std::string key_str = Key(rand_keys[0]); + for (int i = 0; i < 10; i++) { + keys[i] += key_str; + batch.Delete(cfh, keys[i]); + } + + s = db_->Write(writeoptions, &batch); + if (!s.ok()) { + fprintf(stderr, "multidelete error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } else { + thread->stats.AddDeletes(10); + } + + return s; + } + + Status TestDeleteRange(ThreadState* /* thread */, + WriteOptions& /* write_opts */, + const std::vector<int>& /* rand_column_families */, + const std::vector<int64_t>& /* rand_keys */) override { + assert(false); + return Status::NotSupported( + "BatchedOpsStressTest does not support " + "TestDeleteRange"); + } + + void TestIngestExternalFile( + ThreadState* /* thread */, + const std::vector<int>& /* rand_column_families */, + const std::vector<int64_t>& /* rand_keys */) override { + assert(false); + fprintf(stderr, + "BatchedOpsStressTest does not support " + "TestIngestExternalFile\n"); + std::terminate(); + } + + // Given a key K, this gets values for "0"+K, "1"+K, ..., "9"+K + // in the same snapshot, and verifies that all the values are of the form + // V+"0", V+"1", ..., V+"9". + // ASSUMES that BatchedOpsStressTest::TestPut was used to put (K, V) into + // the DB. + Status TestGet(ThreadState* thread, const ReadOptions& readoptions, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + std::string keys[10] = {"0", "1", "2", "3", "4", "5", "6", "7", "8", "9"}; + Slice key_slices[10]; + std::string values[10]; + ReadOptions readoptionscopy = readoptions; + readoptionscopy.snapshot = db_->GetSnapshot(); + std::string key_str = Key(rand_keys[0]); + Slice key = key_str; + auto cfh = column_families_[rand_column_families[0]]; + std::string from_db; + Status s; + for (int i = 0; i < 10; i++) { + keys[i] += key.ToString(); + key_slices[i] = keys[i]; + s = db_->Get(readoptionscopy, cfh, key_slices[i], &from_db); + if (!s.ok() && !s.IsNotFound()) { + fprintf(stderr, "get error: %s\n", s.ToString().c_str()); + values[i] = ""; + thread->stats.AddErrors(1); + // we continue after error rather than exiting so that we can + // find more errors if any + } else if (s.IsNotFound()) { + values[i] = ""; + thread->stats.AddGets(1, 0); + } else { + values[i] = from_db; + + assert(!keys[i].empty()); + assert(!values[i].empty()); + + const char expected = keys[i].front(); + const char actual = values[i].back(); + + if (expected != actual) { + fprintf(stderr, "get error expected = %c actual = %c\n", expected, + actual); + } + + values[i].pop_back(); // get rid of the differing character + + thread->stats.AddGets(1, 1); + } + } + db_->ReleaseSnapshot(readoptionscopy.snapshot); + + // Now that we retrieved all values, check that they all match + for (int i = 1; i < 10; i++) { + if (values[i] != values[0]) { + fprintf(stderr, "get error: inconsistent values for key %s: %s, %s\n", + key.ToString(true).c_str(), StringToHex(values[0]).c_str(), + StringToHex(values[i]).c_str()); + // we continue after error rather than exiting so that we can + // find more errors if any + } + } + + return s; + } + + std::vector<Status> TestMultiGet( + ThreadState* thread, const ReadOptions& readoptions, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + size_t num_keys = rand_keys.size(); + std::vector<Status> ret_status(num_keys); + std::array<std::string, 10> keys = { + {"0", "1", "2", "3", "4", "5", "6", "7", "8", "9"}}; + size_t num_prefixes = keys.size(); + for (size_t rand_key = 0; rand_key < num_keys; ++rand_key) { + std::vector<Slice> key_slices; + std::vector<PinnableSlice> values(num_prefixes); + std::vector<Status> statuses(num_prefixes); + ReadOptions readoptionscopy = readoptions; + readoptionscopy.snapshot = db_->GetSnapshot(); + readoptionscopy.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + std::vector<std::string> key_str; + key_str.reserve(num_prefixes); + key_slices.reserve(num_prefixes); + std::string from_db; + ColumnFamilyHandle* cfh = column_families_[rand_column_families[0]]; + + for (size_t key = 0; key < num_prefixes; ++key) { + key_str.emplace_back(keys[key] + Key(rand_keys[rand_key])); + key_slices.emplace_back(key_str.back()); + } + db_->MultiGet(readoptionscopy, cfh, num_prefixes, key_slices.data(), + values.data(), statuses.data()); + for (size_t i = 0; i < num_prefixes; i++) { + Status s = statuses[i]; + if (!s.ok() && !s.IsNotFound()) { + fprintf(stderr, "multiget error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + ret_status[rand_key] = s; + // we continue after error rather than exiting so that we can + // find more errors if any + } else if (s.IsNotFound()) { + thread->stats.AddGets(1, 0); + ret_status[rand_key] = s; + } else { + assert(!keys[i].empty()); + assert(!values[i].empty()); + + const char expected = keys[i][0]; + const char actual = values[i][values[i].size() - 1]; + + if (expected != actual) { + fprintf(stderr, "multiget error expected = %c actual = %c\n", + expected, actual); + } + + values[i].remove_suffix(1); // get rid of the differing character + + thread->stats.AddGets(1, 1); + } + } + db_->ReleaseSnapshot(readoptionscopy.snapshot); + + // Now that we retrieved all values, check that they all match + for (size_t i = 1; i < num_prefixes; i++) { + if (values[i] != values[0]) { + fprintf(stderr, + "multiget error: inconsistent values for key %s: %s, %s\n", + StringToHex(key_str[i]).c_str(), + StringToHex(values[0].ToString()).c_str(), + StringToHex(values[i].ToString()).c_str()); + // we continue after error rather than exiting so that we can + // find more errors if any + } + } + } + + return ret_status; + } + + // Given a key, this does prefix scans for "0"+P, "1"+P, ..., "9"+P + // in the same snapshot where P is the first FLAGS_prefix_size - 1 bytes + // of the key. Each of these 10 scans returns a series of values; + // each series should be the same length, and it is verified for each + // index i that all the i'th values are of the form V+"0", V+"1", ..., V+"9". + // ASSUMES that MultiPut was used to put (K, V) + Status TestPrefixScan(ThreadState* thread, const ReadOptions& readoptions, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + const std::string key = Key(rand_keys[0]); + + assert(FLAGS_prefix_size > 0); + const size_t prefix_to_use = static_cast<size_t>(FLAGS_prefix_size); + + constexpr size_t num_prefixes = 10; + + std::array<std::string, num_prefixes> prefixes; + std::array<Slice, num_prefixes> prefix_slices; + std::array<ReadOptions, num_prefixes> ro_copies; + std::array<std::string, num_prefixes> upper_bounds; + std::array<Slice, num_prefixes> ub_slices; + std::array<std::unique_ptr<Iterator>, num_prefixes> iters; + + const Snapshot* const snapshot = db_->GetSnapshot(); + + ColumnFamilyHandle* const cfh = column_families_[rand_column_families[0]]; + assert(cfh); + + for (size_t i = 0; i < num_prefixes; ++i) { + prefixes[i] = std::to_string(i) + key; + prefix_slices[i] = Slice(prefixes[i].data(), prefix_to_use); + + ro_copies[i] = readoptions; + ro_copies[i].snapshot = snapshot; + if (thread->rand.OneIn(2) && + GetNextPrefix(prefix_slices[i], &(upper_bounds[i]))) { + // For half of the time, set the upper bound to the next prefix + ub_slices[i] = upper_bounds[i]; + ro_copies[i].iterate_upper_bound = &(ub_slices[i]); + } + + iters[i].reset(db_->NewIterator(ro_copies[i], cfh)); + iters[i]->Seek(prefix_slices[i]); + } + + uint64_t count = 0; + + while (iters[0]->Valid() && iters[0]->key().starts_with(prefix_slices[0])) { + ++count; + + std::array<std::string, num_prefixes> values; + + // get list of all values for this iteration + for (size_t i = 0; i < num_prefixes; ++i) { + // no iterator should finish before the first one + assert(iters[i]->Valid() && + iters[i]->key().starts_with(prefix_slices[i])); + values[i] = iters[i]->value().ToString(); + + // make sure the last character of the value is the expected digit + assert(!prefixes[i].empty()); + assert(!values[i].empty()); + + const char expected = prefixes[i].front(); + const char actual = values[i].back(); + + if (expected != actual) { + fprintf(stderr, "prefix scan error expected = %c actual = %c\n", + expected, actual); + } + + values[i].pop_back(); // get rid of the differing character + + // make sure all values are equivalent + if (values[i] != values[0]) { + fprintf(stderr, + "prefix scan error : %" ROCKSDB_PRIszt + ", inconsistent values for prefix %s: %s, %s\n", + i, prefix_slices[i].ToString(/* hex */ true).c_str(), + StringToHex(values[0]).c_str(), + StringToHex(values[i]).c_str()); + // we continue after error rather than exiting so that we can + // find more errors if any + } + + // make sure value() and columns() are consistent + const WideColumns expected_columns = GenerateExpectedWideColumns( + GetValueBase(iters[i]->value()), iters[i]->value()); + if (iters[i]->columns() != expected_columns) { + fprintf(stderr, + "prefix scan error : %" ROCKSDB_PRIszt + ", value and columns inconsistent for prefix %s: %s\n", + i, prefix_slices[i].ToString(/* hex */ true).c_str(), + DebugString(iters[i]->value(), iters[i]->columns(), + expected_columns) + .c_str()); + } + + iters[i]->Next(); + } + } + + // cleanup iterators and snapshot + for (size_t i = 0; i < num_prefixes; ++i) { + // if the first iterator finished, they should have all finished + assert(!iters[i]->Valid() || + !iters[i]->key().starts_with(prefix_slices[i])); + assert(iters[i]->status().ok()); + } + + db_->ReleaseSnapshot(snapshot); + + thread->stats.AddPrefixes(1, count); + + return Status::OK(); + } + + void VerifyDb(ThreadState* /* thread */) const override {} + + void ContinuouslyVerifyDb(ThreadState* /* thread */) const override {} +}; + +StressTest* CreateBatchedOpsStressTest() { return new BatchedOpsStressTest(); } + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/cf_consistency_stress.cc b/src/rocksdb/db_stress_tool/cf_consistency_stress.cc new file mode 100644 index 000000000..33f7b1f2e --- /dev/null +++ b/src/rocksdb/db_stress_tool/cf_consistency_stress.cc @@ -0,0 +1,640 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" +#include "file/file_util.h" + +namespace ROCKSDB_NAMESPACE { +class CfConsistencyStressTest : public StressTest { + public: + CfConsistencyStressTest() : batch_id_(0) {} + + ~CfConsistencyStressTest() override {} + + bool IsStateTracked() const override { return false; } + + Status TestPut(ThreadState* thread, WriteOptions& write_opts, + const ReadOptions& /* read_opts */, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys, + char (&value)[100]) override { + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + const std::string k = Key(rand_keys[0]); + + const uint32_t value_base = batch_id_.fetch_add(1); + const size_t sz = GenerateValue(value_base, value, sizeof(value)); + const Slice v(value, sz); + + WriteBatch batch; + + const bool use_put_entity = !FLAGS_use_merge && + FLAGS_use_put_entity_one_in > 0 && + (value_base % FLAGS_use_put_entity_one_in) == 0; + + for (auto cf : rand_column_families) { + ColumnFamilyHandle* const cfh = column_families_[cf]; + assert(cfh); + + if (FLAGS_use_merge) { + batch.Merge(cfh, k, v); + } else if (use_put_entity) { + batch.PutEntity(cfh, k, GenerateWideColumns(value_base, v)); + } else { + batch.Put(cfh, k, v); + } + } + + Status s = db_->Write(write_opts, &batch); + + if (!s.ok()) { + fprintf(stderr, "multi put or merge error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } else { + auto num = static_cast<long>(rand_column_families.size()); + thread->stats.AddBytesForWrites(num, (sz + 1) * num); + } + + return s; + } + + Status TestDelete(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + std::string key_str = Key(rand_keys[0]); + Slice key = key_str; + WriteBatch batch; + for (auto cf : rand_column_families) { + ColumnFamilyHandle* cfh = column_families_[cf]; + batch.Delete(cfh, key); + } + Status s = db_->Write(write_opts, &batch); + if (!s.ok()) { + fprintf(stderr, "multidel error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } else { + thread->stats.AddDeletes(static_cast<long>(rand_column_families.size())); + } + return s; + } + + Status TestDeleteRange(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + int64_t rand_key = rand_keys[0]; + auto shared = thread->shared; + int64_t max_key = shared->GetMaxKey(); + if (rand_key > max_key - FLAGS_range_deletion_width) { + rand_key = + thread->rand.Next() % (max_key - FLAGS_range_deletion_width + 1); + } + std::string key_str = Key(rand_key); + Slice key = key_str; + std::string end_key_str = Key(rand_key + FLAGS_range_deletion_width); + Slice end_key = end_key_str; + WriteBatch batch; + for (auto cf : rand_column_families) { + ColumnFamilyHandle* cfh = column_families_[rand_column_families[cf]]; + batch.DeleteRange(cfh, key, end_key); + } + Status s = db_->Write(write_opts, &batch); + if (!s.ok()) { + fprintf(stderr, "multi del range error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } else { + thread->stats.AddRangeDeletions( + static_cast<long>(rand_column_families.size())); + } + return s; + } + + void TestIngestExternalFile( + ThreadState* /* thread */, + const std::vector<int>& /* rand_column_families */, + const std::vector<int64_t>& /* rand_keys */) override { + assert(false); + fprintf(stderr, + "CfConsistencyStressTest does not support TestIngestExternalFile " + "because it's not possible to verify the result\n"); + std::terminate(); + } + + Status TestGet(ThreadState* thread, const ReadOptions& readoptions, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + std::string key_str = Key(rand_keys[0]); + Slice key = key_str; + Status s; + bool is_consistent = true; + + if (thread->rand.OneIn(2)) { + // 1/2 chance, does a random read from random CF + auto cfh = + column_families_[rand_column_families[thread->rand.Next() % + rand_column_families.size()]]; + std::string from_db; + s = db_->Get(readoptions, cfh, key, &from_db); + } else { + // 1/2 chance, comparing one key is the same across all CFs + const Snapshot* snapshot = db_->GetSnapshot(); + ReadOptions readoptionscopy = readoptions; + readoptionscopy.snapshot = snapshot; + + std::string value0; + s = db_->Get(readoptionscopy, column_families_[rand_column_families[0]], + key, &value0); + if (s.ok() || s.IsNotFound()) { + bool found = s.ok(); + for (size_t i = 1; i < rand_column_families.size(); i++) { + std::string value1; + s = db_->Get(readoptionscopy, + column_families_[rand_column_families[i]], key, &value1); + if (!s.ok() && !s.IsNotFound()) { + break; + } + if (!found && s.ok()) { + fprintf(stderr, "Get() return different results with key %s\n", + Slice(key_str).ToString(true).c_str()); + fprintf(stderr, "CF %s is not found\n", + column_family_names_[0].c_str()); + fprintf(stderr, "CF %s returns value %s\n", + column_family_names_[i].c_str(), + Slice(value1).ToString(true).c_str()); + is_consistent = false; + } else if (found && s.IsNotFound()) { + fprintf(stderr, "Get() return different results with key %s\n", + Slice(key_str).ToString(true).c_str()); + fprintf(stderr, "CF %s returns value %s\n", + column_family_names_[0].c_str(), + Slice(value0).ToString(true).c_str()); + fprintf(stderr, "CF %s is not found\n", + column_family_names_[i].c_str()); + is_consistent = false; + } else if (s.ok() && value0 != value1) { + fprintf(stderr, "Get() return different results with key %s\n", + Slice(key_str).ToString(true).c_str()); + fprintf(stderr, "CF %s returns value %s\n", + column_family_names_[0].c_str(), + Slice(value0).ToString(true).c_str()); + fprintf(stderr, "CF %s returns value %s\n", + column_family_names_[i].c_str(), + Slice(value1).ToString(true).c_str()); + is_consistent = false; + } + if (!is_consistent) { + break; + } + } + } + + db_->ReleaseSnapshot(snapshot); + } + if (!is_consistent) { + fprintf(stderr, "TestGet error: is_consistent is false\n"); + thread->stats.AddErrors(1); + // Fail fast to preserve the DB state. + thread->shared->SetVerificationFailure(); + } else if (s.ok()) { + thread->stats.AddGets(1, 1); + } else if (s.IsNotFound()) { + thread->stats.AddGets(1, 0); + } else { + fprintf(stderr, "TestGet error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } + return s; + } + + std::vector<Status> TestMultiGet( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + size_t num_keys = rand_keys.size(); + std::vector<std::string> key_str; + std::vector<Slice> keys; + keys.reserve(num_keys); + key_str.reserve(num_keys); + std::vector<PinnableSlice> values(num_keys); + std::vector<Status> statuses(num_keys); + ColumnFamilyHandle* cfh = column_families_[rand_column_families[0]]; + ReadOptions readoptionscopy = read_opts; + readoptionscopy.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + + for (size_t i = 0; i < num_keys; ++i) { + key_str.emplace_back(Key(rand_keys[i])); + keys.emplace_back(key_str.back()); + } + db_->MultiGet(readoptionscopy, cfh, num_keys, keys.data(), values.data(), + statuses.data()); + for (auto s : statuses) { + if (s.ok()) { + // found case + thread->stats.AddGets(1, 1); + } else if (s.IsNotFound()) { + // not found case + thread->stats.AddGets(1, 0); + } else { + // errors case + fprintf(stderr, "MultiGet error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } + } + return statuses; + } + + Status TestPrefixScan(ThreadState* thread, const ReadOptions& readoptions, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + const std::string key = Key(rand_keys[0]); + + const size_t prefix_to_use = + (FLAGS_prefix_size < 0) ? 7 : static_cast<size_t>(FLAGS_prefix_size); + + const Slice prefix(key.data(), prefix_to_use); + + std::string upper_bound; + Slice ub_slice; + + ReadOptions ro_copy = readoptions; + + // Get the next prefix first and then see if we want to set upper bound. + // We'll use the next prefix in an assertion later on + if (GetNextPrefix(prefix, &upper_bound) && thread->rand.OneIn(2)) { + ub_slice = Slice(upper_bound); + ro_copy.iterate_upper_bound = &ub_slice; + } + + ColumnFamilyHandle* const cfh = + column_families_[rand_column_families[thread->rand.Uniform( + static_cast<int>(rand_column_families.size()))]]; + assert(cfh); + + std::unique_ptr<Iterator> iter(db_->NewIterator(ro_copy, cfh)); + + uint64_t count = 0; + Status s; + + for (iter->Seek(prefix); iter->Valid() && iter->key().starts_with(prefix); + iter->Next()) { + ++count; + + const WideColumns expected_columns = GenerateExpectedWideColumns( + GetValueBase(iter->value()), iter->value()); + if (iter->columns() != expected_columns) { + s = Status::Corruption( + "Value and columns inconsistent", + DebugString(iter->value(), iter->columns(), expected_columns)); + break; + } + } + + assert(prefix_to_use == 0 || + count <= GetPrefixKeyCount(prefix.ToString(), upper_bound)); + + if (s.ok()) { + s = iter->status(); + } + + if (!s.ok()) { + fprintf(stderr, "TestPrefixScan error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + + return s; + } + + thread->stats.AddPrefixes(1, count); + + return Status::OK(); + } + + ColumnFamilyHandle* GetControlCfh(ThreadState* thread, + int /*column_family_id*/ + ) override { + // All column families should contain the same data. Randomly pick one. + return column_families_[thread->rand.Next() % column_families_.size()]; + } + + void VerifyDb(ThreadState* thread) const override { + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions options(FLAGS_verify_checksum, true); + + // We must set total_order_seek to true because we are doing a SeekToFirst + // on a column family whose memtables may support (by default) prefix-based + // iterator. In this case, NewIterator with options.total_order_seek being + // false returns a prefix-based iterator. Calling SeekToFirst using this + // iterator causes the iterator to become invalid. That means we cannot + // iterate the memtable using this iterator any more, although the memtable + // contains the most up-to-date key-values. + options.total_order_seek = true; + + ManagedSnapshot snapshot_guard(db_); + options.snapshot = snapshot_guard.snapshot(); + + const size_t num = column_families_.size(); + + std::vector<std::unique_ptr<Iterator>> iters; + iters.reserve(num); + + for (size_t i = 0; i < num; ++i) { + iters.emplace_back(db_->NewIterator(options, column_families_[i])); + iters.back()->SeekToFirst(); + } + + std::vector<Status> statuses(num, Status::OK()); + + assert(thread); + + auto shared = thread->shared; + assert(shared); + + do { + if (shared->HasVerificationFailedYet()) { + break; + } + + size_t valid_cnt = 0; + + for (size_t i = 0; i < num; ++i) { + const auto& iter = iters[i]; + assert(iter); + + if (iter->Valid()) { + const WideColumns expected_columns = GenerateExpectedWideColumns( + GetValueBase(iter->value()), iter->value()); + if (iter->columns() != expected_columns) { + statuses[i] = Status::Corruption( + "Value and columns inconsistent", + DebugString(iter->value(), iter->columns(), expected_columns)); + } else { + ++valid_cnt; + } + } else { + statuses[i] = iter->status(); + } + } + + if (valid_cnt == 0) { + for (size_t i = 0; i < num; ++i) { + const auto& s = statuses[i]; + if (!s.ok()) { + fprintf(stderr, "Iterator on cf %s has error: %s\n", + column_families_[i]->GetName().c_str(), + s.ToString().c_str()); + shared->SetVerificationFailure(); + } + } + + break; + } + + if (valid_cnt < num) { + shared->SetVerificationFailure(); + + for (size_t i = 0; i < num; ++i) { + assert(iters[i]); + + if (!iters[i]->Valid()) { + if (statuses[i].ok()) { + fprintf(stderr, "Finished scanning cf %s\n", + column_families_[i]->GetName().c_str()); + } else { + fprintf(stderr, "Iterator on cf %s has error: %s\n", + column_families_[i]->GetName().c_str(), + statuses[i].ToString().c_str()); + } + } else { + fprintf(stderr, "cf %s has remaining data to scan\n", + column_families_[i]->GetName().c_str()); + } + } + + break; + } + + if (shared->HasVerificationFailedYet()) { + break; + } + + // If the program reaches here, then all column families' iterators are + // still valid. + assert(valid_cnt == num); + + if (shared->PrintingVerificationResults()) { + continue; + } + + assert(iters[0]); + + const Slice key = iters[0]->key(); + const Slice value = iters[0]->value(); + + int num_mismatched_cfs = 0; + + for (size_t i = 1; i < num; ++i) { + assert(iters[i]); + + const int cmp = key.compare(iters[i]->key()); + + if (cmp != 0) { + ++num_mismatched_cfs; + + if (1 == num_mismatched_cfs) { + fprintf(stderr, "Verification failed\n"); + fprintf(stderr, "Latest Sequence Number: %" PRIu64 "\n", + db_->GetLatestSequenceNumber()); + fprintf(stderr, "[%s] %s => %s\n", + column_families_[0]->GetName().c_str(), + key.ToString(true /* hex */).c_str(), + value.ToString(true /* hex */).c_str()); + } + + fprintf(stderr, "[%s] %s => %s\n", + column_families_[i]->GetName().c_str(), + iters[i]->key().ToString(true /* hex */).c_str(), + iters[i]->value().ToString(true /* hex */).c_str()); + +#ifndef ROCKSDB_LITE + Slice begin_key; + Slice end_key; + if (cmp < 0) { + begin_key = key; + end_key = iters[i]->key(); + } else { + begin_key = iters[i]->key(); + end_key = key; + } + + const auto print_key_versions = [&](ColumnFamilyHandle* cfh) { + constexpr size_t kMaxNumIKeys = 8; + + std::vector<KeyVersion> versions; + const Status s = GetAllKeyVersions(db_, cfh, begin_key, end_key, + kMaxNumIKeys, &versions); + if (!s.ok()) { + fprintf(stderr, "%s\n", s.ToString().c_str()); + return; + } + + assert(cfh); + + fprintf(stderr, + "Internal keys in CF '%s', [%s, %s] (max %" ROCKSDB_PRIszt + ")\n", + cfh->GetName().c_str(), + begin_key.ToString(true /* hex */).c_str(), + end_key.ToString(true /* hex */).c_str(), kMaxNumIKeys); + + for (const KeyVersion& kv : versions) { + fprintf(stderr, " key %s seq %" PRIu64 " type %d\n", + Slice(kv.user_key).ToString(true).c_str(), kv.sequence, + kv.type); + } + }; + + if (1 == num_mismatched_cfs) { + print_key_versions(column_families_[0]); + } + + print_key_versions(column_families_[i]); +#endif // ROCKSDB_LITE + + shared->SetVerificationFailure(); + } + } + + shared->FinishPrintingVerificationResults(); + + for (auto& iter : iters) { + assert(iter); + iter->Next(); + } + } while (true); + } + +#ifndef ROCKSDB_LITE + void ContinuouslyVerifyDb(ThreadState* thread) const override { + assert(thread); + Status status; + + DB* db_ptr = cmp_db_ ? cmp_db_ : db_; + const auto& cfhs = cmp_db_ ? cmp_cfhs_ : column_families_; + + // Take a snapshot to preserve the state of primary db. + ManagedSnapshot snapshot_guard(db_); + + SharedState* shared = thread->shared; + assert(shared); + + if (cmp_db_) { + status = cmp_db_->TryCatchUpWithPrimary(); + if (!status.ok()) { + fprintf(stderr, "TryCatchUpWithPrimary: %s\n", + status.ToString().c_str()); + shared->SetShouldStopTest(); + assert(false); + return; + } + } + + const auto checksum_column_family = [](Iterator* iter, + uint32_t* checksum) -> Status { + assert(nullptr != checksum); + + uint32_t ret = 0; + for (iter->SeekToFirst(); iter->Valid(); iter->Next()) { + ret = crc32c::Extend(ret, iter->key().data(), iter->key().size()); + ret = crc32c::Extend(ret, iter->value().data(), iter->value().size()); + + for (const auto& column : iter->columns()) { + ret = crc32c::Extend(ret, column.name().data(), column.name().size()); + ret = + crc32c::Extend(ret, column.value().data(), column.value().size()); + } + } + + *checksum = ret; + return iter->status(); + }; + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions ropts(FLAGS_verify_checksum, true); + ropts.total_order_seek = true; + if (nullptr == cmp_db_) { + ropts.snapshot = snapshot_guard.snapshot(); + } + uint32_t crc = 0; + { + // Compute crc for all key-values of default column family. + std::unique_ptr<Iterator> it(db_ptr->NewIterator(ropts)); + status = checksum_column_family(it.get(), &crc); + if (!status.ok()) { + fprintf(stderr, "Computing checksum of default cf: %s\n", + status.ToString().c_str()); + assert(false); + } + } + // Since we currently intentionally disallow reading from the secondary + // instance with snapshot, we cannot achieve cross-cf consistency if WAL is + // enabled because there is no guarantee that secondary instance replays + // the primary's WAL to a consistent point where all cfs have the same + // data. + if (status.ok() && FLAGS_disable_wal) { + uint32_t tmp_crc = 0; + for (ColumnFamilyHandle* cfh : cfhs) { + if (cfh == db_ptr->DefaultColumnFamily()) { + continue; + } + std::unique_ptr<Iterator> it(db_ptr->NewIterator(ropts, cfh)); + status = checksum_column_family(it.get(), &tmp_crc); + if (!status.ok() || tmp_crc != crc) { + break; + } + } + if (!status.ok()) { + fprintf(stderr, "status: %s\n", status.ToString().c_str()); + shared->SetShouldStopTest(); + assert(false); + } else if (tmp_crc != crc) { + fprintf(stderr, "tmp_crc=%" PRIu32 " crc=%" PRIu32 "\n", tmp_crc, crc); + shared->SetShouldStopTest(); + assert(false); + } + } + } +#else // ROCKSDB_LITE + void ContinuouslyVerifyDb(ThreadState* /*thread*/) const override {} +#endif // !ROCKSDB_LITE + + std::vector<int> GenerateColumnFamilies( + const int /* num_column_families */, + int /* rand_column_family */) const override { + std::vector<int> ret; + int num = static_cast<int>(column_families_.size()); + int k = 0; + std::generate_n(back_inserter(ret), num, [&k]() -> int { return k++; }); + return ret; + } + + private: + std::atomic<uint32_t> batch_id_; +}; + +StressTest* CreateCfConsistencyStressTest() { + return new CfConsistencyStressTest(); +} + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress.cc b/src/rocksdb/db_stress_tool/db_stress.cc new file mode 100644 index 000000000..2d03f5d26 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress.cc @@ -0,0 +1,25 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifndef GFLAGS +#include <cstdio> + +int main() { + fprintf(stderr, "Please install gflags to run rocksdb tools\n"); + return 1; +} +#else +#include "port/stack_trace.h" +#include "rocksdb/db_stress_tool.h" + +int main(int argc, char** argv) { + ROCKSDB_NAMESPACE::port::InstallStackTraceHandler(); + return ROCKSDB_NAMESPACE::db_stress_tool(argc, argv); +} +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_common.cc b/src/rocksdb/db_stress_tool/db_stress_common.cc new file mode 100644 index 000000000..af8db9e2f --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_common.cc @@ -0,0 +1,460 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. +// + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" + +#include <cmath> + +#include "util/file_checksum_helper.h" +#include "util/xxhash.h" + +ROCKSDB_NAMESPACE::Env* db_stress_listener_env = nullptr; +ROCKSDB_NAMESPACE::Env* db_stress_env = nullptr; +// If non-null, injects read error at a rate specified by the +// read_fault_one_in or write_fault_one_in flag +std::shared_ptr<ROCKSDB_NAMESPACE::FaultInjectionTestFS> fault_fs_guard; +enum ROCKSDB_NAMESPACE::CompressionType compression_type_e = + ROCKSDB_NAMESPACE::kSnappyCompression; +enum ROCKSDB_NAMESPACE::CompressionType bottommost_compression_type_e = + ROCKSDB_NAMESPACE::kSnappyCompression; +enum ROCKSDB_NAMESPACE::ChecksumType checksum_type_e = + ROCKSDB_NAMESPACE::kCRC32c; +enum RepFactory FLAGS_rep_factory = kSkipList; +std::vector<double> sum_probs(100001); +constexpr int64_t zipf_sum_size = 100000; + +namespace ROCKSDB_NAMESPACE { + +// Zipfian distribution is generated based on a pre-calculated array. +// It should be used before start the stress test. +// First, the probability distribution function (PDF) of this Zipfian follows +// power low. P(x) = 1/(x^alpha). +// So we calculate the PDF when x is from 0 to zipf_sum_size in first for loop +// and add the PDF value togetger as c. So we get the total probability in c. +// Next, we calculate inverse CDF of Zipfian and store the value of each in +// an array (sum_probs). The rank is from 0 to zipf_sum_size. For example, for +// integer k, its Zipfian CDF value is sum_probs[k]. +// Third, when we need to get an integer whose probability follows Zipfian +// distribution, we use a rand_seed [0,1] which follows uniform distribution +// as a seed and search it in the sum_probs via binary search. When we find +// the closest sum_probs[i] of rand_seed, i is the integer that in +// [0, zipf_sum_size] following Zipfian distribution with parameter alpha. +// Finally, we can scale i to [0, max_key] scale. +// In order to avoid that hot keys are close to each other and skew towards 0, +// we use Rando64 to shuffle it. +void InitializeHotKeyGenerator(double alpha) { + double c = 0; + for (int64_t i = 1; i <= zipf_sum_size; i++) { + c = c + (1.0 / std::pow(static_cast<double>(i), alpha)); + } + c = 1.0 / c; + + sum_probs[0] = 0; + for (int64_t i = 1; i <= zipf_sum_size; i++) { + sum_probs[i] = + sum_probs[i - 1] + c / std::pow(static_cast<double>(i), alpha); + } +} + +// Generate one key that follows the Zipfian distribution. The skewness +// is decided by the parameter alpha. Input is the rand_seed [0,1] and +// the max of the key to be generated. If we directly return tmp_zipf_seed, +// the closer to 0, the higher probability will be. To randomly distribute +// the hot keys in [0, max_key], we use Random64 to shuffle it. +int64_t GetOneHotKeyID(double rand_seed, int64_t max_key) { + int64_t low = 1, mid, high = zipf_sum_size, zipf = 0; + while (low <= high) { + mid = (low + high) / 2; + if (sum_probs[mid] >= rand_seed && sum_probs[mid - 1] < rand_seed) { + zipf = mid; + break; + } else if (sum_probs[mid] >= rand_seed) { + high = mid - 1; + } else { + low = mid + 1; + } + } + int64_t tmp_zipf_seed = zipf * max_key / zipf_sum_size; + Random64 rand_local(tmp_zipf_seed); + return rand_local.Next() % max_key; +} + +void PoolSizeChangeThread(void* v) { + assert(FLAGS_compaction_thread_pool_adjust_interval > 0); + ThreadState* thread = reinterpret_cast<ThreadState*>(v); + SharedState* shared = thread->shared; + + while (true) { + { + MutexLock l(shared->GetMutex()); + if (shared->ShouldStopBgThread()) { + shared->IncBgThreadsFinished(); + if (shared->BgThreadsFinished()) { + shared->GetCondVar()->SignalAll(); + } + return; + } + } + + auto thread_pool_size_base = FLAGS_max_background_compactions; + auto thread_pool_size_var = FLAGS_compaction_thread_pool_variations; + int new_thread_pool_size = + thread_pool_size_base - thread_pool_size_var + + thread->rand.Next() % (thread_pool_size_var * 2 + 1); + if (new_thread_pool_size < 1) { + new_thread_pool_size = 1; + } + db_stress_env->SetBackgroundThreads(new_thread_pool_size, + ROCKSDB_NAMESPACE::Env::Priority::LOW); + // Sleep up to 3 seconds + db_stress_env->SleepForMicroseconds( + thread->rand.Next() % FLAGS_compaction_thread_pool_adjust_interval * + 1000 + + 1); + } +} + +void DbVerificationThread(void* v) { + assert(FLAGS_continuous_verification_interval > 0); + auto* thread = reinterpret_cast<ThreadState*>(v); + SharedState* shared = thread->shared; + StressTest* stress_test = shared->GetStressTest(); + assert(stress_test != nullptr); + while (true) { + { + MutexLock l(shared->GetMutex()); + if (shared->ShouldStopBgThread()) { + shared->IncBgThreadsFinished(); + if (shared->BgThreadsFinished()) { + shared->GetCondVar()->SignalAll(); + } + return; + } + } + if (!shared->HasVerificationFailedYet()) { + stress_test->ContinuouslyVerifyDb(thread); + } + db_stress_env->SleepForMicroseconds( + thread->rand.Next() % FLAGS_continuous_verification_interval * 1000 + + 1); + } +} + +void PrintKeyValue(int cf, uint64_t key, const char* value, size_t sz) { + if (!FLAGS_verbose) { + return; + } + std::string tmp; + tmp.reserve(sz * 2 + 16); + char buf[4]; + for (size_t i = 0; i < sz; i++) { + snprintf(buf, 4, "%X", value[i]); + tmp.append(buf); + } + auto key_str = Key(key); + Slice key_slice = key_str; + fprintf(stdout, "[CF %d] %s (%" PRIi64 ") == > (%" ROCKSDB_PRIszt ") %s\n", + cf, key_slice.ToString(true).c_str(), key, sz, tmp.c_str()); +} + +// Note that if hot_key_alpha != 0, it generates the key based on Zipfian +// distribution. Keys are randomly scattered to [0, FLAGS_max_key]. It does +// not ensure the order of the keys being generated and the keys does not have +// the active range which is related to FLAGS_active_width. +int64_t GenerateOneKey(ThreadState* thread, uint64_t iteration) { + const double completed_ratio = + static_cast<double>(iteration) / FLAGS_ops_per_thread; + const int64_t base_key = static_cast<int64_t>( + completed_ratio * (FLAGS_max_key - FLAGS_active_width)); + int64_t rand_seed = base_key + thread->rand.Next() % FLAGS_active_width; + int64_t cur_key = rand_seed; + if (FLAGS_hot_key_alpha != 0) { + // If set the Zipfian distribution Alpha to non 0, use Zipfian + double float_rand = + (static_cast<double>(thread->rand.Next() % FLAGS_max_key)) / + FLAGS_max_key; + cur_key = GetOneHotKeyID(float_rand, FLAGS_max_key); + } + return cur_key; +} + +// Note that if hot_key_alpha != 0, it generates the key based on Zipfian +// distribution. Keys being generated are in random order. +// If user want to generate keys based on uniform distribution, user needs to +// set hot_key_alpha == 0. It will generate the random keys in increasing +// order in the key array (ensure key[i] >= key[i+1]) and constrained in a +// range related to FLAGS_active_width. +std::vector<int64_t> GenerateNKeys(ThreadState* thread, int num_keys, + uint64_t iteration) { + const double completed_ratio = + static_cast<double>(iteration) / FLAGS_ops_per_thread; + const int64_t base_key = static_cast<int64_t>( + completed_ratio * (FLAGS_max_key - FLAGS_active_width)); + std::vector<int64_t> keys; + keys.reserve(num_keys); + int64_t next_key = base_key + thread->rand.Next() % FLAGS_active_width; + keys.push_back(next_key); + for (int i = 1; i < num_keys; ++i) { + // Generate the key follows zipfian distribution + if (FLAGS_hot_key_alpha != 0) { + double float_rand = + (static_cast<double>(thread->rand.Next() % FLAGS_max_key)) / + FLAGS_max_key; + next_key = GetOneHotKeyID(float_rand, FLAGS_max_key); + } else { + // This may result in some duplicate keys + next_key = next_key + thread->rand.Next() % + (FLAGS_active_width - (next_key - base_key)); + } + keys.push_back(next_key); + } + return keys; +} + +size_t GenerateValue(uint32_t rand, char* v, size_t max_sz) { + size_t value_sz = + ((rand % kRandomValueMaxFactor) + 1) * FLAGS_value_size_mult; + assert(value_sz <= max_sz && value_sz >= sizeof(uint32_t)); + (void)max_sz; + PutUnaligned(reinterpret_cast<uint32_t*>(v), rand); + for (size_t i = sizeof(uint32_t); i < value_sz; i++) { + v[i] = (char)(rand ^ i); + } + v[value_sz] = '\0'; + return value_sz; // the size of the value set. +} + +uint32_t GetValueBase(Slice s) { + assert(s.size() >= sizeof(uint32_t)); + uint32_t res; + GetUnaligned(reinterpret_cast<const uint32_t*>(s.data()), &res); + return res; +} + +WideColumns GenerateWideColumns(uint32_t value_base, const Slice& slice) { + WideColumns columns; + + constexpr size_t max_columns = 4; + const size_t num_columns = (value_base % max_columns) + 1; + + columns.reserve(num_columns); + + assert(slice.size() >= num_columns); + + columns.emplace_back(kDefaultWideColumnName, slice); + + for (size_t i = 1; i < num_columns; ++i) { + const Slice name(slice.data(), i); + const Slice value(slice.data() + i, slice.size() - i); + + columns.emplace_back(name, value); + } + + return columns; +} + +WideColumns GenerateExpectedWideColumns(uint32_t value_base, + const Slice& slice) { + if (FLAGS_use_put_entity_one_in == 0 || + (value_base % FLAGS_use_put_entity_one_in) != 0) { + return WideColumns{{kDefaultWideColumnName, slice}}; + } + + WideColumns columns = GenerateWideColumns(value_base, slice); + + std::sort(columns.begin(), columns.end(), + [](const WideColumn& lhs, const WideColumn& rhs) { + return lhs.name().compare(rhs.name()) < 0; + }); + + return columns; +} + +std::string GetNowNanos() { + uint64_t t = db_stress_env->NowNanos(); + std::string ret; + PutFixed64(&ret, t); + return ret; +} + +namespace { + +class MyXXH64Checksum : public FileChecksumGenerator { + public: + explicit MyXXH64Checksum(bool big) : big_(big) { + state_ = XXH64_createState(); + XXH64_reset(state_, 0); + } + + virtual ~MyXXH64Checksum() override { XXH64_freeState(state_); } + + void Update(const char* data, size_t n) override { + XXH64_update(state_, data, n); + } + + void Finalize() override { + assert(str_.empty()); + uint64_t digest = XXH64_digest(state_); + // Store as little endian raw bytes + PutFixed64(&str_, digest); + if (big_) { + // Throw in some more data for stress testing (448 bits total) + PutFixed64(&str_, GetSliceHash64(str_)); + PutFixed64(&str_, GetSliceHash64(str_)); + PutFixed64(&str_, GetSliceHash64(str_)); + PutFixed64(&str_, GetSliceHash64(str_)); + PutFixed64(&str_, GetSliceHash64(str_)); + PutFixed64(&str_, GetSliceHash64(str_)); + } + } + + std::string GetChecksum() const override { + assert(!str_.empty()); + return str_; + } + + const char* Name() const override { + return big_ ? "MyBigChecksum" : "MyXXH64Checksum"; + } + + private: + bool big_; + XXH64_state_t* state_; + std::string str_; +}; + +class DbStressChecksumGenFactory : public FileChecksumGenFactory { + std::string default_func_name_; + + std::unique_ptr<FileChecksumGenerator> CreateFromFuncName( + const std::string& func_name) { + std::unique_ptr<FileChecksumGenerator> rv; + if (func_name == "FileChecksumCrc32c") { + rv.reset(new FileChecksumGenCrc32c(FileChecksumGenContext())); + } else if (func_name == "MyXXH64Checksum") { + rv.reset(new MyXXH64Checksum(false /* big */)); + } else if (func_name == "MyBigChecksum") { + rv.reset(new MyXXH64Checksum(true /* big */)); + } else { + // Should be a recognized function when we get here + assert(false); + } + return rv; + } + + public: + explicit DbStressChecksumGenFactory(const std::string& default_func_name) + : default_func_name_(default_func_name) {} + + std::unique_ptr<FileChecksumGenerator> CreateFileChecksumGenerator( + const FileChecksumGenContext& context) override { + if (context.requested_checksum_func_name.empty()) { + return CreateFromFuncName(default_func_name_); + } else { + return CreateFromFuncName(context.requested_checksum_func_name); + } + } + + const char* Name() const override { return "FileChecksumGenCrc32cFactory"; } +}; + +} // namespace + +std::shared_ptr<FileChecksumGenFactory> GetFileChecksumImpl( + const std::string& name) { + // Translate from friendly names to internal names + std::string internal_name; + if (name == "crc32c") { + internal_name = "FileChecksumCrc32c"; + } else if (name == "xxh64") { + internal_name = "MyXXH64Checksum"; + } else if (name == "big") { + internal_name = "MyBigChecksum"; + } else { + assert(name.empty() || name == "none"); + return nullptr; + } + return std::make_shared<DbStressChecksumGenFactory>(internal_name); +} + +Status DeleteFilesInDirectory(const std::string& dirname) { + std::vector<std::string> filenames; + Status s = Env::Default()->GetChildren(dirname, &filenames); + for (size_t i = 0; s.ok() && i < filenames.size(); ++i) { + s = Env::Default()->DeleteFile(dirname + "/" + filenames[i]); + } + return s; +} + +Status SaveFilesInDirectory(const std::string& src_dirname, + const std::string& dst_dirname) { + std::vector<std::string> filenames; + Status s = Env::Default()->GetChildren(src_dirname, &filenames); + for (size_t i = 0; s.ok() && i < filenames.size(); ++i) { + bool is_dir = false; + s = Env::Default()->IsDirectory(src_dirname + "/" + filenames[i], &is_dir); + if (s.ok()) { + if (is_dir) { + continue; + } + s = Env::Default()->LinkFile(src_dirname + "/" + filenames[i], + dst_dirname + "/" + filenames[i]); + } + } + return s; +} + +Status InitUnverifiedSubdir(const std::string& dirname) { + Status s = Env::Default()->FileExists(dirname); + if (s.IsNotFound()) { + return Status::OK(); + } + + const std::string kUnverifiedDirname = dirname + "/unverified"; + if (s.ok()) { + s = Env::Default()->CreateDirIfMissing(kUnverifiedDirname); + } + if (s.ok()) { + // It might already exist with some stale contents. Delete any such + // contents. + s = DeleteFilesInDirectory(kUnverifiedDirname); + } + if (s.ok()) { + s = SaveFilesInDirectory(dirname, kUnverifiedDirname); + } + return s; +} + +Status DestroyUnverifiedSubdir(const std::string& dirname) { + Status s = Env::Default()->FileExists(dirname); + if (s.IsNotFound()) { + return Status::OK(); + } + + const std::string kUnverifiedDirname = dirname + "/unverified"; + if (s.ok()) { + s = Env::Default()->FileExists(kUnverifiedDirname); + } + if (s.IsNotFound()) { + return Status::OK(); + } + + if (s.ok()) { + s = DeleteFilesInDirectory(kUnverifiedDirname); + } + if (s.ok()) { + s = Env::Default()->DeleteDir(kUnverifiedDirname); + } + return s; +} + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_common.h b/src/rocksdb/db_stress_tool/db_stress_common.h new file mode 100644 index 000000000..45f3e9c19 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_common.h @@ -0,0 +1,650 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. +// +// The test uses an array to compare against values written to the database. +// Keys written to the array are in 1:1 correspondence to the actual values in +// the database according to the formula in the function GenerateValue. + +// Space is reserved in the array from 0 to FLAGS_max_key and values are +// randomly written/deleted/read from those positions. During verification we +// compare all the positions in the array. To shorten/elongate the running +// time, you could change the settings: FLAGS_max_key, FLAGS_ops_per_thread, +// (sometimes also FLAGS_threads). +// +// NOTE that if FLAGS_test_batches_snapshots is set, the test will have +// different behavior. See comment of the flag for details. + +#ifdef GFLAGS +#pragma once +#include <fcntl.h> +#include <stdio.h> +#include <stdlib.h> +#include <sys/types.h> + +#include <algorithm> +#include <array> +#include <chrono> +#include <cinttypes> +#include <exception> +#include <queue> +#include <thread> + +#include "db/db_impl/db_impl.h" +#include "db/version_set.h" +#include "db_stress_tool/db_stress_env_wrapper.h" +#include "db_stress_tool/db_stress_listener.h" +#include "db_stress_tool/db_stress_shared_state.h" +#include "db_stress_tool/db_stress_test_base.h" +#include "logging/logging.h" +#include "monitoring/histogram.h" +#include "options/options_helper.h" +#include "port/port.h" +#include "rocksdb/cache.h" +#include "rocksdb/env.h" +#include "rocksdb/slice.h" +#include "rocksdb/slice_transform.h" +#include "rocksdb/statistics.h" +#include "rocksdb/utilities/backup_engine.h" +#include "rocksdb/utilities/checkpoint.h" +#include "rocksdb/utilities/db_ttl.h" +#include "rocksdb/utilities/debug.h" +#include "rocksdb/utilities/options_util.h" +#include "rocksdb/utilities/transaction.h" +#include "rocksdb/utilities/transaction_db.h" +#include "rocksdb/write_batch.h" +#include "test_util/testutil.h" +#include "util/coding.h" +#include "util/compression.h" +#include "util/crc32c.h" +#include "util/gflags_compat.h" +#include "util/mutexlock.h" +#include "util/random.h" +#include "util/string_util.h" +#include "utilities/blob_db/blob_db.h" +#include "utilities/fault_injection_fs.h" +#include "utilities/merge_operators.h" + +using GFLAGS_NAMESPACE::ParseCommandLineFlags; +using GFLAGS_NAMESPACE::RegisterFlagValidator; +using GFLAGS_NAMESPACE::SetUsageMessage; + +DECLARE_uint64(seed); +DECLARE_bool(read_only); +DECLARE_int64(max_key); +DECLARE_double(hot_key_alpha); +DECLARE_int32(max_key_len); +DECLARE_string(key_len_percent_dist); +DECLARE_int32(key_window_scale_factor); +DECLARE_int32(column_families); +DECLARE_string(options_file); +DECLARE_int64(active_width); +DECLARE_bool(test_batches_snapshots); +DECLARE_bool(atomic_flush); +DECLARE_int32(manual_wal_flush_one_in); +DECLARE_bool(test_cf_consistency); +DECLARE_bool(test_multi_ops_txns); +DECLARE_int32(threads); +DECLARE_int32(ttl); +DECLARE_int32(value_size_mult); +DECLARE_int32(compaction_readahead_size); +DECLARE_bool(enable_pipelined_write); +DECLARE_bool(verify_before_write); +DECLARE_bool(histogram); +DECLARE_bool(destroy_db_initially); +DECLARE_bool(verbose); +DECLARE_bool(progress_reports); +DECLARE_uint64(db_write_buffer_size); +DECLARE_int32(write_buffer_size); +DECLARE_int32(max_write_buffer_number); +DECLARE_int32(min_write_buffer_number_to_merge); +DECLARE_int32(max_write_buffer_number_to_maintain); +DECLARE_int64(max_write_buffer_size_to_maintain); +DECLARE_double(memtable_prefix_bloom_size_ratio); +DECLARE_bool(memtable_whole_key_filtering); +DECLARE_int32(open_files); +DECLARE_int64(compressed_cache_size); +DECLARE_int32(compressed_cache_numshardbits); +DECLARE_int32(compaction_style); +DECLARE_int32(compaction_pri); +DECLARE_int32(num_levels); +DECLARE_int32(level0_file_num_compaction_trigger); +DECLARE_int32(level0_slowdown_writes_trigger); +DECLARE_int32(level0_stop_writes_trigger); +DECLARE_int32(block_size); +DECLARE_int32(format_version); +DECLARE_int32(index_block_restart_interval); +DECLARE_bool(disable_auto_compactions); +DECLARE_int32(max_background_compactions); +DECLARE_int32(num_bottom_pri_threads); +DECLARE_int32(compaction_thread_pool_adjust_interval); +DECLARE_int32(compaction_thread_pool_variations); +DECLARE_int32(max_background_flushes); +DECLARE_int32(universal_size_ratio); +DECLARE_int32(universal_min_merge_width); +DECLARE_int32(universal_max_merge_width); +DECLARE_int32(universal_max_size_amplification_percent); +DECLARE_int32(clear_column_family_one_in); +DECLARE_int32(get_live_files_one_in); +DECLARE_int32(get_sorted_wal_files_one_in); +DECLARE_int32(get_current_wal_file_one_in); +DECLARE_int32(set_options_one_in); +DECLARE_int32(set_in_place_one_in); +DECLARE_int64(cache_size); +DECLARE_int32(cache_numshardbits); +DECLARE_bool(cache_index_and_filter_blocks); +DECLARE_bool(charge_compression_dictionary_building_buffer); +DECLARE_bool(charge_filter_construction); +DECLARE_bool(charge_table_reader); +DECLARE_bool(charge_file_metadata); +DECLARE_bool(charge_blob_cache); +DECLARE_int32(top_level_index_pinning); +DECLARE_int32(partition_pinning); +DECLARE_int32(unpartitioned_pinning); +DECLARE_string(cache_type); +DECLARE_uint64(subcompactions); +DECLARE_uint64(periodic_compaction_seconds); +DECLARE_uint64(compaction_ttl); +DECLARE_bool(allow_concurrent_memtable_write); +DECLARE_double(experimental_mempurge_threshold); +DECLARE_bool(enable_write_thread_adaptive_yield); +DECLARE_int32(reopen); +DECLARE_double(bloom_bits); +DECLARE_int32(ribbon_starting_level); +DECLARE_bool(partition_filters); +DECLARE_bool(optimize_filters_for_memory); +DECLARE_bool(detect_filter_construct_corruption); +DECLARE_int32(index_type); +DECLARE_int32(data_block_index_type); +DECLARE_string(db); +DECLARE_string(secondaries_base); +DECLARE_bool(test_secondary); +DECLARE_string(expected_values_dir); +DECLARE_bool(verify_checksum); +DECLARE_bool(mmap_read); +DECLARE_bool(mmap_write); +DECLARE_bool(use_direct_reads); +DECLARE_bool(use_direct_io_for_flush_and_compaction); +DECLARE_bool(mock_direct_io); +DECLARE_bool(statistics); +DECLARE_bool(sync); +DECLARE_bool(use_fsync); +DECLARE_uint64(stats_dump_period_sec); +DECLARE_uint64(bytes_per_sync); +DECLARE_uint64(wal_bytes_per_sync); +DECLARE_int32(kill_random_test); +DECLARE_string(kill_exclude_prefixes); +DECLARE_bool(disable_wal); +DECLARE_uint64(recycle_log_file_num); +DECLARE_int64(target_file_size_base); +DECLARE_int32(target_file_size_multiplier); +DECLARE_uint64(max_bytes_for_level_base); +DECLARE_double(max_bytes_for_level_multiplier); +DECLARE_int32(range_deletion_width); +DECLARE_uint64(rate_limiter_bytes_per_sec); +DECLARE_bool(rate_limit_bg_reads); +DECLARE_bool(rate_limit_user_ops); +DECLARE_bool(rate_limit_auto_wal_flush); +DECLARE_uint64(sst_file_manager_bytes_per_sec); +DECLARE_uint64(sst_file_manager_bytes_per_truncate); +DECLARE_bool(use_txn); +DECLARE_uint64(txn_write_policy); +DECLARE_bool(unordered_write); +DECLARE_int32(backup_one_in); +DECLARE_uint64(backup_max_size); +DECLARE_int32(checkpoint_one_in); +DECLARE_int32(ingest_external_file_one_in); +DECLARE_int32(ingest_external_file_width); +DECLARE_int32(compact_files_one_in); +DECLARE_int32(compact_range_one_in); +DECLARE_int32(mark_for_compaction_one_file_in); +DECLARE_int32(flush_one_in); +DECLARE_int32(pause_background_one_in); +DECLARE_int32(compact_range_width); +DECLARE_int32(acquire_snapshot_one_in); +DECLARE_bool(compare_full_db_state_snapshot); +DECLARE_uint64(snapshot_hold_ops); +DECLARE_bool(long_running_snapshots); +DECLARE_bool(use_multiget); +DECLARE_int32(readpercent); +DECLARE_int32(prefixpercent); +DECLARE_int32(writepercent); +DECLARE_int32(delpercent); +DECLARE_int32(delrangepercent); +DECLARE_int32(nooverwritepercent); +DECLARE_int32(iterpercent); +DECLARE_uint64(num_iterations); +DECLARE_int32(customopspercent); +DECLARE_string(compression_type); +DECLARE_string(bottommost_compression_type); +DECLARE_int32(compression_max_dict_bytes); +DECLARE_int32(compression_zstd_max_train_bytes); +DECLARE_int32(compression_parallel_threads); +DECLARE_uint64(compression_max_dict_buffer_bytes); +DECLARE_bool(compression_use_zstd_dict_trainer); +DECLARE_string(checksum_type); +DECLARE_string(env_uri); +DECLARE_string(fs_uri); +DECLARE_uint64(ops_per_thread); +DECLARE_uint64(log2_keys_per_lock); +DECLARE_uint64(max_manifest_file_size); +DECLARE_bool(in_place_update); +DECLARE_string(memtablerep); +DECLARE_int32(prefix_size); +DECLARE_bool(use_merge); +DECLARE_uint32(use_put_entity_one_in); +DECLARE_bool(use_full_merge_v1); +DECLARE_int32(sync_wal_one_in); +DECLARE_bool(avoid_unnecessary_blocking_io); +DECLARE_bool(write_dbid_to_manifest); +DECLARE_bool(avoid_flush_during_recovery); +DECLARE_uint64(max_write_batch_group_size_bytes); +DECLARE_bool(level_compaction_dynamic_level_bytes); +DECLARE_int32(verify_checksum_one_in); +DECLARE_int32(verify_db_one_in); +DECLARE_int32(continuous_verification_interval); +DECLARE_int32(get_property_one_in); +DECLARE_string(file_checksum_impl); + +#ifndef ROCKSDB_LITE +// Options for StackableDB-based BlobDB +DECLARE_bool(use_blob_db); +DECLARE_uint64(blob_db_min_blob_size); +DECLARE_uint64(blob_db_bytes_per_sync); +DECLARE_uint64(blob_db_file_size); +DECLARE_bool(blob_db_enable_gc); +DECLARE_double(blob_db_gc_cutoff); +#endif // !ROCKSDB_LITE + +// Options for integrated BlobDB +DECLARE_bool(allow_setting_blob_options_dynamically); +DECLARE_bool(enable_blob_files); +DECLARE_uint64(min_blob_size); +DECLARE_uint64(blob_file_size); +DECLARE_string(blob_compression_type); +DECLARE_bool(enable_blob_garbage_collection); +DECLARE_double(blob_garbage_collection_age_cutoff); +DECLARE_double(blob_garbage_collection_force_threshold); +DECLARE_uint64(blob_compaction_readahead_size); +DECLARE_int32(blob_file_starting_level); +DECLARE_bool(use_blob_cache); +DECLARE_bool(use_shared_block_and_blob_cache); +DECLARE_uint64(blob_cache_size); +DECLARE_int32(blob_cache_numshardbits); +DECLARE_int32(prepopulate_blob_cache); + +DECLARE_int32(approximate_size_one_in); +DECLARE_bool(sync_fault_injection); + +DECLARE_bool(best_efforts_recovery); +DECLARE_bool(skip_verifydb); +DECLARE_bool(enable_compaction_filter); +DECLARE_bool(paranoid_file_checks); +DECLARE_bool(fail_if_options_file_error); +DECLARE_uint64(batch_protection_bytes_per_key); +DECLARE_uint32(memtable_protection_bytes_per_key); + +DECLARE_uint64(user_timestamp_size); +DECLARE_string(secondary_cache_uri); +DECLARE_int32(secondary_cache_fault_one_in); + +DECLARE_int32(prepopulate_block_cache); + +DECLARE_bool(two_write_queues); +#ifndef ROCKSDB_LITE +DECLARE_bool(use_only_the_last_commit_time_batch_for_recovery); +DECLARE_uint64(wp_snapshot_cache_bits); +DECLARE_uint64(wp_commit_cache_bits); +#endif // !ROCKSDB_LITE + +DECLARE_bool(adaptive_readahead); +DECLARE_bool(async_io); +DECLARE_string(wal_compression); +DECLARE_bool(verify_sst_unique_id_in_manifest); + +DECLARE_int32(create_timestamped_snapshot_one_in); + +DECLARE_bool(allow_data_in_errors); + +// Tiered storage +DECLARE_bool(enable_tiered_storage); // set last_level_temperature +DECLARE_int64(preclude_last_level_data_seconds); +DECLARE_int64(preserve_internal_time_seconds); + +DECLARE_int32(verify_iterator_with_expected_state_one_in); +DECLARE_bool(preserve_unverified_changes); + +DECLARE_uint64(readahead_size); +DECLARE_uint64(initial_auto_readahead_size); +DECLARE_uint64(max_auto_readahead_size); +DECLARE_uint64(num_file_reads_for_auto_readahead); + +constexpr long KB = 1024; +constexpr int kRandomValueMaxFactor = 3; +constexpr int kValueMaxLen = 100; + +// wrapped posix environment +extern ROCKSDB_NAMESPACE::Env* db_stress_env; +extern ROCKSDB_NAMESPACE::Env* db_stress_listener_env; +extern std::shared_ptr<ROCKSDB_NAMESPACE::FaultInjectionTestFS> fault_fs_guard; + +extern enum ROCKSDB_NAMESPACE::CompressionType compression_type_e; +extern enum ROCKSDB_NAMESPACE::CompressionType bottommost_compression_type_e; +extern enum ROCKSDB_NAMESPACE::ChecksumType checksum_type_e; + +enum RepFactory { kSkipList, kHashSkipList, kVectorRep }; + +inline enum RepFactory StringToRepFactory(const char* ctype) { + assert(ctype); + + if (!strcasecmp(ctype, "skip_list")) + return kSkipList; + else if (!strcasecmp(ctype, "prefix_hash")) + return kHashSkipList; + else if (!strcasecmp(ctype, "vector")) + return kVectorRep; + + fprintf(stdout, "Cannot parse memreptable %s\n", ctype); + return kSkipList; +} + +extern enum RepFactory FLAGS_rep_factory; + +namespace ROCKSDB_NAMESPACE { +inline enum ROCKSDB_NAMESPACE::CompressionType StringToCompressionType( + const char* ctype) { + assert(ctype); + + ROCKSDB_NAMESPACE::CompressionType ret_compression_type; + + if (!strcasecmp(ctype, "disable")) { + ret_compression_type = ROCKSDB_NAMESPACE::kDisableCompressionOption; + } else if (!strcasecmp(ctype, "none")) { + ret_compression_type = ROCKSDB_NAMESPACE::kNoCompression; + } else if (!strcasecmp(ctype, "snappy")) { + ret_compression_type = ROCKSDB_NAMESPACE::kSnappyCompression; + } else if (!strcasecmp(ctype, "zlib")) { + ret_compression_type = ROCKSDB_NAMESPACE::kZlibCompression; + } else if (!strcasecmp(ctype, "bzip2")) { + ret_compression_type = ROCKSDB_NAMESPACE::kBZip2Compression; + } else if (!strcasecmp(ctype, "lz4")) { + ret_compression_type = ROCKSDB_NAMESPACE::kLZ4Compression; + } else if (!strcasecmp(ctype, "lz4hc")) { + ret_compression_type = ROCKSDB_NAMESPACE::kLZ4HCCompression; + } else if (!strcasecmp(ctype, "xpress")) { + ret_compression_type = ROCKSDB_NAMESPACE::kXpressCompression; + } else if (!strcasecmp(ctype, "zstd")) { + ret_compression_type = ROCKSDB_NAMESPACE::kZSTD; + } else { + fprintf(stderr, "Cannot parse compression type '%s'\n", ctype); + ret_compression_type = + ROCKSDB_NAMESPACE::kSnappyCompression; // default value + } + if (ret_compression_type != ROCKSDB_NAMESPACE::kDisableCompressionOption && + !CompressionTypeSupported(ret_compression_type)) { + // Use no compression will be more portable but considering this is + // only a stress test and snappy is widely available. Use snappy here. + ret_compression_type = ROCKSDB_NAMESPACE::kSnappyCompression; + } + return ret_compression_type; +} + +inline enum ROCKSDB_NAMESPACE::ChecksumType StringToChecksumType( + const char* ctype) { + assert(ctype); + auto iter = ROCKSDB_NAMESPACE::checksum_type_string_map.find(ctype); + if (iter != ROCKSDB_NAMESPACE::checksum_type_string_map.end()) { + return iter->second; + } + fprintf(stderr, "Cannot parse checksum type '%s'\n", ctype); + return ROCKSDB_NAMESPACE::kCRC32c; +} + +inline std::string ChecksumTypeToString(ROCKSDB_NAMESPACE::ChecksumType ctype) { + auto iter = std::find_if( + ROCKSDB_NAMESPACE::checksum_type_string_map.begin(), + ROCKSDB_NAMESPACE::checksum_type_string_map.end(), + [&](const std::pair<std::string, ROCKSDB_NAMESPACE::ChecksumType>& + name_and_enum_val) { return name_and_enum_val.second == ctype; }); + assert(iter != ROCKSDB_NAMESPACE::checksum_type_string_map.end()); + return iter->first; +} + +inline std::vector<std::string> SplitString(std::string src) { + std::vector<std::string> ret; + if (src.empty()) { + return ret; + } + size_t pos = 0; + size_t pos_comma; + while ((pos_comma = src.find(',', pos)) != std::string::npos) { + ret.push_back(src.substr(pos, pos_comma - pos)); + pos = pos_comma + 1; + } + ret.push_back(src.substr(pos, src.length())); + return ret; +} + +#ifdef _MSC_VER +#pragma warning(push) +// truncation of constant value on static_cast +#pragma warning(disable : 4309) +#endif +inline bool GetNextPrefix(const ROCKSDB_NAMESPACE::Slice& src, std::string* v) { + std::string ret = src.ToString(); + for (int i = static_cast<int>(ret.size()) - 1; i >= 0; i--) { + if (ret[i] != static_cast<char>(255)) { + ret[i] = ret[i] + 1; + break; + } else if (i != 0) { + ret[i] = 0; + } else { + // all FF. No next prefix + return false; + } + } + *v = ret; + return true; +} +#ifdef _MSC_VER +#pragma warning(pop) +#endif + +// Append `val` to `*key` in fixed-width big-endian format +extern inline void AppendIntToString(uint64_t val, std::string* key) { + // PutFixed64 uses little endian + PutFixed64(key, val); + // Reverse to get big endian + char* int_data = &((*key)[key->size() - sizeof(uint64_t)]); + for (size_t i = 0; i < sizeof(uint64_t) / 2; ++i) { + std::swap(int_data[i], int_data[sizeof(uint64_t) - 1 - i]); + } +} + +// A struct for maintaining the parameters for generating variable length keys +struct KeyGenContext { + // Number of adjacent keys in one cycle of key lengths + uint64_t window; + // Number of keys of each possible length in a given window + std::vector<uint64_t> weights; +}; +extern KeyGenContext key_gen_ctx; + +// Generate a variable length key string from the given int64 val. The +// order of the keys is preserved. The key could be anywhere from 8 to +// max_key_len * 8 bytes. +// The algorithm picks the length based on the +// offset of the val within a configured window and the distribution of the +// number of keys of various lengths in that window. For example, if x, y, x are +// the weights assigned to each possible key length, the keys generated would be +// - {0}...{x-1} +// {(x-1),0}..{(x-1),(y-1)},{(x-1),(y-1),0}..{(x-1),(y-1),(z-1)} and so on. +// Additionally, a trailer of 0-7 bytes could be appended. +extern inline std::string Key(int64_t val) { + uint64_t window = key_gen_ctx.window; + size_t levels = key_gen_ctx.weights.size(); + std::string key; + // Over-reserve and for now do not bother `shrink_to_fit()` since the key + // strings are transient. + key.reserve(FLAGS_max_key_len * 8); + + uint64_t window_idx = static_cast<uint64_t>(val) / window; + uint64_t offset = static_cast<uint64_t>(val) % window; + for (size_t level = 0; level < levels; ++level) { + uint64_t weight = key_gen_ctx.weights[level]; + uint64_t pfx; + if (level == 0) { + pfx = window_idx * weight; + } else { + pfx = 0; + } + pfx += offset >= weight ? weight - 1 : offset; + AppendIntToString(pfx, &key); + if (offset < weight) { + // Use the bottom 3 bits of offset as the number of trailing 'x's in the + // key. If the next key is going to be of the next level, then skip the + // trailer as it would break ordering. If the key length is already at + // max, skip the trailer. + if (offset < weight - 1 && level < levels - 1) { + size_t trailer_len = offset & 0x7; + key.append(trailer_len, 'x'); + } + break; + } + offset -= weight; + } + + return key; +} + +// Given a string key, map it to an index into the expected values buffer +extern inline bool GetIntVal(std::string big_endian_key, uint64_t* key_p) { + size_t size_key = big_endian_key.size(); + std::vector<uint64_t> prefixes; + + assert(size_key <= key_gen_ctx.weights.size() * sizeof(uint64_t)); + + std::string little_endian_key; + little_endian_key.resize(size_key); + for (size_t start = 0; start + sizeof(uint64_t) <= size_key; + start += sizeof(uint64_t)) { + size_t end = start + sizeof(uint64_t); + for (size_t i = 0; i < sizeof(uint64_t); ++i) { + little_endian_key[start + i] = big_endian_key[end - 1 - i]; + } + Slice little_endian_slice = + Slice(&little_endian_key[start], sizeof(uint64_t)); + uint64_t pfx; + if (!GetFixed64(&little_endian_slice, &pfx)) { + return false; + } + prefixes.emplace_back(pfx); + } + + uint64_t key = 0; + for (size_t i = 0; i < prefixes.size(); ++i) { + uint64_t pfx = prefixes[i]; + key += (pfx / key_gen_ctx.weights[i]) * key_gen_ctx.window + + pfx % key_gen_ctx.weights[i]; + if (i < prefixes.size() - 1) { + // The encoding writes a `key_gen_ctx.weights[i] - 1` that counts for + // `key_gen_ctx.weights[i]` when there are more prefixes to come. So we + // need to add back the one here as we're at a non-last prefix. + ++key; + } + } + *key_p = key; + return true; +} + +// Given a string prefix, map it to the first corresponding index in the +// expected values buffer. +inline bool GetFirstIntValInPrefix(std::string big_endian_prefix, + uint64_t* key_p) { + size_t size_key = big_endian_prefix.size(); + // Pad with zeros to make it a multiple of 8. This function may be called + // with a prefix, in which case we return the first index that falls + // inside or outside that prefix, dependeing on whether the prefix is + // the start of upper bound of a scan + unsigned int pad = sizeof(uint64_t) - (size_key % sizeof(uint64_t)); + if (pad < sizeof(uint64_t)) { + big_endian_prefix.append(pad, '\0'); + } + return GetIntVal(std::move(big_endian_prefix), key_p); +} + +extern inline uint64_t GetPrefixKeyCount(const std::string& prefix, + const std::string& ub) { + uint64_t start = 0; + uint64_t end = 0; + + if (!GetFirstIntValInPrefix(prefix, &start) || + !GetFirstIntValInPrefix(ub, &end)) { + return 0; + } + + return end - start; +} + +extern inline std::string StringToHex(const std::string& str) { + std::string result = "0x"; + result.append(Slice(str).ToString(true)); + return result; +} + +// Unified output format for double parameters +extern inline std::string FormatDoubleParam(double param) { + return std::to_string(param); +} + +// Make sure that double parameter is a value we can reproduce by +// re-inputting the value printed. +extern inline void SanitizeDoubleParam(double* param) { + *param = std::atof(FormatDoubleParam(*param).c_str()); +} + +extern void PoolSizeChangeThread(void* v); + +extern void DbVerificationThread(void* v); + +extern void TimestampedSnapshotsThread(void* v); + +extern void PrintKeyValue(int cf, uint64_t key, const char* value, size_t sz); + +extern int64_t GenerateOneKey(ThreadState* thread, uint64_t iteration); + +extern std::vector<int64_t> GenerateNKeys(ThreadState* thread, int num_keys, + uint64_t iteration); + +extern size_t GenerateValue(uint32_t rand, char* v, size_t max_sz); +extern uint32_t GetValueBase(Slice s); + +extern WideColumns GenerateWideColumns(uint32_t value_base, const Slice& slice); +extern WideColumns GenerateExpectedWideColumns(uint32_t value_base, + const Slice& slice); + +extern StressTest* CreateCfConsistencyStressTest(); +extern StressTest* CreateBatchedOpsStressTest(); +extern StressTest* CreateNonBatchedOpsStressTest(); +extern StressTest* CreateMultiOpsTxnsStressTest(); +extern void CheckAndSetOptionsForMultiOpsTxnStressTest(); +extern void InitializeHotKeyGenerator(double alpha); +extern int64_t GetOneHotKeyID(double rand_seed, int64_t max_key); + +extern std::string GetNowNanos(); + +std::shared_ptr<FileChecksumGenFactory> GetFileChecksumImpl( + const std::string& name); + +Status DeleteFilesInDirectory(const std::string& dirname); +Status SaveFilesInDirectory(const std::string& src_dirname, + const std::string& dst_dirname); +Status DestroyUnverifiedSubdir(const std::string& dirname); +Status InitUnverifiedSubdir(const std::string& dirname); +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_compaction_filter.h b/src/rocksdb/db_stress_tool/db_stress_compaction_filter.h new file mode 100644 index 000000000..408bb48f3 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_compaction_filter.h @@ -0,0 +1,96 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#pragma once + +#include "db_stress_tool/db_stress_common.h" +#include "db_stress_tool/db_stress_shared_state.h" +#include "rocksdb/compaction_filter.h" + +namespace ROCKSDB_NAMESPACE { + +// DbStressCompactionFilter is safe to use with db_stress as it does not perform +// any mutation. It only makes `kRemove` decisions for keys that are already +// non-existent according to the `SharedState`. +class DbStressCompactionFilter : public CompactionFilter { + public: + DbStressCompactionFilter(SharedState* state, int cf_id) + : state_(state), cf_id_(cf_id) {} + + Decision FilterV2(int /*level*/, const Slice& key, ValueType /*value_type*/, + const Slice& /*existing_value*/, std::string* /*new_value*/, + std::string* /*skip_until*/) const override { + if (state_ == nullptr) { + return Decision::kKeep; + } + if (key.empty() || ('0' <= key[0] && key[0] <= '9')) { + // It is likely leftover from a test_batches_snapshots run. Below this + // conditional, the test_batches_snapshots key format is not handled + // properly. Just keep it to be safe. + return Decision::kKeep; + } + uint64_t key_num = 0; + { + Slice ukey_without_ts = key; + assert(ukey_without_ts.size() >= FLAGS_user_timestamp_size); + ukey_without_ts.remove_suffix(FLAGS_user_timestamp_size); + [[maybe_unused]] bool ok = + GetIntVal(ukey_without_ts.ToString(), &key_num); + assert(ok); + } + port::Mutex* key_mutex = state_->GetMutexForKey(cf_id_, key_num); + if (!key_mutex->TryLock()) { + return Decision::kKeep; + } + // Reaching here means we acquired the lock. + + bool key_exists = state_->Exists(cf_id_, key_num); + const bool allow_overwrite = state_->AllowsOverwrite(key_num); + + key_mutex->Unlock(); + + if (!key_exists) { + return allow_overwrite ? Decision::kRemove : Decision::kPurge; + } + return Decision::kKeep; + } + + const char* Name() const override { return "DbStressCompactionFilter"; } + + private: + SharedState* const state_; + const int cf_id_; +}; + +class DbStressCompactionFilterFactory : public CompactionFilterFactory { + public: + DbStressCompactionFilterFactory() : state_(nullptr) {} + + void SetSharedState(SharedState* state) { + MutexLock state_mutex_guard(&state_mutex_); + state_ = state; + } + + std::unique_ptr<CompactionFilter> CreateCompactionFilter( + const CompactionFilter::Context& context) override { + MutexLock state_mutex_guard(&state_mutex_); + return std::unique_ptr<CompactionFilter>( + new DbStressCompactionFilter(state_, context.column_family_id)); + } + + const char* Name() const override { + return "DbStressCompactionFilterFactory"; + } + + private: + port::Mutex state_mutex_; + SharedState* state_; +}; + +} // namespace ROCKSDB_NAMESPACE diff --git a/src/rocksdb/db_stress_tool/db_stress_driver.cc b/src/rocksdb/db_stress_tool/db_stress_driver.cc new file mode 100644 index 000000000..ed1240e00 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_driver.cc @@ -0,0 +1,212 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. +// + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" +#include "utilities/fault_injection_fs.h" + +namespace ROCKSDB_NAMESPACE { +void ThreadBody(void* v) { + ThreadState* thread = reinterpret_cast<ThreadState*>(v); + SharedState* shared = thread->shared; + + if (!FLAGS_skip_verifydb && shared->ShouldVerifyAtBeginning()) { + thread->shared->GetStressTest()->VerifyDb(thread); + } + { + MutexLock l(shared->GetMutex()); + shared->IncInitialized(); + if (shared->AllInitialized()) { + shared->GetCondVar()->SignalAll(); + } + while (!shared->Started()) { + shared->GetCondVar()->Wait(); + } + } + thread->shared->GetStressTest()->OperateDb(thread); + + { + MutexLock l(shared->GetMutex()); + shared->IncOperated(); + if (shared->AllOperated()) { + shared->GetCondVar()->SignalAll(); + } + while (!shared->VerifyStarted()) { + shared->GetCondVar()->Wait(); + } + } + + if (!FLAGS_skip_verifydb) { + thread->shared->GetStressTest()->VerifyDb(thread); + } + + { + MutexLock l(shared->GetMutex()); + shared->IncDone(); + if (shared->AllDone()) { + shared->GetCondVar()->SignalAll(); + } + } +} + +bool RunStressTest(StressTest* stress) { + SystemClock* clock = db_stress_env->GetSystemClock().get(); + + SharedState shared(db_stress_env, stress); + + if (shared.ShouldVerifyAtBeginning() && FLAGS_preserve_unverified_changes) { + Status s = InitUnverifiedSubdir(FLAGS_db); + if (s.ok() && !FLAGS_expected_values_dir.empty()) { + s = InitUnverifiedSubdir(FLAGS_expected_values_dir); + } + if (!s.ok()) { + fprintf(stderr, "Failed to setup unverified state dir: %s\n", + s.ToString().c_str()); + exit(1); + } + } + + stress->InitDb(&shared); + stress->FinishInitDb(&shared); + + if (FLAGS_sync_fault_injection) { + fault_fs_guard->SetFilesystemDirectWritable(false); + } + if (FLAGS_write_fault_one_in) { + fault_fs_guard->EnableWriteErrorInjection(); + } + + uint32_t n = FLAGS_threads; + uint64_t now = clock->NowMicros(); + fprintf(stdout, "%s Initializing worker threads\n", + clock->TimeToString(now / 1000000).c_str()); + + shared.SetThreads(n); + + if (FLAGS_compaction_thread_pool_adjust_interval > 0) { + shared.IncBgThreads(); + } + + if (FLAGS_continuous_verification_interval > 0) { + shared.IncBgThreads(); + } + + std::vector<ThreadState*> threads(n); + for (uint32_t i = 0; i < n; i++) { + threads[i] = new ThreadState(i, &shared); + db_stress_env->StartThread(ThreadBody, threads[i]); + } + + ThreadState bg_thread(0, &shared); + if (FLAGS_compaction_thread_pool_adjust_interval > 0) { + db_stress_env->StartThread(PoolSizeChangeThread, &bg_thread); + } + + ThreadState continuous_verification_thread(0, &shared); + if (FLAGS_continuous_verification_interval > 0) { + db_stress_env->StartThread(DbVerificationThread, + &continuous_verification_thread); + } + + // Each thread goes through the following states: + // initializing -> wait for others to init -> read/populate/depopulate + // wait for others to operate -> verify -> done + + { + MutexLock l(shared.GetMutex()); + while (!shared.AllInitialized()) { + shared.GetCondVar()->Wait(); + } + if (shared.ShouldVerifyAtBeginning()) { + if (shared.HasVerificationFailedYet()) { + fprintf(stderr, "Crash-recovery verification failed :(\n"); + } else { + fprintf(stdout, "Crash-recovery verification passed :)\n"); + Status s = DestroyUnverifiedSubdir(FLAGS_db); + if (s.ok() && !FLAGS_expected_values_dir.empty()) { + s = DestroyUnverifiedSubdir(FLAGS_expected_values_dir); + } + if (!s.ok()) { + fprintf(stderr, "Failed to cleanup unverified state dir: %s\n", + s.ToString().c_str()); + exit(1); + } + } + } + + // This is after the verification step to avoid making all those `Get()`s + // and `MultiGet()`s contend on the DB-wide trace mutex. + if (!FLAGS_expected_values_dir.empty()) { + stress->TrackExpectedState(&shared); + } + + now = clock->NowMicros(); + fprintf(stdout, "%s Starting database operations\n", + clock->TimeToString(now / 1000000).c_str()); + + shared.SetStart(); + shared.GetCondVar()->SignalAll(); + while (!shared.AllOperated()) { + shared.GetCondVar()->Wait(); + } + + now = clock->NowMicros(); + if (FLAGS_test_batches_snapshots) { + fprintf(stdout, "%s Limited verification already done during gets\n", + clock->TimeToString((uint64_t)now / 1000000).c_str()); + } else if (FLAGS_skip_verifydb) { + fprintf(stdout, "%s Verification skipped\n", + clock->TimeToString((uint64_t)now / 1000000).c_str()); + } else { + fprintf(stdout, "%s Starting verification\n", + clock->TimeToString((uint64_t)now / 1000000).c_str()); + } + + shared.SetStartVerify(); + shared.GetCondVar()->SignalAll(); + while (!shared.AllDone()) { + shared.GetCondVar()->Wait(); + } + } + + for (unsigned int i = 1; i < n; i++) { + threads[0]->stats.Merge(threads[i]->stats); + } + threads[0]->stats.Report("Stress Test"); + + for (unsigned int i = 0; i < n; i++) { + delete threads[i]; + threads[i] = nullptr; + } + now = clock->NowMicros(); + if (!FLAGS_skip_verifydb && !FLAGS_test_batches_snapshots && + !shared.HasVerificationFailedYet()) { + fprintf(stdout, "%s Verification successful\n", + clock->TimeToString(now / 1000000).c_str()); + } + stress->PrintStatistics(); + + if (FLAGS_compaction_thread_pool_adjust_interval > 0 || + FLAGS_continuous_verification_interval > 0) { + MutexLock l(shared.GetMutex()); + shared.SetShouldStopBgThread(); + while (!shared.BgThreadsFinished()) { + shared.GetCondVar()->Wait(); + } + } + + if (shared.HasVerificationFailedYet()) { + fprintf(stderr, "Verification failed :(\n"); + return false; + } + return true; +} +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_driver.h b/src/rocksdb/db_stress_tool/db_stress_driver.h new file mode 100644 index 000000000..ff701fcb2 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_driver.h @@ -0,0 +1,17 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#pragma once +#include "db_stress_tool/db_stress_test_base.h" +namespace ROCKSDB_NAMESPACE { +extern void ThreadBody(void* /*thread_state*/); +extern bool RunStressTest(StressTest*); +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_env_wrapper.h b/src/rocksdb/db_stress_tool/db_stress_env_wrapper.h new file mode 100644 index 000000000..21f6db2ab --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_env_wrapper.h @@ -0,0 +1,42 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#pragma once +#include "db_stress_tool/db_stress_common.h" + +namespace ROCKSDB_NAMESPACE { +class DbStressEnvWrapper : public EnvWrapper { + public: + explicit DbStressEnvWrapper(Env* t) : EnvWrapper(t) {} + static const char* kClassName() { return "DbStressEnv"; } + const char* Name() const override { return kClassName(); } + + Status DeleteFile(const std::string& f) override { + // We determine whether it is a manifest file by searching a strong, + // so that there will be false positive if the directory path contains the + // keyword but it is unlikely. + // Checkpoint, backup, and restore directories needs to be exempted. + if (!if_preserve_all_manifests || + f.find("MANIFEST-") == std::string::npos || + f.find("checkpoint") != std::string::npos || + f.find(".backup") != std::string::npos || + f.find(".restore") != std::string::npos) { + return target()->DeleteFile(f); + } + // Rename the file instead of deletion to keep the history, and + // at the same time it is not visible to RocksDB. + return target()->RenameFile(f, f + "_renamed_"); + } + + // If true, all manifest files will not be delted in DeleteFile(). + bool if_preserve_all_manifests = true; +}; +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_gflags.cc b/src/rocksdb/db_stress_tool/db_stress_gflags.cc new file mode 100644 index 000000000..7adc66509 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_gflags.cc @@ -0,0 +1,1074 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" + +static bool ValidateUint32Range(const char* flagname, uint64_t value) { + if (value > std::numeric_limits<uint32_t>::max()) { + fprintf(stderr, "Invalid value for --%s: %lu, overflow\n", flagname, + (unsigned long)value); + return false; + } + return true; +} + +DEFINE_uint64(seed, 2341234, + "Seed for PRNG. When --nooverwritepercent is " + "nonzero and --expected_values_dir is nonempty, this value " + "must be fixed across invocations."); +static const bool FLAGS_seed_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_seed, &ValidateUint32Range); + +DEFINE_bool(read_only, false, "True if open DB in read-only mode during tests"); + +DEFINE_int64(max_key, 1 * KB * KB, + "Max number of key/values to place in database"); + +DEFINE_int32(max_key_len, 3, "Maximum length of a key in 8-byte units"); + +DEFINE_string(key_len_percent_dist, "", + "Percentages of keys of various lengths. For example, 1,30,69 " + "means 1% of keys are 8 bytes, 30% are 16 bytes, and 69% are " + "24 bytes. If not specified, it will be evenly distributed"); + +DEFINE_int32(key_window_scale_factor, 10, + "This value will be multiplied by 100 to come up with a window " + "size for varying the key length"); + +DEFINE_int32(column_families, 10, "Number of column families"); + +DEFINE_double( + hot_key_alpha, 0, + "Use Zipfian distribution to generate the key " + "distribution. If it is not specified, write path will use random " + "distribution to generate the keys. The parameter is [0, double_max]). " + "However, the larger alpha is, the more shewed will be. If alpha is " + "larger than 2, it is likely that only 1 key will be accessed. The " + "Recommended value is [0.8-1.5]. The distribution is also related to " + "max_key and total iterations of generating the hot key. "); + +DEFINE_string( + options_file, "", + "The path to a RocksDB options file. If specified, then db_stress will " + "run with the RocksDB options in the default column family of the " + "specified options file. Note that, when an options file is provided, " + "db_stress will ignore the flag values for all options that may be passed " + "via options file."); + +DEFINE_int64( + active_width, 0, + "Number of keys in active span of the key-range at any given time. The " + "span begins with its left endpoint at key 0, gradually moves rightwards, " + "and ends with its right endpoint at max_key. If set to 0, active_width " + "will be sanitized to be equal to max_key."); + +// TODO(noetzli) Add support for single deletes +DEFINE_bool(test_batches_snapshots, false, + "If set, the test uses MultiGet(), MultiPut() and MultiDelete()" + " which read/write/delete multiple keys in a batch. In this mode," + " we do not verify db content by comparing the content with the " + "pre-allocated array. Instead, we do partial verification inside" + " MultiGet() by checking various values in a batch. Benefit of" + " this mode:\n" + "\t(a) No need to acquire mutexes during writes (less cache " + "flushes in multi-core leading to speed up)\n" + "\t(b) No long validation at the end (more speed up)\n" + "\t(c) Test snapshot and atomicity of batch writes"); + +DEFINE_bool(atomic_flush, false, + "If set, enables atomic flush in the options.\n"); + +DEFINE_int32( + manual_wal_flush_one_in, 0, + "If non-zero, then `FlushWAL(bool sync)`, where `bool sync` is randomly " + "decided, will be explictly called in db stress once for every N ops " + "on average. Setting `manual_wal_flush_one_in` to be greater than 0 " + "implies `Options::manual_wal_flush = true` is set."); + +DEFINE_bool(test_cf_consistency, false, + "If set, runs the stress test dedicated to verifying writes to " + "multiple column families are consistent. Setting this implies " + "`atomic_flush=true` is set true if `disable_wal=false`.\n"); + +DEFINE_bool(test_multi_ops_txns, false, + "If set, runs stress test dedicated to verifying multi-ops " + "transactions on a simple relational table with primary and " + "secondary index."); + +DEFINE_int32(threads, 32, "Number of concurrent threads to run."); + +DEFINE_int32(ttl, -1, + "Opens the db with this ttl value if this is not -1. " + "Carefully specify a large value such that verifications on " + "deleted values don't fail"); + +DEFINE_int32(value_size_mult, 8, + "Size of value will be this number times rand_int(1,3) bytes"); + +DEFINE_int32(compaction_readahead_size, 0, "Compaction readahead size"); + +DEFINE_bool(enable_pipelined_write, false, "Pipeline WAL/memtable writes"); + +DEFINE_bool(verify_before_write, false, "Verify before write"); + +DEFINE_bool(histogram, false, "Print histogram of operation timings"); + +DEFINE_bool(destroy_db_initially, true, + "Destroys the database dir before start if this is true"); + +DEFINE_bool(verbose, false, "Verbose"); + +DEFINE_bool(progress_reports, true, + "If true, db_stress will report number of finished operations"); + +DEFINE_uint64(db_write_buffer_size, + ROCKSDB_NAMESPACE::Options().db_write_buffer_size, + "Number of bytes to buffer in all memtables before compacting"); + +DEFINE_int32( + write_buffer_size, + static_cast<int32_t>(ROCKSDB_NAMESPACE::Options().write_buffer_size), + "Number of bytes to buffer in memtable before compacting"); + +DEFINE_int32(max_write_buffer_number, + ROCKSDB_NAMESPACE::Options().max_write_buffer_number, + "The number of in-memory memtables. " + "Each memtable is of size FLAGS_write_buffer_size."); + +DEFINE_int32(min_write_buffer_number_to_merge, + ROCKSDB_NAMESPACE::Options().min_write_buffer_number_to_merge, + "The minimum number of write buffers that will be merged together " + "before writing to storage. This is cheap because it is an " + "in-memory merge. If this feature is not enabled, then all these " + "write buffers are flushed to L0 as separate files and this " + "increases read amplification because a get request has to check " + "in all of these files. Also, an in-memory merge may result in " + "writing less data to storage if there are duplicate records in" + " each of these individual write buffers."); + +DEFINE_int32(max_write_buffer_number_to_maintain, + ROCKSDB_NAMESPACE::Options().max_write_buffer_number_to_maintain, + "The total maximum number of write buffers to maintain in memory " + "including copies of buffers that have already been flushed. " + "Unlike max_write_buffer_number, this parameter does not affect " + "flushing. This controls the minimum amount of write history " + "that will be available in memory for conflict checking when " + "Transactions are used. If this value is too low, some " + "transactions may fail at commit time due to not being able to " + "determine whether there were any write conflicts. Setting this " + "value to 0 will cause write buffers to be freed immediately " + "after they are flushed. If this value is set to -1, " + "'max_write_buffer_number' will be used."); + +DEFINE_int64(max_write_buffer_size_to_maintain, + ROCKSDB_NAMESPACE::Options().max_write_buffer_size_to_maintain, + "The total maximum size of write buffers to maintain in memory " + "including copies of buffers that have already been flushed. " + "Unlike max_write_buffer_number, this parameter does not affect " + "flushing. This controls the minimum amount of write history " + "that will be available in memory for conflict checking when " + "Transactions are used. If this value is too low, some " + "transactions may fail at commit time due to not being able to " + "determine whether there were any write conflicts. Setting this " + "value to 0 will cause write buffers to be freed immediately " + "after they are flushed. If this value is set to -1, " + "'max_write_buffer_number' will be used."); + +DEFINE_double(memtable_prefix_bloom_size_ratio, + ROCKSDB_NAMESPACE::Options().memtable_prefix_bloom_size_ratio, + "creates prefix blooms for memtables, each with size " + "`write_buffer_size * memtable_prefix_bloom_size_ratio`."); + +DEFINE_bool(memtable_whole_key_filtering, + ROCKSDB_NAMESPACE::Options().memtable_whole_key_filtering, + "Enable whole key filtering in memtables."); + +DEFINE_int32(open_files, ROCKSDB_NAMESPACE::Options().max_open_files, + "Maximum number of files to keep open at the same time " + "(use default if == 0)"); + +DEFINE_int64(compressed_cache_size, 0, + "Number of bytes to use as a cache of compressed data." + " 0 means use default settings."); + +DEFINE_int32( + compressed_cache_numshardbits, -1, + "Number of shards for the compressed block cache is 2 ** " + "compressed_cache_numshardbits. Negative value means default settings. " + "This is applied only if compressed_cache_size is greater than 0."); + +DEFINE_int32(compaction_style, ROCKSDB_NAMESPACE::Options().compaction_style, + ""); + +DEFINE_int32(compaction_pri, ROCKSDB_NAMESPACE::Options().compaction_pri, + "Which file from a level should be picked to merge to the next " + "level in level-based compaction"); + +DEFINE_int32(num_levels, ROCKSDB_NAMESPACE::Options().num_levels, + "Number of levels in the DB"); + +DEFINE_int32(level0_file_num_compaction_trigger, + ROCKSDB_NAMESPACE::Options().level0_file_num_compaction_trigger, + "Level0 compaction start trigger"); + +DEFINE_int32(level0_slowdown_writes_trigger, + ROCKSDB_NAMESPACE::Options().level0_slowdown_writes_trigger, + "Number of files in level-0 that will slow down writes"); + +DEFINE_int32(level0_stop_writes_trigger, + ROCKSDB_NAMESPACE::Options().level0_stop_writes_trigger, + "Number of files in level-0 that will trigger put stop."); + +DEFINE_int32(block_size, + static_cast<int32_t>( + ROCKSDB_NAMESPACE::BlockBasedTableOptions().block_size), + "Number of bytes in a block."); + +DEFINE_int32(format_version, + static_cast<int32_t>( + ROCKSDB_NAMESPACE::BlockBasedTableOptions().format_version), + "Format version of SST files."); + +DEFINE_int32( + index_block_restart_interval, + ROCKSDB_NAMESPACE::BlockBasedTableOptions().index_block_restart_interval, + "Number of keys between restart points " + "for delta encoding of keys in index block."); + +DEFINE_bool(disable_auto_compactions, + ROCKSDB_NAMESPACE::Options().disable_auto_compactions, + "If true, RocksDB internally will not trigger compactions."); + +DEFINE_int32(max_background_compactions, + ROCKSDB_NAMESPACE::Options().max_background_compactions, + "The maximum number of concurrent background compactions " + "that can occur in parallel."); + +DEFINE_int32(num_bottom_pri_threads, 0, + "The number of threads in the bottom-priority thread pool (used " + "by universal compaction only)."); + +DEFINE_int32(compaction_thread_pool_adjust_interval, 0, + "The interval (in milliseconds) to adjust compaction thread pool " + "size. Don't change it periodically if the value is 0."); + +DEFINE_int32(compaction_thread_pool_variations, 2, + "Range of background thread pool size variations when adjusted " + "periodically."); + +DEFINE_int32(max_background_flushes, + ROCKSDB_NAMESPACE::Options().max_background_flushes, + "The maximum number of concurrent background flushes " + "that can occur in parallel."); + +DEFINE_int32(universal_size_ratio, 0, + "The ratio of file sizes that trigger" + " compaction in universal style"); + +DEFINE_int32(universal_min_merge_width, 0, + "The minimum number of files to " + "compact in universal style compaction"); + +DEFINE_int32(universal_max_merge_width, 0, + "The max number of files to compact" + " in universal style compaction"); + +DEFINE_int32(universal_max_size_amplification_percent, 0, + "The max size amplification for universal style compaction"); + +DEFINE_int32(clear_column_family_one_in, 1000000, + "With a chance of 1/N, delete a column family and then recreate " + "it again. If N == 0, never drop/create column families. " + "When test_batches_snapshots is true, this flag has no effect"); + +DEFINE_int32(get_live_files_one_in, 1000000, + "With a chance of 1/N, call GetLiveFiles to verify if it returns " + "correctly. If N == 0, do not call the interface."); + +DEFINE_int32( + get_sorted_wal_files_one_in, 1000000, + "With a chance of 1/N, call GetSortedWalFiles to verify if it returns " + "correctly. (Note that this API may legitimately return an error.) If N == " + "0, do not call the interface."); + +DEFINE_int32( + get_current_wal_file_one_in, 1000000, + "With a chance of 1/N, call GetCurrentWalFile to verify if it returns " + "correctly. (Note that this API may legitimately return an error.) If N == " + "0, do not call the interface."); + +DEFINE_int32(set_options_one_in, 0, + "With a chance of 1/N, change some random options"); + +DEFINE_int32(set_in_place_one_in, 0, + "With a chance of 1/N, toggle in place support option"); + +DEFINE_int64(cache_size, 2LL * KB * KB * KB, + "Number of bytes to use as a cache of uncompressed data."); + +DEFINE_int32(cache_numshardbits, 6, + "Number of shards for the block cache" + " is 2 ** cache_numshardbits. Negative means use default settings." + " This is applied only if FLAGS_cache_size is greater than 0."); + +DEFINE_bool(cache_index_and_filter_blocks, false, + "True if indexes/filters should be cached in block cache."); + +DEFINE_bool(charge_compression_dictionary_building_buffer, false, + "Setting for " + "CacheEntryRoleOptions::charged of " + "CacheEntryRole::kCompressionDictionaryBuildingBuffer"); + +DEFINE_bool(charge_filter_construction, false, + "Setting for " + "CacheEntryRoleOptions::charged of " + "CacheEntryRole::kFilterConstruction"); + +DEFINE_bool(charge_table_reader, false, + "Setting for " + "CacheEntryRoleOptions::charged of " + "CacheEntryRole::kBlockBasedTableReader"); + +DEFINE_bool(charge_file_metadata, false, + "Setting for " + "CacheEntryRoleOptions::charged of " + "kFileMetadata"); + +DEFINE_bool(charge_blob_cache, false, + "Setting for " + "CacheEntryRoleOptions::charged of " + "kBlobCache"); + +DEFINE_int32( + top_level_index_pinning, + static_cast<int32_t>(ROCKSDB_NAMESPACE::PinningTier::kFallback), + "Type of pinning for top-level indexes into metadata partitions (see " + "`enum PinningTier` in table.h)"); + +DEFINE_int32( + partition_pinning, + static_cast<int32_t>(ROCKSDB_NAMESPACE::PinningTier::kFallback), + "Type of pinning for metadata partitions (see `enum PinningTier` in " + "table.h)"); + +DEFINE_int32( + unpartitioned_pinning, + static_cast<int32_t>(ROCKSDB_NAMESPACE::PinningTier::kFallback), + "Type of pinning for unpartitioned metadata blocks (see `enum PinningTier` " + "in table.h)"); + +DEFINE_string(cache_type, "lru_cache", "Type of block cache."); + +DEFINE_uint64(subcompactions, 1, + "Maximum number of subcompactions to divide L0-L1 compactions " + "into."); + +DEFINE_uint64(periodic_compaction_seconds, 1000, + "Files older than this value will be picked up for compaction."); + +DEFINE_uint64(compaction_ttl, 1000, + "Files older than TTL will be compacted to the next level."); + +DEFINE_bool(allow_concurrent_memtable_write, false, + "Allow multi-writers to update mem tables in parallel."); + +DEFINE_double(experimental_mempurge_threshold, 0.0, + "Maximum estimated useful payload that triggers a " + "mempurge process to collect memtable garbage bytes."); + +DEFINE_bool(enable_write_thread_adaptive_yield, true, + "Use a yielding spin loop for brief writer thread waits."); + +#ifndef ROCKSDB_LITE +// Options for StackableDB-based BlobDB +DEFINE_bool(use_blob_db, false, "[Stacked BlobDB] Use BlobDB."); + +DEFINE_uint64( + blob_db_min_blob_size, + ROCKSDB_NAMESPACE::blob_db::BlobDBOptions().min_blob_size, + "[Stacked BlobDB] Smallest blob to store in a file. Blobs " + "smaller than this will be inlined with the key in the LSM tree."); + +DEFINE_uint64( + blob_db_bytes_per_sync, + ROCKSDB_NAMESPACE::blob_db::BlobDBOptions().bytes_per_sync, + "[Stacked BlobDB] Sync blob files once per every N bytes written."); + +DEFINE_uint64(blob_db_file_size, + ROCKSDB_NAMESPACE::blob_db::BlobDBOptions().blob_file_size, + "[Stacked BlobDB] Target size of each blob file."); + +DEFINE_bool( + blob_db_enable_gc, + ROCKSDB_NAMESPACE::blob_db::BlobDBOptions().enable_garbage_collection, + "[Stacked BlobDB] Enable BlobDB garbage collection."); + +DEFINE_double( + blob_db_gc_cutoff, + ROCKSDB_NAMESPACE::blob_db::BlobDBOptions().garbage_collection_cutoff, + "[Stacked BlobDB] Cutoff ratio for BlobDB garbage collection."); +#endif // !ROCKSDB_LITE + +// Options for integrated BlobDB +DEFINE_bool(allow_setting_blob_options_dynamically, false, + "[Integrated BlobDB] Allow setting blob options dynamically."); + +DEFINE_bool( + enable_blob_files, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions().enable_blob_files, + "[Integrated BlobDB] Enable writing large values to separate blob files."); + +DEFINE_uint64(min_blob_size, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions().min_blob_size, + "[Integrated BlobDB] The size of the smallest value to be stored " + "separately in a blob file."); + +DEFINE_uint64(blob_file_size, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions().blob_file_size, + "[Integrated BlobDB] The size limit for blob files."); + +DEFINE_string(blob_compression_type, "none", + "[Integrated BlobDB] The compression algorithm to use for large " + "values stored in blob files."); + +DEFINE_bool(enable_blob_garbage_collection, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions() + .enable_blob_garbage_collection, + "[Integrated BlobDB] Enable blob garbage collection."); + +DEFINE_double(blob_garbage_collection_age_cutoff, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions() + .blob_garbage_collection_age_cutoff, + "[Integrated BlobDB] The cutoff in terms of blob file age for " + "garbage collection."); + +DEFINE_double(blob_garbage_collection_force_threshold, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions() + .blob_garbage_collection_force_threshold, + "[Integrated BlobDB] The threshold for the ratio of garbage in " + "the oldest blob files for forcing garbage collection."); + +DEFINE_uint64(blob_compaction_readahead_size, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions() + .blob_compaction_readahead_size, + "[Integrated BlobDB] Compaction readahead for blob files."); + +DEFINE_int32( + blob_file_starting_level, + ROCKSDB_NAMESPACE::AdvancedColumnFamilyOptions().blob_file_starting_level, + "[Integrated BlobDB] Enable writing blob files during flushes and " + "compactions starting from the specified level."); + +DEFINE_bool(use_blob_cache, false, "[Integrated BlobDB] Enable blob cache."); + +DEFINE_bool( + use_shared_block_and_blob_cache, true, + "[Integrated BlobDB] Use a shared backing cache for both block " + "cache and blob cache. It only takes effect if use_blob_cache is enabled."); + +DEFINE_uint64( + blob_cache_size, 2LL * KB * KB * KB, + "[Integrated BlobDB] Number of bytes to use as a cache of blobs. It only " + "takes effect if the block and blob caches are different " + "(use_shared_block_and_blob_cache = false)."); + +DEFINE_int32(blob_cache_numshardbits, 6, + "[Integrated BlobDB] Number of shards for the blob cache is 2 ** " + "blob_cache_numshardbits. Negative means use default settings. " + "It only takes effect if blob_cache_size is greater than 0, and " + "the block and blob caches are different " + "(use_shared_block_and_blob_cache = false)."); + +DEFINE_int32(prepopulate_blob_cache, 0, + "[Integrated BlobDB] Pre-populate hot/warm blobs in blob cache. 0 " + "to disable and 1 to insert during flush."); + +DEFINE_bool(enable_tiered_storage, false, "Set last_level_temperature"); + +DEFINE_int64(preclude_last_level_data_seconds, 0, + "Preclude data from the last level. Used with tiered storage " + "feature to preclude new data from comacting to the last level."); + +DEFINE_int64( + preserve_internal_time_seconds, 0, + "Preserve internal time information which is attached to each SST."); + +static const bool FLAGS_subcompactions_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_subcompactions, &ValidateUint32Range); + +static bool ValidateInt32Positive(const char* flagname, int32_t value) { + if (value < 0) { + fprintf(stderr, "Invalid value for --%s: %d, must be >=0\n", flagname, + value); + return false; + } + return true; +} +DEFINE_int32(reopen, 10, "Number of times database reopens"); +static const bool FLAGS_reopen_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_reopen, &ValidateInt32Positive); + +DEFINE_double(bloom_bits, 10, + "Bloom filter bits per key. " + "Negative means use default settings."); + +DEFINE_int32( + ribbon_starting_level, 999, + "Use Bloom filter on levels below specified and Ribbon beginning on level " + "specified. Flush is considered level -1. 999 or more -> always Bloom. 0 " + "-> Ribbon except Bloom for flush. -1 -> always Ribbon."); + +DEFINE_bool(partition_filters, false, + "use partitioned filters " + "for block-based table"); + +DEFINE_bool( + optimize_filters_for_memory, + ROCKSDB_NAMESPACE::BlockBasedTableOptions().optimize_filters_for_memory, + "Minimize memory footprint of filters"); + +DEFINE_bool( + detect_filter_construct_corruption, + ROCKSDB_NAMESPACE::BlockBasedTableOptions() + .detect_filter_construct_corruption, + "Detect corruption during new Bloom Filter and Ribbon Filter construction"); + +DEFINE_int32( + index_type, + static_cast<int32_t>( + ROCKSDB_NAMESPACE::BlockBasedTableOptions().index_type), + "Type of block-based table index (see `enum IndexType` in table.h)"); + +DEFINE_int32( + data_block_index_type, + static_cast<int32_t>( + ROCKSDB_NAMESPACE::BlockBasedTableOptions().data_block_index_type), + "Index type for data blocks (see `enum DataBlockIndexType` in table.h)"); + +DEFINE_string(db, "", "Use the db with the following name."); + +DEFINE_string(secondaries_base, "", + "Use this path as the base path for secondary instances."); + +DEFINE_bool(test_secondary, false, + "If true, start an additional secondary instance which can be used " + "for verification."); + +DEFINE_string( + expected_values_dir, "", + "Dir where files containing info about the latest/historical values will " + "be stored. If provided and non-empty, the DB state will be verified " + "against values from these files after recovery. --max_key and " + "--column_family must be kept the same across invocations of this program " + "that use the same --expected_values_dir. Currently historical values are " + "only tracked when --sync_fault_injection is set. See --seed and " + "--nooverwritepercent for further requirements."); + +DEFINE_bool(verify_checksum, false, + "Verify checksum for every block read from storage"); + +DEFINE_bool(mmap_read, ROCKSDB_NAMESPACE::Options().allow_mmap_reads, + "Allow reads to occur via mmap-ing files"); + +DEFINE_bool(mmap_write, ROCKSDB_NAMESPACE::Options().allow_mmap_writes, + "Allow writes to occur via mmap-ing files"); + +DEFINE_bool(use_direct_reads, ROCKSDB_NAMESPACE::Options().use_direct_reads, + "Use O_DIRECT for reading data"); + +DEFINE_bool(use_direct_io_for_flush_and_compaction, + ROCKSDB_NAMESPACE::Options().use_direct_io_for_flush_and_compaction, + "Use O_DIRECT for writing data"); + +DEFINE_bool(mock_direct_io, false, + "Mock direct IO by not using O_DIRECT for direct IO read"); + +DEFINE_bool(statistics, false, "Create database statistics"); + +DEFINE_bool(sync, false, "Sync all writes to disk"); + +DEFINE_bool(use_fsync, false, "If true, issue fsync instead of fdatasync"); + +DEFINE_uint64(bytes_per_sync, ROCKSDB_NAMESPACE::Options().bytes_per_sync, + "If nonzero, sync SST file data incrementally after every " + "`bytes_per_sync` bytes are written"); + +DEFINE_uint64(wal_bytes_per_sync, + ROCKSDB_NAMESPACE::Options().wal_bytes_per_sync, + "If nonzero, sync WAL file data incrementally after every " + "`bytes_per_sync` bytes are written"); + +DEFINE_int32(kill_random_test, 0, + "If non-zero, kill at various points in source code with " + "probability 1/this"); +static const bool FLAGS_kill_random_test_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_kill_random_test, &ValidateInt32Positive); + +DEFINE_string(kill_exclude_prefixes, "", + "If non-empty, kill points with prefix in the list given will be" + " skipped. Items are comma-separated."); +extern std::vector<std::string> rocksdb_kill_exclude_prefixes; + +DEFINE_bool(disable_wal, false, "If true, do not write WAL for write."); + +DEFINE_uint64(recycle_log_file_num, + ROCKSDB_NAMESPACE::Options().recycle_log_file_num, + "Number of old WAL files to keep around for later recycling"); + +DEFINE_int64(target_file_size_base, + ROCKSDB_NAMESPACE::Options().target_file_size_base, + "Target level-1 file size for compaction"); + +DEFINE_int32(target_file_size_multiplier, 1, + "A multiplier to compute target level-N file size (N >= 2)"); + +DEFINE_uint64(max_bytes_for_level_base, + ROCKSDB_NAMESPACE::Options().max_bytes_for_level_base, + "Max bytes for level-1"); + +DEFINE_double(max_bytes_for_level_multiplier, 2, + "A multiplier to compute max bytes for level-N (N >= 2)"); + +DEFINE_int32(range_deletion_width, 10, + "The width of the range deletion intervals."); + +DEFINE_uint64(rate_limiter_bytes_per_sec, 0, "Set options.rate_limiter value."); + +DEFINE_bool(rate_limit_bg_reads, false, + "Use options.rate_limiter on compaction reads"); + +DEFINE_bool(rate_limit_user_ops, false, + "When true use Env::IO_USER priority level to charge internal rate " + "limiter for reads associated with user operations."); + +DEFINE_bool(rate_limit_auto_wal_flush, false, + "When true use Env::IO_USER priority level to charge internal rate " + "limiter for automatic WAL flush (`Options::manual_wal_flush` == " + "false) after the user " + "write operation."); + +DEFINE_uint64(sst_file_manager_bytes_per_sec, 0, + "Set `Options::sst_file_manager` to delete at this rate. By " + "default the deletion rate is unbounded."); + +DEFINE_uint64(sst_file_manager_bytes_per_truncate, 0, + "Set `Options::sst_file_manager` to delete in chunks of this " + "many bytes. By default whole files will be deleted."); + +DEFINE_bool(use_txn, false, + "Use TransactionDB. Currently the default write policy is " + "TxnDBWritePolicy::WRITE_PREPARED"); + +DEFINE_uint64(txn_write_policy, 0, + "The transaction write policy. Default is " + "TxnDBWritePolicy::WRITE_COMMITTED. Note that this should not be " + "changed accross crashes."); + +DEFINE_bool(unordered_write, false, + "Turn on the unordered_write feature. This options is currently " + "tested only in combination with use_txn=true and " + "txn_write_policy=TxnDBWritePolicy::WRITE_PREPARED."); + +DEFINE_int32(backup_one_in, 0, + "If non-zero, then CreateNewBackup() will be called once for " + "every N operations on average. 0 indicates CreateNewBackup() " + "is disabled."); + +DEFINE_uint64(backup_max_size, 100 * 1024 * 1024, + "If non-zero, skip checking backup/restore when DB size in " + "bytes exceeds this setting."); + +DEFINE_int32(checkpoint_one_in, 0, + "If non-zero, then CreateCheckpoint() will be called once for " + "every N operations on average. 0 indicates CreateCheckpoint() " + "is disabled."); + +DEFINE_int32(ingest_external_file_one_in, 0, + "If non-zero, then IngestExternalFile() will be called once for " + "every N operations on average. 0 indicates IngestExternalFile() " + "is disabled."); + +DEFINE_int32(ingest_external_file_width, 100, + "The width of the ingested external files."); + +DEFINE_int32(compact_files_one_in, 0, + "If non-zero, then CompactFiles() will be called once for every N " + "operations on average. 0 indicates CompactFiles() is disabled."); + +DEFINE_int32(compact_range_one_in, 0, + "If non-zero, then CompactRange() will be called once for every N " + "operations on average. 0 indicates CompactRange() is disabled."); + +DEFINE_int32(mark_for_compaction_one_file_in, 0, + "A `TablePropertiesCollectorFactory` will be registered, which " + "creates a `TablePropertiesCollector` with `NeedCompact()` " + "returning true once for every N files on average. 0 or negative " + "mean `NeedCompact()` always returns false."); + +DEFINE_int32(flush_one_in, 0, + "If non-zero, then Flush() will be called once for every N ops " + "on average. 0 indicates calls to Flush() are disabled."); + +DEFINE_int32(pause_background_one_in, 0, + "If non-zero, then PauseBackgroundWork()+Continue will be called " + "once for every N ops on average. 0 disables."); + +DEFINE_int32(compact_range_width, 10000, + "The width of the ranges passed to CompactRange()."); + +DEFINE_int32(acquire_snapshot_one_in, 0, + "If non-zero, then acquires a snapshot once every N operations on " + "average."); + +DEFINE_bool(compare_full_db_state_snapshot, false, + "If set we compare state of entire db (in one of the threads) with" + "each snapshot."); + +DEFINE_uint64(snapshot_hold_ops, 0, + "If non-zero, then releases snapshots N operations after they're " + "acquired."); + +DEFINE_bool(long_running_snapshots, false, + "If set, hold on some some snapshots for much longer time."); + +DEFINE_bool(use_multiget, false, + "If set, use the batched MultiGet API for reads"); + +static bool ValidateInt32Percent(const char* flagname, int32_t value) { + if (value < 0 || value > 100) { + fprintf(stderr, "Invalid value for --%s: %d, 0<= pct <=100 \n", flagname, + value); + return false; + } + return true; +} + +DEFINE_int32(readpercent, 10, + "Ratio of reads to total workload (expressed as a percentage)"); +static const bool FLAGS_readpercent_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_readpercent, &ValidateInt32Percent); + +DEFINE_int32(prefixpercent, 20, + "Ratio of prefix iterators to total workload (expressed as a" + " percentage)"); +static const bool FLAGS_prefixpercent_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_prefixpercent, &ValidateInt32Percent); + +DEFINE_int32(writepercent, 45, + "Ratio of writes to total workload (expressed as a percentage)"); +static const bool FLAGS_writepercent_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_writepercent, &ValidateInt32Percent); + +DEFINE_int32(delpercent, 15, + "Ratio of deletes to total workload (expressed as a percentage)"); +static const bool FLAGS_delpercent_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_delpercent, &ValidateInt32Percent); + +DEFINE_int32(delrangepercent, 0, + "Ratio of range deletions to total workload (expressed as a " + "percentage). Cannot be used with test_batches_snapshots"); +static const bool FLAGS_delrangepercent_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_delrangepercent, &ValidateInt32Percent); + +DEFINE_int32(nooverwritepercent, 60, + "Ratio of keys without overwrite to total workload (expressed as " + "a percentage). When --expected_values_dir is nonempty, must " + "keep this value constant across invocations."); +static const bool FLAGS_nooverwritepercent_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_nooverwritepercent, &ValidateInt32Percent); + +DEFINE_int32(iterpercent, 10, + "Ratio of iterations to total workload" + " (expressed as a percentage)"); +static const bool FLAGS_iterpercent_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_iterpercent, &ValidateInt32Percent); + +DEFINE_uint64(num_iterations, 10, "Number of iterations per MultiIterate run"); +static const bool FLAGS_num_iterations_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_num_iterations, &ValidateUint32Range); + +DEFINE_int32( + customopspercent, 0, + "Ratio of custom operations to total workload (expressed as a percentage)"); + +DEFINE_string(compression_type, "snappy", + "Algorithm to use to compress the database"); + +DEFINE_int32(compression_max_dict_bytes, 0, + "Maximum size of dictionary used to prime the compression " + "library."); + +DEFINE_int32(compression_zstd_max_train_bytes, 0, + "Maximum size of training data passed to zstd's dictionary " + "trainer."); + +DEFINE_int32(compression_parallel_threads, 1, + "Number of threads for parallel compression."); + +DEFINE_uint64(compression_max_dict_buffer_bytes, 0, + "Buffering limit for SST file data to sample for dictionary " + "compression."); + +DEFINE_bool( + compression_use_zstd_dict_trainer, true, + "Use zstd's trainer to generate dictionary. If the options is false, " + "zstd's finalizeDictionary() API is used to generate dictionary. " + "ZSTD 1.4.5+ is required. If ZSTD 1.4.5+ is not linked with the binary, " + "this flag will have the default value true."); + +DEFINE_string(bottommost_compression_type, "disable", + "Algorithm to use to compress bottommost level of the database. " + "\"disable\" means disabling the feature"); + +DEFINE_string(checksum_type, "kCRC32c", "Algorithm to use to checksum blocks"); + +DEFINE_string(env_uri, "", + "URI for env lookup. Mutually exclusive with --fs_uri"); + +DEFINE_string(fs_uri, "", + "URI for registry Filesystem lookup. Mutually exclusive" + " with --env_uri." + " Creates a default environment with the specified filesystem."); + +DEFINE_uint64(ops_per_thread, 1200000, "Number of operations per thread."); +static const bool FLAGS_ops_per_thread_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_ops_per_thread, &ValidateUint32Range); + +DEFINE_uint64(log2_keys_per_lock, 2, "Log2 of number of keys per lock"); +static const bool FLAGS_log2_keys_per_lock_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_log2_keys_per_lock, &ValidateUint32Range); + +DEFINE_uint64(max_manifest_file_size, 16384, "Maximum size of a MANIFEST file"); + +DEFINE_bool(in_place_update, false, "On true, does inplace update in memtable"); + +DEFINE_string(memtablerep, "skip_list", ""); + +inline static bool ValidatePrefixSize(const char* flagname, int32_t value) { + if (value < -1 || value > 8) { + fprintf(stderr, "Invalid value for --%s: %d. -1 <= PrefixSize <= 8\n", + flagname, value); + return false; + } + return true; +} +DEFINE_int32(prefix_size, 7, + "Control the prefix size for HashSkipListRep. " + "-1 is disabled."); +static const bool FLAGS_prefix_size_dummy __attribute__((__unused__)) = + RegisterFlagValidator(&FLAGS_prefix_size, &ValidatePrefixSize); + +DEFINE_bool(use_merge, false, + "On true, replaces all writes with a Merge " + "that behaves like a Put"); + +DEFINE_uint32(use_put_entity_one_in, 0, + "If greater than zero, PutEntity will be used once per every N " + "write ops on average."); + +DEFINE_bool(use_full_merge_v1, false, + "On true, use a merge operator that implement the deprecated " + "version of FullMerge"); + +DEFINE_int32(sync_wal_one_in, 0, + "If non-zero, then SyncWAL() will be called once for every N ops " + "on average. 0 indicates that calls to SyncWAL() are disabled."); + +DEFINE_bool(avoid_unnecessary_blocking_io, + ROCKSDB_NAMESPACE::Options().avoid_unnecessary_blocking_io, + "If true, some expensive cleaning up operations will be moved from " + "user reads to high-pri background threads."); + +DEFINE_bool(write_dbid_to_manifest, + ROCKSDB_NAMESPACE::Options().write_dbid_to_manifest, + "Write DB_ID to manifest"); + +DEFINE_bool(avoid_flush_during_recovery, + ROCKSDB_NAMESPACE::Options().avoid_flush_during_recovery, + "Avoid flush during recovery"); + +DEFINE_uint64(max_write_batch_group_size_bytes, + ROCKSDB_NAMESPACE::Options().max_write_batch_group_size_bytes, + "Max write batch group size"); + +DEFINE_bool(level_compaction_dynamic_level_bytes, + ROCKSDB_NAMESPACE::Options().level_compaction_dynamic_level_bytes, + "Use dynamic level"); + +DEFINE_int32(verify_checksum_one_in, 0, + "If non-zero, then DB::VerifyChecksum() will be called to do" + " checksum verification of all the files in the database once for" + " every N ops on average. 0 indicates that calls to" + " VerifyChecksum() are disabled."); +DEFINE_int32(verify_db_one_in, 0, + "If non-zero, call VerifyDb() once for every N ops. 0 indicates " + "that VerifyDb() will not be called in OperateDb(). Note that " + "enabling this can slow down tests."); + +DEFINE_int32(continuous_verification_interval, 1000, + "While test is running, verify db every N milliseconds. 0 " + "disables continuous verification."); + +DEFINE_int32(approximate_size_one_in, 64, + "If non-zero, DB::GetApproximateSizes() will be called against" + " random key ranges."); + +DEFINE_int32(read_fault_one_in, 1000, + "On non-zero, enables fault injection on read"); + +DEFINE_int32(get_property_one_in, 1000, + "If non-zero, then DB::GetProperty() will be called to get various" + " properties for every N ops on average. 0 indicates that" + " GetProperty() will be not be called."); + +DEFINE_bool(sync_fault_injection, false, + "If true, FaultInjectionTestFS will be used for write operations, " + "and unsynced data in DB will lost after crash. In such a case we " + "track DB changes in a trace file (\"*.trace\") in " + "--expected_values_dir for verifying there are no holes in the " + "recovered data."); + +DEFINE_bool(best_efforts_recovery, false, + "If true, use best efforts recovery."); +DEFINE_bool(skip_verifydb, false, + "If true, skip VerifyDb() calls and Get()/Iterator verifications" + "against expected state."); + +DEFINE_bool(enable_compaction_filter, false, + "If true, configures a compaction filter that returns a kRemove " + "decision for deleted keys."); + +DEFINE_bool(paranoid_file_checks, true, + "After writing every SST file, reopen it and read all the keys " + "and validate checksums"); + +DEFINE_bool(fail_if_options_file_error, false, + "Fail operations that fail to detect or properly persist options " + "file."); + +DEFINE_uint64(batch_protection_bytes_per_key, 0, + "If nonzero, enables integrity protection in `WriteBatch` at the " + "specified number of bytes per key. Currently the only supported " + "nonzero value is eight."); + +DEFINE_uint32( + memtable_protection_bytes_per_key, 0, + "If nonzero, enables integrity protection in memtable entries at the " + "specified number of bytes per key. Currently the supported " + "nonzero values are 1, 2, 4 and 8."); + +DEFINE_string(file_checksum_impl, "none", + "Name of an implementation for file_checksum_gen_factory, or " + "\"none\" for null."); + +DEFINE_int32(write_fault_one_in, 0, + "On non-zero, enables fault injection on write"); + +DEFINE_uint64(user_timestamp_size, 0, + "Number of bytes for a user-defined timestamp. Currently, only " + "8-byte is supported"); + +DEFINE_int32(open_metadata_write_fault_one_in, 0, + "On non-zero, enables fault injection on file metadata write " + "during DB reopen."); + +#ifndef ROCKSDB_LITE +DEFINE_string(secondary_cache_uri, "", + "Full URI for creating a customized secondary cache object"); +DEFINE_int32(secondary_cache_fault_one_in, 0, + "On non-zero, enables fault injection in secondary cache inserts" + " and lookups"); +#endif // ROCKSDB_LITE +DEFINE_int32(open_write_fault_one_in, 0, + "On non-zero, enables fault injection on file writes " + "during DB reopen."); +DEFINE_int32(open_read_fault_one_in, 0, + "On non-zero, enables fault injection on file reads " + "during DB reopen."); +DEFINE_int32(injest_error_severity, 1, + "The severity of the injested IO Error. 1 is soft error (e.g. " + "retryable error), 2 is fatal error, and the default is " + "retryable error."); +DEFINE_int32(prepopulate_block_cache, + static_cast<int32_t>(ROCKSDB_NAMESPACE::BlockBasedTableOptions:: + PrepopulateBlockCache::kDisable), + "Options related to cache warming (see `enum " + "PrepopulateBlockCache` in table.h)"); + +DEFINE_bool(two_write_queues, false, + "Set to true to enable two write queues. Default: false"); +#ifndef ROCKSDB_LITE + +DEFINE_bool(use_only_the_last_commit_time_batch_for_recovery, false, + "If true, the commit-time write batch will not be immediately " + "inserted into the memtables. Default: false"); + +DEFINE_uint64( + wp_snapshot_cache_bits, 7ull, + "Number of bits to represent write-prepared transaction db's snapshot " + "cache. Default: 7 (128 entries)"); + +DEFINE_uint64(wp_commit_cache_bits, 23ull, + "Number of bits to represent write-prepared transaction db's " + "commit cache. Default: 23 (8M entries)"); +#endif // !ROCKSDB_LITE + +DEFINE_bool(adaptive_readahead, false, + "Carry forward internal auto readahead size from one file to next " + "file at each level during iteration"); +DEFINE_bool( + async_io, false, + "Does asynchronous prefetching when internal auto readahead is enabled"); + +DEFINE_string(wal_compression, "none", + "Algorithm to use for WAL compression. none to disable."); + +DEFINE_bool( + verify_sst_unique_id_in_manifest, false, + "Enable DB options `verify_sst_unique_id_in_manifest`, if true, during " + "DB-open try verifying the SST unique id between MANIFEST and SST " + "properties."); + +DEFINE_int32( + create_timestamped_snapshot_one_in, 0, + "On non-zero, create timestamped snapshots upon transaction commits."); + +DEFINE_bool(allow_data_in_errors, + ROCKSDB_NAMESPACE::Options().allow_data_in_errors, + "If true, allow logging data, e.g. key, value in LOG files."); + +DEFINE_int32(verify_iterator_with_expected_state_one_in, 0, + "If non-zero, when TestIterate() is to be called, there is a " + "1/verify_iterator_with_expected_state_one_in " + "chance that the iterator is verified against the expected state " + "file, instead of comparing keys between two iterators."); + +DEFINE_uint64(readahead_size, 0, "Iterator readahead size"); +DEFINE_uint64(initial_auto_readahead_size, 0, + "Initial auto readahead size for prefetching during Iteration"); +DEFINE_uint64(max_auto_readahead_size, 0, + "Max auto readahead size for prefetching during Iteration"); +DEFINE_uint64( + num_file_reads_for_auto_readahead, 0, + "Num of sequential reads to enable auto prefetching during Iteration"); + +DEFINE_bool( + preserve_unverified_changes, false, + "DB files of the current run will all be preserved in `FLAGS_db`. DB files " + "from the last run will be preserved in `FLAGS_db/unverified` until the " + "first verification succeeds. Expected state files from the last run will " + "be preserved similarly under `FLAGS_expected_values_dir/unverified` when " + "`--expected_values_dir` is nonempty."); + +DEFINE_uint64(stats_dump_period_sec, + ROCKSDB_NAMESPACE::Options().stats_dump_period_sec, + "Gap between printing stats to log in seconds"); + +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_listener.cc b/src/rocksdb/db_stress_tool/db_stress_listener.cc new file mode 100644 index 000000000..578f21c41 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_listener.cc @@ -0,0 +1,191 @@ +// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). + +#include "db_stress_tool/db_stress_listener.h" + +#include <cstdint> + +#include "file/file_util.h" +#include "rocksdb/file_system.h" +#include "util/coding_lean.h" + +namespace ROCKSDB_NAMESPACE { + +#ifdef GFLAGS +#ifndef ROCKSDB_LITE + +// TODO: consider using expected_values_dir instead, but this is more +// convenient for now. +UniqueIdVerifier::UniqueIdVerifier(const std::string& db_name, Env* env) + : path_(db_name + "/.unique_ids") { + // We expect such a small number of files generated during this test + // (thousands?), checking full 192-bit IDs for uniqueness is a very + // weak check. For a stronger check, we pick a specific 64-bit + // subsequence from the ID to check for uniqueness. All bits of the + // ID should be high quality, and 64 bits should be unique with + // very good probability for the quantities in this test. + offset_ = Random::GetTLSInstance()->Uniform(17); // 0 to 16 + + const std::shared_ptr<FileSystem> fs = env->GetFileSystem(); + IOOptions opts; + + Status st = fs->CreateDirIfMissing(db_name, opts, nullptr); + if (!st.ok()) { + fprintf(stderr, "Failed to create directory %s: %s\n", db_name.c_str(), + st.ToString().c_str()); + exit(1); + } + + // Avoid relying on ReopenWritableFile which is not supported by all + // file systems. Create a new file and copy the old file contents to it. + std::string tmp_path = path_ + ".tmp"; + st = fs->FileExists(tmp_path, opts, /*dbg*/ nullptr); + if (st.IsNotFound()) { + st = fs->RenameFile(path_, tmp_path, opts, /*dbg*/ nullptr); + // Either it should succeed or fail because src path doesn't exist + assert(st.ok() || st.IsPathNotFound()); + } else { + // If path_ and tmp_path both exist, retain tmp_path as its + // guaranteed to be more complete. The order of operations are - + // 1. Rename path_ to tmp_path + // 2. Parse tmp_path contents + // 3. Create path_ + // 4. Copy tmp_path contents to path_ + // 5. Delete tmp_path + st = fs->DeleteFile(path_, opts, /*dbg*/ nullptr); + assert(st.ok() || st.IsPathNotFound()); + } + + uint64_t size = 0; + { + std::unique_ptr<FSSequentialFile> reader; + Status s = fs->NewSequentialFile(tmp_path, FileOptions(), &reader, + /*dbg*/ nullptr); + if (s.ok()) { + // Load from file + std::string id(24U, '\0'); + Slice result; + for (;;) { + s = reader->Read(id.size(), opts, &result, &id[0], /*dbg*/ nullptr); + if (!s.ok()) { + fprintf(stderr, "Error reading unique id file: %s\n", + s.ToString().c_str()); + assert(false); + } + if (result.size() < id.size()) { + // EOF + if (result.size() != 0) { + // Corrupt file. Not a DB bug but could happen if OS doesn't provide + // good guarantees on process crash. + fprintf(stdout, "Warning: clearing corrupt unique id file\n"); + id_set_.clear(); + reader.reset(); + s = fs->DeleteFile(tmp_path, opts, /*dbg*/ nullptr); + assert(s.ok()); + size = 0; + } + break; + } + size += 24U; + VerifyNoWrite(id); + } + } else { + // Newly created is ok. + // But FileSystem doesn't tell us whether non-existence was the cause of + // the failure. (Issue #9021) + Status s2 = fs->FileExists(tmp_path, opts, /*dbg*/ nullptr); + if (!s2.IsNotFound()) { + fprintf(stderr, "Error opening unique id file: %s\n", + s.ToString().c_str()); + assert(false); + } + size = 0; + } + } + fprintf(stdout, "(Re-)verified %zu unique IDs\n", id_set_.size()); + + std::unique_ptr<FSWritableFile> file_writer; + st = fs->NewWritableFile(path_, FileOptions(), &file_writer, /*dbg*/ nullptr); + if (!st.ok()) { + fprintf(stderr, "Error creating the unique ids file: %s\n", + st.ToString().c_str()); + assert(false); + } + data_file_writer_.reset( + new WritableFileWriter(std::move(file_writer), path_, FileOptions())); + + if (size > 0) { + st = CopyFile(fs.get(), tmp_path, data_file_writer_, size, + /*use_fsync*/ true, /*io_tracer*/ nullptr, + /*temparature*/ Temperature::kHot); + if (!st.ok()) { + fprintf(stderr, "Error copying contents of old unique id file: %s\n", + st.ToString().c_str()); + assert(false); + } + } + st = fs->DeleteFile(tmp_path, opts, /*dbg*/ nullptr); + assert(st.ok() || st.IsPathNotFound()); +} + +UniqueIdVerifier::~UniqueIdVerifier() { + IOStatus s = data_file_writer_->Close(); + assert(s.ok()); +} + +void UniqueIdVerifier::VerifyNoWrite(const std::string& id) { + assert(id.size() == 24); + bool is_new = id_set_.insert(DecodeFixed64(&id[offset_])).second; + if (!is_new) { + fprintf(stderr, + "Duplicate partial unique ID found (offset=%zu, count=%zu)\n", + offset_, id_set_.size()); + assert(false); + } +} + +void UniqueIdVerifier::Verify(const std::string& id) { + assert(id.size() == 24); + std::lock_guard<std::mutex> lock(mutex_); + // If we accumulate more than ~4 million IDs, there would be > 1 in 1M + // natural chance of collision. Thus, simply stop checking at that point. + if (id_set_.size() >= 4294967) { + return; + } + IOStatus s = data_file_writer_->Append(Slice(id)); + if (!s.ok()) { + fprintf(stderr, "Error writing to unique id file: %s\n", + s.ToString().c_str()); + assert(false); + } + s = data_file_writer_->Flush(); + if (!s.ok()) { + fprintf(stderr, "Error flushing unique id file: %s\n", + s.ToString().c_str()); + assert(false); + } + VerifyNoWrite(id); +} + +void DbStressListener::VerifyTableFileUniqueId( + const TableProperties& new_file_properties, const std::string& file_path) { + // Verify unique ID + std::string id; + // Unit tests verify that GetUniqueIdFromTableProperties returns just a + // substring of this, and we're only going to pull out 64 bits, so using + // GetExtendedUniqueIdFromTableProperties is arguably stronger testing here. + Status s = GetExtendedUniqueIdFromTableProperties(new_file_properties, &id); + if (!s.ok()) { + fprintf(stderr, "Error getting SST unique id for %s: %s\n", + file_path.c_str(), s.ToString().c_str()); + assert(false); + } + unique_ids_.Verify(id); +} + +#endif // !ROCKSDB_LITE +#endif // GFLAGS + +} // namespace ROCKSDB_NAMESPACE diff --git a/src/rocksdb/db_stress_tool/db_stress_listener.h b/src/rocksdb/db_stress_tool/db_stress_listener.h new file mode 100644 index 000000000..faced3172 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_listener.h @@ -0,0 +1,271 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). + +#ifdef GFLAGS +#pragma once + +#include <mutex> +#include <unordered_set> + +#include "file/filename.h" +#include "file/writable_file_writer.h" +#include "rocksdb/db.h" +#include "rocksdb/env.h" +#include "rocksdb/file_system.h" +#include "rocksdb/listener.h" +#include "rocksdb/table_properties.h" +#include "rocksdb/unique_id.h" +#include "util/gflags_compat.h" +#include "util/random.h" + +DECLARE_int32(compact_files_one_in); + +namespace ROCKSDB_NAMESPACE { + +#ifndef ROCKSDB_LITE +// Verify across process executions that all seen IDs are unique +class UniqueIdVerifier { + public: + explicit UniqueIdVerifier(const std::string& db_name, Env* env); + ~UniqueIdVerifier(); + + void Verify(const std::string& id); + + private: + void VerifyNoWrite(const std::string& id); + + private: + std::mutex mutex_; + // IDs persisted to a hidden file inside DB dir + std::string path_; + std::unique_ptr<WritableFileWriter> data_file_writer_; + // Starting byte for which 8 bytes to check in memory within 24 byte ID + size_t offset_; + // Working copy of the set of 8 byte pieces + std::unordered_set<uint64_t> id_set_; +}; + +class DbStressListener : public EventListener { + public: + DbStressListener(const std::string& db_name, + const std::vector<DbPath>& db_paths, + const std::vector<ColumnFamilyDescriptor>& column_families, + Env* env) + : db_name_(db_name), + db_paths_(db_paths), + column_families_(column_families), + num_pending_file_creations_(0), + unique_ids_(db_name, env) {} + + const char* Name() const override { return kClassName(); } + static const char* kClassName() { return "DBStressListener"; } + + ~DbStressListener() override { assert(num_pending_file_creations_ == 0); } + void OnFlushCompleted(DB* /*db*/, const FlushJobInfo& info) override { + assert(IsValidColumnFamilyName(info.cf_name)); + VerifyFilePath(info.file_path); + // pretending doing some work here + RandomSleep(); + } + + void OnFlushBegin(DB* /*db*/, + const FlushJobInfo& /*flush_job_info*/) override { + RandomSleep(); + } + + void OnTableFileDeleted(const TableFileDeletionInfo& /*info*/) override { + RandomSleep(); + } + + void OnCompactionBegin(DB* /*db*/, const CompactionJobInfo& /*ci*/) override { + RandomSleep(); + } + + void OnCompactionCompleted(DB* /*db*/, const CompactionJobInfo& ci) override { + assert(IsValidColumnFamilyName(ci.cf_name)); + assert(ci.input_files.size() + ci.output_files.size() > 0U); + for (const auto& file_path : ci.input_files) { + VerifyFilePath(file_path); + } + for (const auto& file_path : ci.output_files) { + VerifyFilePath(file_path); + } + // pretending doing some work here + RandomSleep(); + } + + void OnTableFileCreationStarted( + const TableFileCreationBriefInfo& /*info*/) override { + ++num_pending_file_creations_; + } + + void OnTableFileCreated(const TableFileCreationInfo& info) override { + assert(info.db_name == db_name_); + assert(IsValidColumnFamilyName(info.cf_name)); + assert(info.job_id > 0 || FLAGS_compact_files_one_in > 0); + if (info.status.ok()) { + assert(info.file_size > 0); + VerifyFilePath(info.file_path); + assert(info.table_properties.data_size > 0 || + info.table_properties.num_range_deletions > 0); + assert(info.table_properties.raw_key_size > 0); + assert(info.table_properties.num_entries > 0); + VerifyTableFileUniqueId(info.table_properties, info.file_path); + } + --num_pending_file_creations_; + } + + void OnMemTableSealed(const MemTableInfo& /*info*/) override { + RandomSleep(); + } + + void OnColumnFamilyHandleDeletionStarted( + ColumnFamilyHandle* /*handle*/) override { + RandomSleep(); + } + + void OnExternalFileIngested(DB* /*db*/, + const ExternalFileIngestionInfo& info) override { + RandomSleep(); + // Here we assume that each generated external file is ingested + // exactly once (or thrown away in case of crash) + VerifyTableFileUniqueId(info.table_properties, info.internal_file_path); + } + + void OnBackgroundError(BackgroundErrorReason /* reason */, + Status* /* bg_error */) override { + RandomSleep(); + } + + void OnStallConditionsChanged(const WriteStallInfo& /*info*/) override { + RandomSleep(); + } + + void OnFileReadFinish(const FileOperationInfo& info) override { + // Even empty callback is valuable because sometimes some locks are + // released in order to make the callback. + + // Sleep carefully here as it is a frequent operation and we don't want + // to slow down the tests. We always sleep when the read is large. + // When read is small, sleep in a small chance. + size_t length_read = info.length; + if (length_read >= 1000000 || Random::GetTLSInstance()->OneIn(1000)) { + RandomSleep(); + } + } + + void OnFileWriteFinish(const FileOperationInfo& info) override { + // Even empty callback is valuable because sometimes some locks are + // released in order to make the callback. + + // Sleep carefully here as it is a frequent operation and we don't want + // to slow down the tests. When the write is large, always sleep. + // Otherwise, sleep in a relatively small chance. + size_t length_write = info.length; + if (length_write >= 1000000 || Random::GetTLSInstance()->OneIn(64)) { + RandomSleep(); + } + } + + bool ShouldBeNotifiedOnFileIO() override { + RandomSleep(); + return static_cast<bool>(Random::GetTLSInstance()->OneIn(1)); + } + + void OnErrorRecoveryBegin(BackgroundErrorReason /* reason */, + Status /* bg_error */, + bool* /* auto_recovery */) override { + RandomSleep(); + } + + void OnErrorRecoveryCompleted(Status /* old_bg_error */) override { + RandomSleep(); + } + + protected: + bool IsValidColumnFamilyName(const std::string& cf_name) const { + if (cf_name == kDefaultColumnFamilyName) { + return true; + } + // The column family names in the stress tests are numbers. + for (size_t i = 0; i < cf_name.size(); ++i) { + if (cf_name[i] < '0' || cf_name[i] > '9') { + return false; + } + } + return true; + } + + void VerifyFileDir(const std::string& file_dir) { +#ifndef NDEBUG + if (db_name_ == file_dir) { + return; + } + for (const auto& db_path : db_paths_) { + if (db_path.path == file_dir) { + return; + } + } + for (auto& cf : column_families_) { + for (const auto& cf_path : cf.options.cf_paths) { + if (cf_path.path == file_dir) { + return; + } + } + } + assert(false); +#else + (void)file_dir; +#endif // !NDEBUG + } + + void VerifyFileName(const std::string& file_name) { +#ifndef NDEBUG + uint64_t file_number; + FileType file_type; + bool result = ParseFileName(file_name, &file_number, &file_type); + assert(result); + assert(file_type == kTableFile); +#else + (void)file_name; +#endif // !NDEBUG + } + + void VerifyFilePath(const std::string& file_path) { +#ifndef NDEBUG + size_t pos = file_path.find_last_of("/"); + if (pos == std::string::npos) { + VerifyFileName(file_path); + } else { + if (pos > 0) { + VerifyFileDir(file_path.substr(0, pos)); + } + VerifyFileName(file_path.substr(pos)); + } +#else + (void)file_path; +#endif // !NDEBUG + } + + // Unique id is verified using the TableProperties. file_path is only used + // for reporting. + void VerifyTableFileUniqueId(const TableProperties& new_file_properties, + const std::string& file_path); + + void RandomSleep() { + std::this_thread::sleep_for( + std::chrono::microseconds(Random::GetTLSInstance()->Uniform(5000))); + } + + private: + std::string db_name_; + std::vector<DbPath> db_paths_; + std::vector<ColumnFamilyDescriptor> column_families_; + std::atomic<int> num_pending_file_creations_; + UniqueIdVerifier unique_ids_; +}; +#endif // !ROCKSDB_LITE +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_shared_state.cc b/src/rocksdb/db_stress_tool/db_stress_shared_state.cc new file mode 100644 index 000000000..a27f6ac73 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_shared_state.cc @@ -0,0 +1,17 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. +// + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_shared_state.h" + +namespace ROCKSDB_NAMESPACE { +thread_local bool SharedState::ignore_read_error; +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_shared_state.h b/src/rocksdb/db_stress_tool/db_stress_shared_state.h new file mode 100644 index 000000000..5565c6221 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_shared_state.h @@ -0,0 +1,427 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors + +#ifdef GFLAGS +#pragma once + +#include "db_stress_tool/db_stress_stat.h" +#include "db_stress_tool/expected_state.h" +// SyncPoint is not supported in Released Windows Mode. +#if !(defined NDEBUG) || !defined(OS_WIN) +#include "test_util/sync_point.h" +#endif // !(defined NDEBUG) || !defined(OS_WIN) +#include "util/gflags_compat.h" + +DECLARE_uint64(seed); +DECLARE_int64(max_key); +DECLARE_uint64(log2_keys_per_lock); +DECLARE_int32(threads); +DECLARE_int32(column_families); +DECLARE_int32(nooverwritepercent); +DECLARE_string(expected_values_dir); +DECLARE_int32(clear_column_family_one_in); +DECLARE_bool(test_batches_snapshots); +DECLARE_int32(compaction_thread_pool_adjust_interval); +DECLARE_int32(continuous_verification_interval); +DECLARE_int32(read_fault_one_in); +DECLARE_int32(write_fault_one_in); +DECLARE_int32(open_metadata_write_fault_one_in); +DECLARE_int32(open_write_fault_one_in); +DECLARE_int32(open_read_fault_one_in); + +DECLARE_int32(injest_error_severity); + +namespace ROCKSDB_NAMESPACE { +class StressTest; + +// State shared by all concurrent executions of the same benchmark. +class SharedState { + public: + // indicates a key may have any value (or not be present) as an operation on + // it is incomplete. + static constexpr uint32_t UNKNOWN_SENTINEL = 0xfffffffe; + // indicates a key should definitely be deleted + static constexpr uint32_t DELETION_SENTINEL = 0xffffffff; + + // Errors when reading filter blocks are ignored, so we use a thread + // local variable updated via sync points to keep track of errors injected + // while reading filter blocks in order to ignore the Get/MultiGet result + // for those calls + static thread_local bool ignore_read_error; + + SharedState(Env* /*env*/, StressTest* stress_test) + : cv_(&mu_), + seed_(static_cast<uint32_t>(FLAGS_seed)), + max_key_(FLAGS_max_key), + log2_keys_per_lock_(static_cast<uint32_t>(FLAGS_log2_keys_per_lock)), + num_threads_(0), + num_initialized_(0), + num_populated_(0), + vote_reopen_(0), + num_done_(0), + start_(false), + start_verify_(false), + num_bg_threads_(0), + should_stop_bg_thread_(false), + bg_thread_finished_(0), + stress_test_(stress_test), + verification_failure_(false), + should_stop_test_(false), + no_overwrite_ids_(GenerateNoOverwriteIds()), + expected_state_manager_(nullptr), + printing_verification_results_(false), + start_timestamp_(Env::Default()->NowNanos()) { + Status status; + // TODO: We should introduce a way to explicitly disable verification + // during shutdown. When that is disabled and FLAGS_expected_values_dir + // is empty (disabling verification at startup), we can skip tracking + // expected state. Only then should we permit bypassing the below feature + // compatibility checks. + if (!FLAGS_expected_values_dir.empty()) { + if (!std::atomic<uint32_t>{}.is_lock_free()) { + status = Status::InvalidArgument( + "Cannot use --expected_values_dir on platforms without lock-free " + "std::atomic<uint32_t>"); + } + if (status.ok() && FLAGS_clear_column_family_one_in > 0) { + status = Status::InvalidArgument( + "Cannot use --expected_values_dir on when " + "--clear_column_family_one_in is greater than zero."); + } + } + if (status.ok()) { + if (FLAGS_expected_values_dir.empty()) { + expected_state_manager_.reset( + new AnonExpectedStateManager(FLAGS_max_key, FLAGS_column_families)); + } else { + expected_state_manager_.reset(new FileExpectedStateManager( + FLAGS_max_key, FLAGS_column_families, FLAGS_expected_values_dir)); + } + status = expected_state_manager_->Open(); + } + if (!status.ok()) { + fprintf(stderr, "Failed setting up expected state with error: %s\n", + status.ToString().c_str()); + exit(1); + } + + if (FLAGS_test_batches_snapshots) { + fprintf(stdout, "No lock creation because test_batches_snapshots set\n"); + return; + } + + long num_locks = static_cast<long>(max_key_ >> log2_keys_per_lock_); + if (max_key_ & ((1 << log2_keys_per_lock_) - 1)) { + num_locks++; + } + fprintf(stdout, "Creating %ld locks\n", num_locks * FLAGS_column_families); + key_locks_.resize(FLAGS_column_families); + + for (int i = 0; i < FLAGS_column_families; ++i) { + key_locks_[i].reset(new port::Mutex[num_locks]); + } + if (FLAGS_read_fault_one_in) { +#ifdef NDEBUG + // Unsupported in release mode because it relies on + // `IGNORE_STATUS_IF_ERROR` to distinguish faults not expected to lead to + // failure. + fprintf(stderr, + "Cannot set nonzero value for --read_fault_one_in in " + "release mode."); + exit(1); +#else // NDEBUG + SyncPoint::GetInstance()->SetCallBack("FaultInjectionIgnoreError", + IgnoreReadErrorCallback); + SyncPoint::GetInstance()->EnableProcessing(); +#endif // NDEBUG + } + } + + ~SharedState() { +#ifndef NDEBUG + if (FLAGS_read_fault_one_in) { + SyncPoint::GetInstance()->ClearAllCallBacks(); + SyncPoint::GetInstance()->DisableProcessing(); + } +#endif + } + + port::Mutex* GetMutex() { return &mu_; } + + port::CondVar* GetCondVar() { return &cv_; } + + StressTest* GetStressTest() const { return stress_test_; } + + int64_t GetMaxKey() const { return max_key_; } + + uint32_t GetNumThreads() const { return num_threads_; } + + void SetThreads(int num_threads) { num_threads_ = num_threads; } + + void IncInitialized() { num_initialized_++; } + + void IncOperated() { num_populated_++; } + + void IncDone() { num_done_++; } + + void IncVotedReopen() { vote_reopen_ = (vote_reopen_ + 1) % num_threads_; } + + bool AllInitialized() const { return num_initialized_ >= num_threads_; } + + bool AllOperated() const { return num_populated_ >= num_threads_; } + + bool AllDone() const { return num_done_ >= num_threads_; } + + bool AllVotedReopen() { return (vote_reopen_ == 0); } + + void SetStart() { start_ = true; } + + void SetStartVerify() { start_verify_ = true; } + + bool Started() const { return start_; } + + bool VerifyStarted() const { return start_verify_; } + + void SetVerificationFailure() { verification_failure_.store(true); } + + bool HasVerificationFailedYet() const { return verification_failure_.load(); } + + void SetShouldStopTest() { should_stop_test_.store(true); } + + bool ShouldStopTest() const { return should_stop_test_.load(); } + + // Returns a lock covering `key` in `cf`. + port::Mutex* GetMutexForKey(int cf, int64_t key) { + return &key_locks_[cf][key >> log2_keys_per_lock_]; + } + + // Acquires locks for all keys in `cf`. + void LockColumnFamily(int cf) { + for (int i = 0; i < max_key_ >> log2_keys_per_lock_; ++i) { + key_locks_[cf][i].Lock(); + } + } + + // Releases locks for all keys in `cf`. + void UnlockColumnFamily(int cf) { + for (int i = 0; i < max_key_ >> log2_keys_per_lock_; ++i) { + key_locks_[cf][i].Unlock(); + } + } + + // Returns a collection of mutex locks covering the key range [start, end) in + // `cf`. + std::vector<std::unique_ptr<MutexLock>> GetLocksForKeyRange(int cf, + int64_t start, + int64_t end) { + std::vector<std::unique_ptr<MutexLock>> range_locks; + + if (start >= end) { + return range_locks; + } + + const int64_t start_idx = start >> log2_keys_per_lock_; + + int64_t end_idx = end >> log2_keys_per_lock_; + if ((end & ((1 << log2_keys_per_lock_) - 1)) == 0) { + --end_idx; + } + + for (int64_t idx = start_idx; idx <= end_idx; ++idx) { + range_locks.emplace_back( + std::make_unique<MutexLock>(&key_locks_[cf][idx])); + } + + return range_locks; + } + + Status SaveAtAndAfter(DB* db) { + return expected_state_manager_->SaveAtAndAfter(db); + } + + bool HasHistory() { return expected_state_manager_->HasHistory(); } + + Status Restore(DB* db) { return expected_state_manager_->Restore(db); } + + // Requires external locking covering all keys in `cf`. + void ClearColumnFamily(int cf) { + return expected_state_manager_->ClearColumnFamily(cf); + } + + // @param pending True if the update may have started but is not yet + // guaranteed finished. This is useful for crash-recovery testing when the + // process may crash before updating the expected values array. + // + // Requires external locking covering `key` in `cf`. + void Put(int cf, int64_t key, uint32_t value_base, bool pending) { + return expected_state_manager_->Put(cf, key, value_base, pending); + } + + // Requires external locking covering `key` in `cf`. + uint32_t Get(int cf, int64_t key) const { + return expected_state_manager_->Get(cf, key); + } + + // @param pending See comment above Put() + // Returns true if the key was not yet deleted. + // + // Requires external locking covering `key` in `cf`. + bool Delete(int cf, int64_t key, bool pending) { + return expected_state_manager_->Delete(cf, key, pending); + } + + // @param pending See comment above Put() + // Returns true if the key was not yet deleted. + // + // Requires external locking covering `key` in `cf`. + bool SingleDelete(int cf, int64_t key, bool pending) { + return expected_state_manager_->Delete(cf, key, pending); + } + + // @param pending See comment above Put() + // Returns number of keys deleted by the call. + // + // Requires external locking covering keys in `[begin_key, end_key)` in `cf`. + int DeleteRange(int cf, int64_t begin_key, int64_t end_key, bool pending) { + return expected_state_manager_->DeleteRange(cf, begin_key, end_key, + pending); + } + + bool AllowsOverwrite(int64_t key) const { + return no_overwrite_ids_.find(key) == no_overwrite_ids_.end(); + } + + // Requires external locking covering `key` in `cf`. + bool Exists(int cf, int64_t key) { + return expected_state_manager_->Exists(cf, key); + } + + uint32_t GetSeed() const { return seed_; } + + void SetShouldStopBgThread() { should_stop_bg_thread_ = true; } + + bool ShouldStopBgThread() { return should_stop_bg_thread_; } + + void IncBgThreads() { ++num_bg_threads_; } + + void IncBgThreadsFinished() { ++bg_thread_finished_; } + + bool BgThreadsFinished() const { + return bg_thread_finished_ == num_bg_threads_; + } + + bool ShouldVerifyAtBeginning() const { + return !FLAGS_expected_values_dir.empty(); + } + + bool PrintingVerificationResults() { + bool tmp = false; + return !printing_verification_results_.compare_exchange_strong( + tmp, true, std::memory_order_relaxed); + } + + void FinishPrintingVerificationResults() { + printing_verification_results_.store(false, std::memory_order_relaxed); + } + + uint64_t GetStartTimestamp() const { return start_timestamp_; } + + private: + static void IgnoreReadErrorCallback(void*) { ignore_read_error = true; } + + // Pick random keys in each column family that will not experience overwrite. + std::unordered_set<int64_t> GenerateNoOverwriteIds() const { + fprintf(stdout, "Choosing random keys with no overwrite\n"); + // Start with the identity permutation. Subsequent iterations of + // for loop below will start with perm of previous for loop + std::vector<int64_t> permutation(max_key_); + for (int64_t i = 0; i < max_key_; ++i) { + permutation[i] = i; + } + // Now do the Knuth shuffle + const int64_t num_no_overwrite_keys = + (max_key_ * FLAGS_nooverwritepercent) / 100; + // Only need to figure out first num_no_overwrite_keys of permutation + std::unordered_set<int64_t> ret; + ret.reserve(num_no_overwrite_keys); + Random64 rnd(seed_); + for (int64_t i = 0; i < num_no_overwrite_keys; i++) { + assert(i < max_key_); + int64_t rand_index = i + rnd.Next() % (max_key_ - i); + // Swap i and rand_index; + int64_t temp = permutation[i]; + permutation[i] = permutation[rand_index]; + permutation[rand_index] = temp; + // Fill no_overwrite_ids_ with the first num_no_overwrite_keys of + // permutation + ret.insert(permutation[i]); + } + return ret; + } + + port::Mutex mu_; + port::CondVar cv_; + const uint32_t seed_; + const int64_t max_key_; + const uint32_t log2_keys_per_lock_; + int num_threads_; + long num_initialized_; + long num_populated_; + long vote_reopen_; + long num_done_; + bool start_; + bool start_verify_; + int num_bg_threads_; + bool should_stop_bg_thread_; + int bg_thread_finished_; + StressTest* stress_test_; + std::atomic<bool> verification_failure_; + std::atomic<bool> should_stop_test_; + + // Keys that should not be overwritten + const std::unordered_set<int64_t> no_overwrite_ids_; + + std::unique_ptr<ExpectedStateManager> expected_state_manager_; + // Cannot store `port::Mutex` directly in vector since it is not copyable + // and storing it in the container may require copying depending on the impl. + std::vector<std::unique_ptr<port::Mutex[]>> key_locks_; + std::atomic<bool> printing_verification_results_; + const uint64_t start_timestamp_; +}; + +// Per-thread state for concurrent executions of the same benchmark. +struct ThreadState { + uint32_t tid; // 0..n-1 + Random rand; // Has different seeds for different threads + SharedState* shared; + Stats stats; + struct SnapshotState { + const Snapshot* snapshot; + // The cf from which we did a Get at this snapshot + int cf_at; + // The name of the cf at the time that we did a read + std::string cf_at_name; + // The key with which we did a Get at this snapshot + std::string key; + // The status of the Get + Status status; + // The value of the Get + std::string value; + // optional state of all keys in the db + std::vector<bool>* key_vec; + + std::string timestamp; + }; + std::queue<std::pair<uint64_t, SnapshotState>> snapshot_queue; + + ThreadState(uint32_t index, SharedState* _shared) + : tid(index), rand(1000 + index + _shared->GetSeed()), shared(_shared) {} +}; +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_stat.cc b/src/rocksdb/db_stress_tool/db_stress_stat.cc new file mode 100644 index 000000000..6a7883a52 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_stat.cc @@ -0,0 +1,17 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). + +#ifdef GFLAGS + +#include "db_stress_tool/db_stress_stat.h" + +namespace ROCKSDB_NAMESPACE { + +std::shared_ptr<ROCKSDB_NAMESPACE::Statistics> dbstats; +std::shared_ptr<ROCKSDB_NAMESPACE::Statistics> dbstats_secondaries; + +} // namespace ROCKSDB_NAMESPACE + +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_stat.h b/src/rocksdb/db_stress_tool/db_stress_stat.h new file mode 100644 index 000000000..5b38c6e2b --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_stat.h @@ -0,0 +1,219 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). + +#pragma once +#include <cinttypes> +#include <memory> +#include <queue> +#include <unordered_set> + +#include "monitoring/histogram.h" +#include "port/port.h" +#include "rocksdb/snapshot.h" +#include "rocksdb/statistics.h" +#include "rocksdb/system_clock.h" +#include "util/gflags_compat.h" +#include "util/random.h" + +DECLARE_bool(histogram); +DECLARE_bool(progress_reports); + +namespace ROCKSDB_NAMESPACE { + +// Database statistics +extern std::shared_ptr<ROCKSDB_NAMESPACE::Statistics> dbstats; +extern std::shared_ptr<ROCKSDB_NAMESPACE::Statistics> dbstats_secondaries; + +class Stats { + private: + uint64_t start_; + uint64_t finish_; + double seconds_; + long done_; + long gets_; + long prefixes_; + long writes_; + long deletes_; + size_t single_deletes_; + long iterator_size_sums_; + long founds_; + long iterations_; + long range_deletions_; + long covered_by_range_deletions_; + long errors_; + long verified_errors_; + long num_compact_files_succeed_; + long num_compact_files_failed_; + int next_report_; + size_t bytes_; + uint64_t last_op_finish_; + HistogramImpl hist_; + + public: + Stats() {} + + void Start() { + next_report_ = 100; + hist_.Clear(); + done_ = 0; + gets_ = 0; + prefixes_ = 0; + writes_ = 0; + deletes_ = 0; + single_deletes_ = 0; + iterator_size_sums_ = 0; + founds_ = 0; + iterations_ = 0; + range_deletions_ = 0; + covered_by_range_deletions_ = 0; + errors_ = 0; + verified_errors_ = 0; + bytes_ = 0; + seconds_ = 0; + num_compact_files_succeed_ = 0; + num_compact_files_failed_ = 0; + start_ = SystemClock::Default()->NowMicros(); + last_op_finish_ = start_; + finish_ = start_; + } + + void Merge(const Stats& other) { + hist_.Merge(other.hist_); + done_ += other.done_; + gets_ += other.gets_; + prefixes_ += other.prefixes_; + writes_ += other.writes_; + deletes_ += other.deletes_; + single_deletes_ += other.single_deletes_; + iterator_size_sums_ += other.iterator_size_sums_; + founds_ += other.founds_; + iterations_ += other.iterations_; + range_deletions_ += other.range_deletions_; + covered_by_range_deletions_ = other.covered_by_range_deletions_; + errors_ += other.errors_; + verified_errors_ += other.verified_errors_; + bytes_ += other.bytes_; + seconds_ += other.seconds_; + num_compact_files_succeed_ += other.num_compact_files_succeed_; + num_compact_files_failed_ += other.num_compact_files_failed_; + if (other.start_ < start_) start_ = other.start_; + if (other.finish_ > finish_) finish_ = other.finish_; + } + + void Stop() { + finish_ = SystemClock::Default()->NowMicros(); + seconds_ = (finish_ - start_) * 1e-6; + } + + void FinishedSingleOp() { + if (FLAGS_histogram) { + auto now = SystemClock::Default()->NowMicros(); + auto micros = now - last_op_finish_; + hist_.Add(micros); + if (micros > 20000) { + fprintf(stdout, "long op: %" PRIu64 " micros%30s\r", micros, ""); + } + last_op_finish_ = now; + } + + done_++; + if (FLAGS_progress_reports) { + if (done_ >= next_report_) { + if (next_report_ < 1000) + next_report_ += 100; + else if (next_report_ < 5000) + next_report_ += 500; + else if (next_report_ < 10000) + next_report_ += 1000; + else if (next_report_ < 50000) + next_report_ += 5000; + else if (next_report_ < 100000) + next_report_ += 10000; + else if (next_report_ < 500000) + next_report_ += 50000; + else + next_report_ += 100000; + fprintf(stdout, "... finished %ld ops%30s\r", done_, ""); + } + } + } + + void AddBytesForWrites(long nwrites, size_t nbytes) { + writes_ += nwrites; + bytes_ += nbytes; + } + + void AddGets(long ngets, long nfounds) { + founds_ += nfounds; + gets_ += ngets; + } + + void AddPrefixes(long nprefixes, long count) { + prefixes_ += nprefixes; + iterator_size_sums_ += count; + } + + void AddIterations(long n) { iterations_ += n; } + + void AddDeletes(long n) { deletes_ += n; } + + void AddSingleDeletes(size_t n) { single_deletes_ += n; } + + void AddRangeDeletions(long n) { range_deletions_ += n; } + + void AddCoveredByRangeDeletions(long n) { covered_by_range_deletions_ += n; } + + void AddErrors(long n) { errors_ += n; } + + void AddVerifiedErrors(long n) { verified_errors_ += n; } + + void AddNumCompactFilesSucceed(long n) { num_compact_files_succeed_ += n; } + + void AddNumCompactFilesFailed(long n) { num_compact_files_failed_ += n; } + + void Report(const char* name) { + std::string extra; + if (bytes_ < 1 || done_ < 1) { + fprintf(stderr, "No writes or ops?\n"); + return; + } + + double elapsed = (finish_ - start_) * 1e-6; + double bytes_mb = bytes_ / 1048576.0; + double rate = bytes_mb / elapsed; + double throughput = (double)done_ / elapsed; + + fprintf(stdout, "%-12s: ", name); + fprintf(stdout, "%.3f micros/op %ld ops/sec\n", seconds_ * 1e6 / done_, + (long)throughput); + fprintf(stdout, "%-12s: Wrote %.2f MB (%.2f MB/sec) (%ld%% of %ld ops)\n", + "", bytes_mb, rate, (100 * writes_) / done_, done_); + fprintf(stdout, "%-12s: Wrote %ld times\n", "", writes_); + fprintf(stdout, "%-12s: Deleted %ld times\n", "", deletes_); + fprintf(stdout, "%-12s: Single deleted %" ROCKSDB_PRIszt " times\n", "", + single_deletes_); + fprintf(stdout, "%-12s: %ld read and %ld found the key\n", "", gets_, + founds_); + fprintf(stdout, "%-12s: Prefix scanned %ld times\n", "", prefixes_); + fprintf(stdout, "%-12s: Iterator size sum is %ld\n", "", + iterator_size_sums_); + fprintf(stdout, "%-12s: Iterated %ld times\n", "", iterations_); + fprintf(stdout, "%-12s: Deleted %ld key-ranges\n", "", range_deletions_); + fprintf(stdout, "%-12s: Range deletions covered %ld keys\n", "", + covered_by_range_deletions_); + + fprintf(stdout, "%-12s: Got errors %ld times\n", "", errors_); + fprintf(stdout, "%-12s: %ld CompactFiles() succeed\n", "", + num_compact_files_succeed_); + fprintf(stdout, "%-12s: %ld CompactFiles() did not succeed\n", "", + num_compact_files_failed_); + + if (FLAGS_histogram) { + fprintf(stdout, "Microseconds per op:\n%s\n", hist_.ToString().c_str()); + } + fflush(stdout); + } +}; +} // namespace ROCKSDB_NAMESPACE diff --git a/src/rocksdb/db_stress_tool/db_stress_table_properties_collector.h b/src/rocksdb/db_stress_tool/db_stress_table_properties_collector.h new file mode 100644 index 000000000..d1758cbb4 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_table_properties_collector.h @@ -0,0 +1,65 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). + +#pragma once + +#include "rocksdb/table.h" +#include "util/gflags_compat.h" +#include "util/random.h" + +DECLARE_int32(mark_for_compaction_one_file_in); + +namespace ROCKSDB_NAMESPACE { + +// A `DbStressTablePropertiesCollector` ignores what keys/values were added to +// the table, adds no properties to the table, and decides at random whether the +// table will be marked for compaction according to +// `FLAGS_mark_for_compaction_one_file_in`. +class DbStressTablePropertiesCollector : public TablePropertiesCollector { + public: + DbStressTablePropertiesCollector() + : need_compact_(Random::GetTLSInstance()->OneInOpt( + FLAGS_mark_for_compaction_one_file_in)) {} + + virtual Status AddUserKey(const Slice& /* key */, const Slice& /* value */, + EntryType /*type*/, SequenceNumber /*seq*/, + uint64_t /*file_size*/) override { + return Status::OK(); + } + + virtual Status Finish(UserCollectedProperties* /* properties */) override { + return Status::OK(); + } + + virtual UserCollectedProperties GetReadableProperties() const override { + return UserCollectedProperties{}; + } + + virtual const char* Name() const override { + return "DbStressTablePropertiesCollector"; + } + + virtual bool NeedCompact() const override { return need_compact_; } + + private: + const bool need_compact_; +}; + +// A `DbStressTablePropertiesCollectorFactory` creates +// `DbStressTablePropertiesCollectorFactory`s. +class DbStressTablePropertiesCollectorFactory + : public TablePropertiesCollectorFactory { + public: + virtual TablePropertiesCollector* CreateTablePropertiesCollector( + TablePropertiesCollectorFactory::Context /* context */) override { + return new DbStressTablePropertiesCollector(); + } + + virtual const char* Name() const override { + return "DbStressTablePropertiesCollectorFactory"; + } +}; + +} // namespace ROCKSDB_NAMESPACE diff --git a/src/rocksdb/db_stress_tool/db_stress_test_base.cc b/src/rocksdb/db_stress_tool/db_stress_test_base.cc new file mode 100644 index 000000000..e51b43176 --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_test_base.cc @@ -0,0 +1,3383 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. +// + +#include <ios> + +#include "util/compression.h" +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" +#include "db_stress_tool/db_stress_compaction_filter.h" +#include "db_stress_tool/db_stress_driver.h" +#include "db_stress_tool/db_stress_table_properties_collector.h" +#include "rocksdb/convenience.h" +#include "rocksdb/filter_policy.h" +#include "rocksdb/secondary_cache.h" +#include "rocksdb/sst_file_manager.h" +#include "rocksdb/types.h" +#include "rocksdb/utilities/object_registry.h" +#include "rocksdb/utilities/write_batch_with_index.h" +#include "test_util/testutil.h" +#include "util/cast_util.h" +#include "utilities/backup/backup_engine_impl.h" +#include "utilities/fault_injection_fs.h" +#include "utilities/fault_injection_secondary_cache.h" + +namespace ROCKSDB_NAMESPACE { + +namespace { + +std::shared_ptr<const FilterPolicy> CreateFilterPolicy() { + if (FLAGS_bloom_bits < 0) { + return BlockBasedTableOptions().filter_policy; + } + const FilterPolicy* new_policy; + if (FLAGS_ribbon_starting_level >= 999) { + // Use Bloom API + new_policy = NewBloomFilterPolicy(FLAGS_bloom_bits, false); + } else { + new_policy = NewRibbonFilterPolicy( + FLAGS_bloom_bits, /* bloom_before_level */ FLAGS_ribbon_starting_level); + } + return std::shared_ptr<const FilterPolicy>(new_policy); +} + +} // namespace + +StressTest::StressTest() + : cache_(NewCache(FLAGS_cache_size, FLAGS_cache_numshardbits)), + compressed_cache_(NewLRUCache(FLAGS_compressed_cache_size, + FLAGS_compressed_cache_numshardbits)), + filter_policy_(CreateFilterPolicy()), + db_(nullptr), +#ifndef ROCKSDB_LITE + txn_db_(nullptr), +#endif + db_aptr_(nullptr), + clock_(db_stress_env->GetSystemClock().get()), + new_column_family_name_(1), + num_times_reopened_(0), + db_preload_finished_(false), + cmp_db_(nullptr), + is_db_stopped_(false) { + if (FLAGS_destroy_db_initially) { + std::vector<std::string> files; + db_stress_env->GetChildren(FLAGS_db, &files); + for (unsigned int i = 0; i < files.size(); i++) { + if (Slice(files[i]).starts_with("heap-")) { + db_stress_env->DeleteFile(FLAGS_db + "/" + files[i]); + } + } + + Options options; + options.env = db_stress_env; + // Remove files without preserving manfiest files +#ifndef ROCKSDB_LITE + const Status s = !FLAGS_use_blob_db + ? DestroyDB(FLAGS_db, options) + : blob_db::DestroyBlobDB(FLAGS_db, options, + blob_db::BlobDBOptions()); +#else + const Status s = DestroyDB(FLAGS_db, options); +#endif // !ROCKSDB_LITE + + if (!s.ok()) { + fprintf(stderr, "Cannot destroy original db: %s\n", s.ToString().c_str()); + exit(1); + } + } +} + +StressTest::~StressTest() { + for (auto cf : column_families_) { + delete cf; + } + column_families_.clear(); + delete db_; + + for (auto* cf : cmp_cfhs_) { + delete cf; + } + cmp_cfhs_.clear(); + delete cmp_db_; +} + +std::shared_ptr<Cache> StressTest::NewCache(size_t capacity, + int32_t num_shard_bits) { + ConfigOptions config_options; + if (capacity <= 0) { + return nullptr; + } + + if (FLAGS_cache_type == "clock_cache") { + fprintf(stderr, "Old clock cache implementation has been removed.\n"); + exit(1); + } else if (FLAGS_cache_type == "hyper_clock_cache") { + return HyperClockCacheOptions(static_cast<size_t>(capacity), + FLAGS_block_size /*estimated_entry_charge*/, + num_shard_bits) + .MakeSharedCache(); + } else if (FLAGS_cache_type == "lru_cache") { + LRUCacheOptions opts; + opts.capacity = capacity; + opts.num_shard_bits = num_shard_bits; +#ifndef ROCKSDB_LITE + std::shared_ptr<SecondaryCache> secondary_cache; + if (!FLAGS_secondary_cache_uri.empty()) { + Status s = SecondaryCache::CreateFromString( + config_options, FLAGS_secondary_cache_uri, &secondary_cache); + if (secondary_cache == nullptr) { + fprintf(stderr, + "No secondary cache registered matching string: %s status=%s\n", + FLAGS_secondary_cache_uri.c_str(), s.ToString().c_str()); + exit(1); + } + if (FLAGS_secondary_cache_fault_one_in > 0) { + secondary_cache = std::make_shared<FaultInjectionSecondaryCache>( + secondary_cache, static_cast<uint32_t>(FLAGS_seed), + FLAGS_secondary_cache_fault_one_in); + } + opts.secondary_cache = secondary_cache; + } +#endif + return NewLRUCache(opts); + } else { + fprintf(stderr, "Cache type not supported."); + exit(1); + } +} + +std::vector<std::string> StressTest::GetBlobCompressionTags() { + std::vector<std::string> compression_tags{"kNoCompression"}; + + if (Snappy_Supported()) { + compression_tags.emplace_back("kSnappyCompression"); + } + if (LZ4_Supported()) { + compression_tags.emplace_back("kLZ4Compression"); + } + if (ZSTD_Supported()) { + compression_tags.emplace_back("kZSTD"); + } + + return compression_tags; +} + +bool StressTest::BuildOptionsTable() { + if (FLAGS_set_options_one_in <= 0) { + return true; + } + + std::unordered_map<std::string, std::vector<std::string>> options_tbl = { + {"write_buffer_size", + {std::to_string(options_.write_buffer_size), + std::to_string(options_.write_buffer_size * 2), + std::to_string(options_.write_buffer_size * 4)}}, + {"max_write_buffer_number", + {std::to_string(options_.max_write_buffer_number), + std::to_string(options_.max_write_buffer_number * 2), + std::to_string(options_.max_write_buffer_number * 4)}}, + {"arena_block_size", + { + std::to_string(options_.arena_block_size), + std::to_string(options_.write_buffer_size / 4), + std::to_string(options_.write_buffer_size / 8), + }}, + {"memtable_huge_page_size", {"0", std::to_string(2 * 1024 * 1024)}}, + {"max_successive_merges", {"0", "2", "4"}}, + {"inplace_update_num_locks", {"100", "200", "300"}}, + // TODO: re-enable once internal task T124324915 is fixed. + // {"experimental_mempurge_threshold", {"0.0", "1.0"}}, + // TODO(ljin): enable test for this option + // {"disable_auto_compactions", {"100", "200", "300"}}, + {"level0_file_num_compaction_trigger", + { + std::to_string(options_.level0_file_num_compaction_trigger), + std::to_string(options_.level0_file_num_compaction_trigger + 2), + std::to_string(options_.level0_file_num_compaction_trigger + 4), + }}, + {"level0_slowdown_writes_trigger", + { + std::to_string(options_.level0_slowdown_writes_trigger), + std::to_string(options_.level0_slowdown_writes_trigger + 2), + std::to_string(options_.level0_slowdown_writes_trigger + 4), + }}, + {"level0_stop_writes_trigger", + { + std::to_string(options_.level0_stop_writes_trigger), + std::to_string(options_.level0_stop_writes_trigger + 2), + std::to_string(options_.level0_stop_writes_trigger + 4), + }}, + {"max_compaction_bytes", + { + std::to_string(options_.target_file_size_base * 5), + std::to_string(options_.target_file_size_base * 15), + std::to_string(options_.target_file_size_base * 100), + }}, + {"target_file_size_base", + { + std::to_string(options_.target_file_size_base), + std::to_string(options_.target_file_size_base * 2), + std::to_string(options_.target_file_size_base * 4), + }}, + {"target_file_size_multiplier", + { + std::to_string(options_.target_file_size_multiplier), + "1", + "2", + }}, + {"max_bytes_for_level_base", + { + std::to_string(options_.max_bytes_for_level_base / 2), + std::to_string(options_.max_bytes_for_level_base), + std::to_string(options_.max_bytes_for_level_base * 2), + }}, + {"max_bytes_for_level_multiplier", + { + std::to_string(options_.max_bytes_for_level_multiplier), + "1", + "2", + }}, + {"max_sequential_skip_in_iterations", {"4", "8", "12"}}, + }; + + if (FLAGS_allow_setting_blob_options_dynamically) { + options_tbl.emplace("enable_blob_files", + std::vector<std::string>{"false", "true"}); + options_tbl.emplace("min_blob_size", + std::vector<std::string>{"0", "8", "16"}); + options_tbl.emplace("blob_file_size", + std::vector<std::string>{"1M", "16M", "256M", "1G"}); + options_tbl.emplace("blob_compression_type", GetBlobCompressionTags()); + options_tbl.emplace("enable_blob_garbage_collection", + std::vector<std::string>{"false", "true"}); + options_tbl.emplace( + "blob_garbage_collection_age_cutoff", + std::vector<std::string>{"0.0", "0.25", "0.5", "0.75", "1.0"}); + options_tbl.emplace("blob_garbage_collection_force_threshold", + std::vector<std::string>{"0.5", "0.75", "1.0"}); + options_tbl.emplace("blob_compaction_readahead_size", + std::vector<std::string>{"0", "1M", "4M"}); + options_tbl.emplace("blob_file_starting_level", + std::vector<std::string>{"0", "1", "2"}); + options_tbl.emplace("prepopulate_blob_cache", + std::vector<std::string>{"kDisable", "kFlushOnly"}); + } + + options_table_ = std::move(options_tbl); + + for (const auto& iter : options_table_) { + options_index_.push_back(iter.first); + } + return true; +} + +void StressTest::InitDb(SharedState* shared) { + uint64_t now = clock_->NowMicros(); + fprintf(stdout, "%s Initializing db_stress\n", + clock_->TimeToString(now / 1000000).c_str()); + PrintEnv(); + Open(shared); + BuildOptionsTable(); +} + +void StressTest::FinishInitDb(SharedState* shared) { + if (FLAGS_read_only) { + uint64_t now = clock_->NowMicros(); + fprintf(stdout, "%s Preloading db with %" PRIu64 " KVs\n", + clock_->TimeToString(now / 1000000).c_str(), FLAGS_max_key); + PreloadDbAndReopenAsReadOnly(FLAGS_max_key, shared); + } + + if (shared->HasHistory()) { + // The way it works right now is, if there's any history, that means the + // previous run mutating the DB had all its operations traced, in which case + // we should always be able to `Restore()` the expected values to match the + // `db_`'s current seqno. + Status s = shared->Restore(db_); + if (!s.ok()) { + fprintf(stderr, "Error restoring historical expected values: %s\n", + s.ToString().c_str()); + exit(1); + } + } +#ifndef ROCKSDB_LITE + if (FLAGS_use_txn) { + // It's OK here without sync because unsynced data cannot be lost at this + // point + // - even with sync_fault_injection=1 as the + // file is still directly writable until after FinishInitDb() + ProcessRecoveredPreparedTxns(shared); + } +#endif + if (FLAGS_enable_compaction_filter) { + auto* compaction_filter_factory = + reinterpret_cast<DbStressCompactionFilterFactory*>( + options_.compaction_filter_factory.get()); + assert(compaction_filter_factory); + // This must be called only after any potential `SharedState::Restore()` has + // completed in order for the `compaction_filter_factory` to operate on the + // correct latest values file. + compaction_filter_factory->SetSharedState(shared); + fprintf(stdout, "Compaction filter factory: %s\n", + compaction_filter_factory->Name()); + } +} + +void StressTest::TrackExpectedState(SharedState* shared) { + // For `FLAGS_manual_wal_flush_one_inWAL` + // data can be lost when `manual_wal_flush_one_in > 0` and `FlushWAL()` is not + // explictly called by users of RocksDB (in our case, db stress). + // Therefore recovery from such potential WAL data loss is a prefix recovery + // that requires tracing + if ((FLAGS_sync_fault_injection || FLAGS_disable_wal || + FLAGS_manual_wal_flush_one_in > 0) && + IsStateTracked()) { + Status s = shared->SaveAtAndAfter(db_); + if (!s.ok()) { + fprintf(stderr, "Error enabling history tracing: %s\n", + s.ToString().c_str()); + exit(1); + } + } +} + +Status StressTest::AssertSame(DB* db, ColumnFamilyHandle* cf, + ThreadState::SnapshotState& snap_state) { + Status s; + if (cf->GetName() != snap_state.cf_at_name) { + return s; + } + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions ropt; + ropt.snapshot = snap_state.snapshot; + Slice ts; + if (!snap_state.timestamp.empty()) { + ts = snap_state.timestamp; + ropt.timestamp = &ts; + } + PinnableSlice exp_v(&snap_state.value); + exp_v.PinSelf(); + PinnableSlice v; + s = db->Get(ropt, cf, snap_state.key, &v); + if (!s.ok() && !s.IsNotFound()) { + return s; + } + if (snap_state.status != s) { + return Status::Corruption( + "The snapshot gave inconsistent results for key " + + std::to_string(Hash(snap_state.key.c_str(), snap_state.key.size(), 0)) + + " in cf " + cf->GetName() + ": (" + snap_state.status.ToString() + + ") vs. (" + s.ToString() + ")"); + } + if (s.ok()) { + if (exp_v != v) { + return Status::Corruption("The snapshot gave inconsistent values: (" + + exp_v.ToString() + ") vs. (" + v.ToString() + + ")"); + } + } + if (snap_state.key_vec != nullptr) { + // When `prefix_extractor` is set, seeking to beginning and scanning + // across prefixes are only supported with `total_order_seek` set. + ropt.total_order_seek = true; + std::unique_ptr<Iterator> iterator(db->NewIterator(ropt)); + std::unique_ptr<std::vector<bool>> tmp_bitvec( + new std::vector<bool>(FLAGS_max_key)); + for (iterator->SeekToFirst(); iterator->Valid(); iterator->Next()) { + uint64_t key_val; + if (GetIntVal(iterator->key().ToString(), &key_val)) { + (*tmp_bitvec.get())[key_val] = true; + } + } + if (!std::equal(snap_state.key_vec->begin(), snap_state.key_vec->end(), + tmp_bitvec.get()->begin())) { + return Status::Corruption("Found inconsistent keys at this snapshot"); + } + } + return Status::OK(); +} + +void StressTest::VerificationAbort(SharedState* shared, std::string msg, + Status s) const { + fprintf(stderr, "Verification failed: %s. Status is %s\n", msg.c_str(), + s.ToString().c_str()); + shared->SetVerificationFailure(); +} + +void StressTest::VerificationAbort(SharedState* shared, std::string msg, int cf, + int64_t key) const { + auto key_str = Key(key); + Slice key_slice = key_str; + fprintf(stderr, + "Verification failed for column family %d key %s (%" PRIi64 "): %s\n", + cf, key_slice.ToString(true).c_str(), key, msg.c_str()); + shared->SetVerificationFailure(); +} + +void StressTest::VerificationAbort(SharedState* shared, std::string msg, int cf, + int64_t key, Slice value_from_db, + Slice value_from_expected) const { + auto key_str = Key(key); + fprintf(stderr, + "Verification failed for column family %d key %s (%" PRIi64 + "): value_from_db: %s, value_from_expected: %s, msg: %s\n", + cf, Slice(key_str).ToString(true).c_str(), key, + value_from_db.ToString(true).c_str(), + value_from_expected.ToString(true).c_str(), msg.c_str()); + shared->SetVerificationFailure(); +} + +void StressTest::VerificationAbort(SharedState* shared, int cf, int64_t key, + const Slice& value, + const WideColumns& columns, + const WideColumns& expected_columns) const { + assert(shared); + + auto key_str = Key(key); + + fprintf(stderr, + "Verification failed for column family %d key %s (%" PRIi64 + "): Value and columns inconsistent: %s\n", + cf, Slice(key_str).ToString(/* hex */ true).c_str(), key, + DebugString(value, columns, expected_columns).c_str()); + + shared->SetVerificationFailure(); +} + +std::string StressTest::DebugString(const Slice& value, + const WideColumns& columns, + const WideColumns& expected_columns) { + std::ostringstream oss; + + oss << "value: " << value.ToString(/* hex */ true); + + auto dump = [](const WideColumns& cols, std::ostream& os) { + if (cols.empty()) { + return; + } + + os << std::hex; + + auto it = cols.begin(); + os << *it; + for (++it; it != cols.end(); ++it) { + os << ' ' << *it; + } + }; + + oss << ", columns: "; + dump(columns, oss); + + oss << ", expected_columns: "; + dump(expected_columns, oss); + + return oss.str(); +} + +void StressTest::PrintStatistics() { + if (dbstats) { + fprintf(stdout, "STATISTICS:\n%s\n", dbstats->ToString().c_str()); + } + if (dbstats_secondaries) { + fprintf(stdout, "Secondary instances STATISTICS:\n%s\n", + dbstats_secondaries->ToString().c_str()); + } +} + +// Currently PreloadDb has to be single-threaded. +void StressTest::PreloadDbAndReopenAsReadOnly(int64_t number_of_keys, + SharedState* shared) { + WriteOptions write_opts; + write_opts.disableWAL = FLAGS_disable_wal; + if (FLAGS_sync) { + write_opts.sync = true; + } + if (FLAGS_rate_limit_auto_wal_flush) { + write_opts.rate_limiter_priority = Env::IO_USER; + } + char value[100]; + int cf_idx = 0; + Status s; + for (auto cfh : column_families_) { + for (int64_t k = 0; k != number_of_keys; ++k) { + const std::string key = Key(k); + + constexpr uint32_t value_base = 0; + const size_t sz = GenerateValue(value_base, value, sizeof(value)); + + const Slice v(value, sz); + + shared->Put(cf_idx, k, value_base, true /* pending */); + + std::string ts; + if (FLAGS_user_timestamp_size > 0) { + ts = GetNowNanos(); + } + + if (FLAGS_use_merge) { + if (!FLAGS_use_txn) { + if (FLAGS_user_timestamp_size > 0) { + s = db_->Merge(write_opts, cfh, key, ts, v); + } else { + s = db_->Merge(write_opts, cfh, key, v); + } + } else { +#ifndef ROCKSDB_LITE + Transaction* txn; + s = NewTxn(write_opts, &txn); + if (s.ok()) { + s = txn->Merge(cfh, key, v); + if (s.ok()) { + s = CommitTxn(txn); + } + } +#endif + } + } else if (FLAGS_use_put_entity_one_in > 0) { + s = db_->PutEntity(write_opts, cfh, key, + GenerateWideColumns(value_base, v)); + } else { + if (!FLAGS_use_txn) { + if (FLAGS_user_timestamp_size > 0) { + s = db_->Put(write_opts, cfh, key, ts, v); + } else { + s = db_->Put(write_opts, cfh, key, v); + } + } else { +#ifndef ROCKSDB_LITE + Transaction* txn; + s = NewTxn(write_opts, &txn); + if (s.ok()) { + s = txn->Put(cfh, key, v); + if (s.ok()) { + s = CommitTxn(txn); + } + } +#endif + } + } + + shared->Put(cf_idx, k, value_base, false /* pending */); + if (!s.ok()) { + break; + } + } + if (!s.ok()) { + break; + } + ++cf_idx; + } + if (s.ok()) { + s = db_->Flush(FlushOptions(), column_families_); + } + if (s.ok()) { + for (auto cf : column_families_) { + delete cf; + } + column_families_.clear(); + delete db_; + db_ = nullptr; +#ifndef ROCKSDB_LITE + txn_db_ = nullptr; +#endif + + db_preload_finished_.store(true); + auto now = clock_->NowMicros(); + fprintf(stdout, "%s Reopening database in read-only\n", + clock_->TimeToString(now / 1000000).c_str()); + // Reopen as read-only, can ignore all options related to updates + Open(shared); + } else { + fprintf(stderr, "Failed to preload db"); + exit(1); + } +} + +Status StressTest::SetOptions(ThreadState* thread) { + assert(FLAGS_set_options_one_in > 0); + std::unordered_map<std::string, std::string> opts; + std::string name = + options_index_[thread->rand.Next() % options_index_.size()]; + int value_idx = thread->rand.Next() % options_table_[name].size(); + if (name == "level0_file_num_compaction_trigger" || + name == "level0_slowdown_writes_trigger" || + name == "level0_stop_writes_trigger") { + opts["level0_file_num_compaction_trigger"] = + options_table_["level0_file_num_compaction_trigger"][value_idx]; + opts["level0_slowdown_writes_trigger"] = + options_table_["level0_slowdown_writes_trigger"][value_idx]; + opts["level0_stop_writes_trigger"] = + options_table_["level0_stop_writes_trigger"][value_idx]; + } else { + opts[name] = options_table_[name][value_idx]; + } + + int rand_cf_idx = thread->rand.Next() % FLAGS_column_families; + auto cfh = column_families_[rand_cf_idx]; + return db_->SetOptions(cfh, opts); +} + +#ifndef ROCKSDB_LITE +void StressTest::ProcessRecoveredPreparedTxns(SharedState* shared) { + assert(txn_db_); + std::vector<Transaction*> recovered_prepared_trans; + txn_db_->GetAllPreparedTransactions(&recovered_prepared_trans); + for (Transaction* txn : recovered_prepared_trans) { + ProcessRecoveredPreparedTxnsHelper(txn, shared); + delete txn; + } + recovered_prepared_trans.clear(); + txn_db_->GetAllPreparedTransactions(&recovered_prepared_trans); + assert(recovered_prepared_trans.size() == 0); +} + +void StressTest::ProcessRecoveredPreparedTxnsHelper(Transaction* txn, + SharedState* shared) { + thread_local Random rand(static_cast<uint32_t>(FLAGS_seed)); + for (size_t i = 0; i < column_families_.size(); ++i) { + std::unique_ptr<WBWIIterator> wbwi_iter( + txn->GetWriteBatch()->NewIterator(column_families_[i])); + for (wbwi_iter->SeekToFirst(); wbwi_iter->Valid(); wbwi_iter->Next()) { + uint64_t key_val; + if (GetIntVal(wbwi_iter->Entry().key.ToString(), &key_val)) { + shared->Put(static_cast<int>(i) /* cf_idx */, key_val, + 0 /* value_base */, true /* pending */); + } + } + } + if (rand.OneIn(2)) { + Status s = txn->Commit(); + assert(s.ok()); + } else { + Status s = txn->Rollback(); + assert(s.ok()); + } +} + +Status StressTest::NewTxn(WriteOptions& write_opts, Transaction** txn) { + if (!FLAGS_use_txn) { + return Status::InvalidArgument("NewTxn when FLAGS_use_txn is not set"); + } + write_opts.disableWAL = FLAGS_disable_wal; + static std::atomic<uint64_t> txn_id = {0}; + TransactionOptions txn_options; + txn_options.use_only_the_last_commit_time_batch_for_recovery = + FLAGS_use_only_the_last_commit_time_batch_for_recovery; + txn_options.lock_timeout = 600000; // 10 min + txn_options.deadlock_detect = true; + *txn = txn_db_->BeginTransaction(write_opts, txn_options); + auto istr = std::to_string(txn_id.fetch_add(1)); + Status s = (*txn)->SetName("xid" + istr); + return s; +} + +Status StressTest::CommitTxn(Transaction* txn, ThreadState* thread) { + if (!FLAGS_use_txn) { + return Status::InvalidArgument("CommitTxn when FLAGS_use_txn is not set"); + } + assert(txn_db_); + Status s = txn->Prepare(); + std::shared_ptr<const Snapshot> timestamped_snapshot; + if (s.ok()) { + if (thread && FLAGS_create_timestamped_snapshot_one_in && + thread->rand.OneIn(FLAGS_create_timestamped_snapshot_one_in)) { + uint64_t ts = db_stress_env->NowNanos(); + s = txn->CommitAndTryCreateSnapshot(/*notifier=*/nullptr, ts, + ×tamped_snapshot); + + std::pair<Status, std::shared_ptr<const Snapshot>> res; + if (thread->tid == 0) { + uint64_t now = db_stress_env->NowNanos(); + res = txn_db_->CreateTimestampedSnapshot(now); + if (res.first.ok()) { + assert(res.second); + assert(res.second->GetTimestamp() == now); + if (timestamped_snapshot) { + assert(res.second->GetTimestamp() > + timestamped_snapshot->GetTimestamp()); + } + } else { + assert(!res.second); + } + } + } else { + s = txn->Commit(); + } + } + if (thread && FLAGS_create_timestamped_snapshot_one_in > 0 && + thread->rand.OneInOpt(50000)) { + uint64_t now = db_stress_env->NowNanos(); + constexpr uint64_t time_diff = static_cast<uint64_t>(1000) * 1000 * 1000; + txn_db_->ReleaseTimestampedSnapshotsOlderThan(now - time_diff); + } + delete txn; + return s; +} + +Status StressTest::RollbackTxn(Transaction* txn) { + if (!FLAGS_use_txn) { + return Status::InvalidArgument( + "RollbackTxn when FLAGS_use_txn is not" + " set"); + } + Status s = txn->Rollback(); + delete txn; + return s; +} +#endif + +void StressTest::OperateDb(ThreadState* thread) { + ReadOptions read_opts(FLAGS_verify_checksum, true); + read_opts.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + read_opts.async_io = FLAGS_async_io; + read_opts.adaptive_readahead = FLAGS_adaptive_readahead; + read_opts.readahead_size = FLAGS_readahead_size; + WriteOptions write_opts; + if (FLAGS_rate_limit_auto_wal_flush) { + write_opts.rate_limiter_priority = Env::IO_USER; + } + auto shared = thread->shared; + char value[100]; + std::string from_db; + if (FLAGS_sync) { + write_opts.sync = true; + } + write_opts.disableWAL = FLAGS_disable_wal; + write_opts.protection_bytes_per_key = FLAGS_batch_protection_bytes_per_key; + const int prefix_bound = static_cast<int>(FLAGS_readpercent) + + static_cast<int>(FLAGS_prefixpercent); + const int write_bound = prefix_bound + static_cast<int>(FLAGS_writepercent); + const int del_bound = write_bound + static_cast<int>(FLAGS_delpercent); + const int delrange_bound = + del_bound + static_cast<int>(FLAGS_delrangepercent); + const int iterate_bound = + delrange_bound + static_cast<int>(FLAGS_iterpercent); + + const uint64_t ops_per_open = FLAGS_ops_per_thread / (FLAGS_reopen + 1); + +#ifndef NDEBUG + if (FLAGS_read_fault_one_in) { + fault_fs_guard->SetThreadLocalReadErrorContext(thread->shared->GetSeed(), + FLAGS_read_fault_one_in); + } +#endif // NDEBUG + if (FLAGS_write_fault_one_in) { + IOStatus error_msg; + if (FLAGS_injest_error_severity <= 1 || FLAGS_injest_error_severity > 2) { + error_msg = IOStatus::IOError("Retryable IO Error"); + error_msg.SetRetryable(true); + } else if (FLAGS_injest_error_severity == 2) { + // Ingest the fatal error + error_msg = IOStatus::IOError("Fatal IO Error"); + error_msg.SetDataLoss(true); + } + std::vector<FileType> types = {FileType::kTableFile, + FileType::kDescriptorFile, + FileType::kCurrentFile}; + fault_fs_guard->SetRandomWriteError( + thread->shared->GetSeed(), FLAGS_write_fault_one_in, error_msg, + /*inject_for_all_file_types=*/false, types); + } + thread->stats.Start(); + for (int open_cnt = 0; open_cnt <= FLAGS_reopen; ++open_cnt) { + if (thread->shared->HasVerificationFailedYet() || + thread->shared->ShouldStopTest()) { + break; + } + if (open_cnt != 0) { + thread->stats.FinishedSingleOp(); + MutexLock l(thread->shared->GetMutex()); + while (!thread->snapshot_queue.empty()) { + db_->ReleaseSnapshot(thread->snapshot_queue.front().second.snapshot); + delete thread->snapshot_queue.front().second.key_vec; + thread->snapshot_queue.pop(); + } + thread->shared->IncVotedReopen(); + if (thread->shared->AllVotedReopen()) { + thread->shared->GetStressTest()->Reopen(thread); + thread->shared->GetCondVar()->SignalAll(); + } else { + thread->shared->GetCondVar()->Wait(); + } + // Commenting this out as we don't want to reset stats on each open. + // thread->stats.Start(); + } + + for (uint64_t i = 0; i < ops_per_open; i++) { + if (thread->shared->HasVerificationFailedYet()) { + break; + } + + // Change Options + if (thread->rand.OneInOpt(FLAGS_set_options_one_in)) { + SetOptions(thread); + } + + if (thread->rand.OneInOpt(FLAGS_set_in_place_one_in)) { + options_.inplace_update_support ^= options_.inplace_update_support; + } + + if (thread->tid == 0 && FLAGS_verify_db_one_in > 0 && + thread->rand.OneIn(FLAGS_verify_db_one_in)) { + ContinuouslyVerifyDb(thread); + if (thread->shared->ShouldStopTest()) { + break; + } + } + + MaybeClearOneColumnFamily(thread); + + if (thread->rand.OneInOpt(FLAGS_manual_wal_flush_one_in)) { + bool sync = thread->rand.OneIn(2) ? true : false; + Status s = db_->FlushWAL(sync); + if (!s.ok() && !(sync && s.IsNotSupported())) { + fprintf(stderr, "FlushWAL(sync=%s) failed: %s\n", + (sync ? "true" : "false"), s.ToString().c_str()); + } + } + + if (thread->rand.OneInOpt(FLAGS_sync_wal_one_in)) { + Status s = db_->SyncWAL(); + if (!s.ok() && !s.IsNotSupported()) { + fprintf(stderr, "SyncWAL() failed: %s\n", s.ToString().c_str()); + } + } + + int rand_column_family = thread->rand.Next() % FLAGS_column_families; + ColumnFamilyHandle* column_family = column_families_[rand_column_family]; + + if (thread->rand.OneInOpt(FLAGS_compact_files_one_in)) { + TestCompactFiles(thread, column_family); + } + + int64_t rand_key = GenerateOneKey(thread, i); + std::string keystr = Key(rand_key); + Slice key = keystr; + + if (thread->rand.OneInOpt(FLAGS_compact_range_one_in)) { + TestCompactRange(thread, rand_key, key, column_family); + if (thread->shared->HasVerificationFailedYet()) { + break; + } + } + + std::vector<int> rand_column_families = + GenerateColumnFamilies(FLAGS_column_families, rand_column_family); + + if (thread->rand.OneInOpt(FLAGS_flush_one_in)) { + Status status = TestFlush(rand_column_families); + if (!status.ok()) { + fprintf(stdout, "Unable to perform Flush(): %s\n", + status.ToString().c_str()); + } + } + +#ifndef ROCKSDB_LITE + // Verify GetLiveFiles with a 1 in N chance. + if (thread->rand.OneInOpt(FLAGS_get_live_files_one_in) && + !FLAGS_write_fault_one_in) { + Status status = VerifyGetLiveFiles(); + if (!status.ok()) { + VerificationAbort(shared, "VerifyGetLiveFiles status not OK", status); + } + } + + // Verify GetSortedWalFiles with a 1 in N chance. + if (thread->rand.OneInOpt(FLAGS_get_sorted_wal_files_one_in)) { + Status status = VerifyGetSortedWalFiles(); + if (!status.ok()) { + VerificationAbort(shared, "VerifyGetSortedWalFiles status not OK", + status); + } + } + + // Verify GetCurrentWalFile with a 1 in N chance. + if (thread->rand.OneInOpt(FLAGS_get_current_wal_file_one_in)) { + Status status = VerifyGetCurrentWalFile(); + if (!status.ok()) { + VerificationAbort(shared, "VerifyGetCurrentWalFile status not OK", + status); + } + } +#endif // !ROCKSDB_LITE + + if (thread->rand.OneInOpt(FLAGS_pause_background_one_in)) { + Status status = TestPauseBackground(thread); + if (!status.ok()) { + VerificationAbort( + shared, "Pause/ContinueBackgroundWork status not OK", status); + } + } + +#ifndef ROCKSDB_LITE + if (thread->rand.OneInOpt(FLAGS_verify_checksum_one_in)) { + Status status = db_->VerifyChecksum(); + if (!status.ok()) { + VerificationAbort(shared, "VerifyChecksum status not OK", status); + } + } + + if (thread->rand.OneInOpt(FLAGS_get_property_one_in)) { + TestGetProperty(thread); + } +#endif + + std::vector<int64_t> rand_keys = GenerateKeys(rand_key); + + if (thread->rand.OneInOpt(FLAGS_ingest_external_file_one_in)) { + TestIngestExternalFile(thread, rand_column_families, rand_keys); + } + + if (thread->rand.OneInOpt(FLAGS_backup_one_in)) { + // Beyond a certain DB size threshold, this test becomes heavier than + // it's worth. + uint64_t total_size = 0; + if (FLAGS_backup_max_size > 0) { + std::vector<FileAttributes> files; + db_stress_env->GetChildrenFileAttributes(FLAGS_db, &files); + for (auto& file : files) { + total_size += file.size_bytes; + } + } + + if (total_size <= FLAGS_backup_max_size) { + Status s = TestBackupRestore(thread, rand_column_families, rand_keys); + if (!s.ok()) { + VerificationAbort(shared, "Backup/restore gave inconsistent state", + s); + } + } + } + + if (thread->rand.OneInOpt(FLAGS_checkpoint_one_in)) { + Status s = TestCheckpoint(thread, rand_column_families, rand_keys); + if (!s.ok()) { + VerificationAbort(shared, "Checkpoint gave inconsistent state", s); + } + } + +#ifndef ROCKSDB_LITE + if (thread->rand.OneInOpt(FLAGS_approximate_size_one_in)) { + Status s = + TestApproximateSize(thread, i, rand_column_families, rand_keys); + if (!s.ok()) { + VerificationAbort(shared, "ApproximateSize Failed", s); + } + } +#endif // !ROCKSDB_LITE + if (thread->rand.OneInOpt(FLAGS_acquire_snapshot_one_in)) { + TestAcquireSnapshot(thread, rand_column_family, keystr, i); + } + + /*always*/ { + Status s = MaybeReleaseSnapshots(thread, i); + if (!s.ok()) { + VerificationAbort(shared, "Snapshot gave inconsistent state", s); + } + } + + // Assign timestamps if necessary. + std::string read_ts_str; + Slice read_ts; + if (FLAGS_user_timestamp_size > 0) { + read_ts_str = GetNowNanos(); + read_ts = read_ts_str; + read_opts.timestamp = &read_ts; + } + + int prob_op = thread->rand.Uniform(100); + // Reset this in case we pick something other than a read op. We don't + // want to use a stale value when deciding at the beginning of the loop + // whether to vote to reopen + if (prob_op >= 0 && prob_op < static_cast<int>(FLAGS_readpercent)) { + assert(0 <= prob_op); + // OPERATION read + if (FLAGS_use_multiget) { + // Leave room for one more iteration of the loop with a single key + // batch. This is to ensure that each thread does exactly the same + // number of ops + int multiget_batch_size = static_cast<int>( + std::min(static_cast<uint64_t>(thread->rand.Uniform(64)), + FLAGS_ops_per_thread - i - 1)); + // If its the last iteration, ensure that multiget_batch_size is 1 + multiget_batch_size = std::max(multiget_batch_size, 1); + rand_keys = GenerateNKeys(thread, multiget_batch_size, i); + TestMultiGet(thread, read_opts, rand_column_families, rand_keys); + i += multiget_batch_size - 1; + } else { + TestGet(thread, read_opts, rand_column_families, rand_keys); + } + } else if (prob_op < prefix_bound) { + assert(static_cast<int>(FLAGS_readpercent) <= prob_op); + // OPERATION prefix scan + // keys are 8 bytes long, prefix size is FLAGS_prefix_size. There are + // (8 - FLAGS_prefix_size) bytes besides the prefix. So there will + // be 2 ^ ((8 - FLAGS_prefix_size) * 8) possible keys with the same + // prefix + TestPrefixScan(thread, read_opts, rand_column_families, rand_keys); + } else if (prob_op < write_bound) { + assert(prefix_bound <= prob_op); + // OPERATION write + TestPut(thread, write_opts, read_opts, rand_column_families, rand_keys, + value); + } else if (prob_op < del_bound) { + assert(write_bound <= prob_op); + // OPERATION delete + TestDelete(thread, write_opts, rand_column_families, rand_keys); + } else if (prob_op < delrange_bound) { + assert(del_bound <= prob_op); + // OPERATION delete range + TestDeleteRange(thread, write_opts, rand_column_families, rand_keys); + } else if (prob_op < iterate_bound) { + assert(delrange_bound <= prob_op); + // OPERATION iterate + if (!FLAGS_skip_verifydb && + thread->rand.OneInOpt( + FLAGS_verify_iterator_with_expected_state_one_in)) { + TestIterateAgainstExpected(thread, read_opts, rand_column_families, + rand_keys); + } else { + int num_seeks = static_cast<int>( + std::min(static_cast<uint64_t>(thread->rand.Uniform(4)), + FLAGS_ops_per_thread - i - 1)); + rand_keys = GenerateNKeys(thread, num_seeks, i); + i += num_seeks - 1; + TestIterate(thread, read_opts, rand_column_families, rand_keys); + } + } else { + assert(iterate_bound <= prob_op); + TestCustomOperations(thread, rand_column_families); + } + thread->stats.FinishedSingleOp(); + } + } + while (!thread->snapshot_queue.empty()) { + db_->ReleaseSnapshot(thread->snapshot_queue.front().second.snapshot); + delete thread->snapshot_queue.front().second.key_vec; + thread->snapshot_queue.pop(); + } + + thread->stats.Stop(); +} + +#ifndef ROCKSDB_LITE +// Generated a list of keys that close to boundaries of SST keys. +// If there isn't any SST file in the DB, return empty list. +std::vector<std::string> StressTest::GetWhiteBoxKeys(ThreadState* thread, + DB* db, + ColumnFamilyHandle* cfh, + size_t num_keys) { + ColumnFamilyMetaData cfmd; + db->GetColumnFamilyMetaData(cfh, &cfmd); + std::vector<std::string> boundaries; + for (const LevelMetaData& lmd : cfmd.levels) { + for (const SstFileMetaData& sfmd : lmd.files) { + // If FLAGS_user_timestamp_size > 0, then both smallestkey and largestkey + // have timestamps. + const auto& skey = sfmd.smallestkey; + const auto& lkey = sfmd.largestkey; + assert(skey.size() >= FLAGS_user_timestamp_size); + assert(lkey.size() >= FLAGS_user_timestamp_size); + boundaries.push_back( + skey.substr(0, skey.size() - FLAGS_user_timestamp_size)); + boundaries.push_back( + lkey.substr(0, lkey.size() - FLAGS_user_timestamp_size)); + } + } + if (boundaries.empty()) { + return {}; + } + + std::vector<std::string> ret; + for (size_t j = 0; j < num_keys; j++) { + std::string k = + boundaries[thread->rand.Uniform(static_cast<int>(boundaries.size()))]; + if (thread->rand.OneIn(3)) { + // Reduce one byte from the string + for (int i = static_cast<int>(k.length()) - 1; i >= 0; i--) { + uint8_t cur = k[i]; + if (cur > 0) { + k[i] = static_cast<char>(cur - 1); + break; + } else if (i > 0) { + k[i] = 0xFFu; + } + } + } else if (thread->rand.OneIn(2)) { + // Add one byte to the string + for (int i = static_cast<int>(k.length()) - 1; i >= 0; i--) { + uint8_t cur = k[i]; + if (cur < 255) { + k[i] = static_cast<char>(cur + 1); + break; + } else if (i > 0) { + k[i] = 0x00; + } + } + } + ret.push_back(k); + } + return ret; +} +#endif // !ROCKSDB_LITE + +// Given a key K, this creates an iterator which scans to K and then +// does a random sequence of Next/Prev operations. +Status StressTest::TestIterate(ThreadState* thread, + const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) { + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + ManagedSnapshot snapshot_guard(db_); + + ReadOptions ro = read_opts; + ro.snapshot = snapshot_guard.snapshot(); + + std::string read_ts_str; + Slice read_ts_slice; + MaybeUseOlderTimestampForRangeScan(thread, read_ts_str, read_ts_slice, ro); + + bool expect_total_order = false; + if (thread->rand.OneIn(16)) { + // When prefix extractor is used, it's useful to cover total order seek. + ro.total_order_seek = true; + expect_total_order = true; + } else if (thread->rand.OneIn(4)) { + ro.total_order_seek = false; + ro.auto_prefix_mode = true; + expect_total_order = true; + } else if (options_.prefix_extractor.get() == nullptr) { + expect_total_order = true; + } + + std::string upper_bound_str; + Slice upper_bound; + if (thread->rand.OneIn(16)) { + // With a 1/16 chance, set an iterator upper bound. + // Note: upper_bound can be smaller than the seek key. + const int64_t rand_upper_key = GenerateOneKey(thread, FLAGS_ops_per_thread); + upper_bound_str = Key(rand_upper_key); + upper_bound = Slice(upper_bound_str); + ro.iterate_upper_bound = &upper_bound; + } + std::string lower_bound_str; + Slice lower_bound; + if (thread->rand.OneIn(16)) { + // With a 1/16 chance, enable iterator lower bound. + // Note: lower_bound can be greater than the seek key. + const int64_t rand_lower_key = GenerateOneKey(thread, FLAGS_ops_per_thread); + lower_bound_str = Key(rand_lower_key); + lower_bound = Slice(lower_bound_str); + ro.iterate_lower_bound = &lower_bound; + } + + ColumnFamilyHandle* const cfh = column_families_[rand_column_families[0]]; + assert(cfh); + + std::unique_ptr<Iterator> iter(db_->NewIterator(ro, cfh)); + + std::vector<std::string> key_strs; + if (thread->rand.OneIn(16)) { + // Generate keys close to lower or upper bound of SST files. + key_strs = GetWhiteBoxKeys(thread, db_, cfh, rand_keys.size()); + } + if (key_strs.empty()) { + // Use the random keys passed in. + for (int64_t rkey : rand_keys) { + key_strs.push_back(Key(rkey)); + } + } + + std::string op_logs; + constexpr size_t kOpLogsLimit = 10000; + + for (const std::string& key_str : key_strs) { + if (op_logs.size() > kOpLogsLimit) { + // Shouldn't take too much memory for the history log. Clear it. + op_logs = "(cleared...)\n"; + } + + if (ro.iterate_upper_bound != nullptr && thread->rand.OneIn(2)) { + // With a 1/2 chance, change the upper bound. + // It is possible that it is changed before first use, but there is no + // problem with that. + const int64_t rand_upper_key = + GenerateOneKey(thread, FLAGS_ops_per_thread); + upper_bound_str = Key(rand_upper_key); + upper_bound = Slice(upper_bound_str); + } + if (ro.iterate_lower_bound != nullptr && thread->rand.OneIn(4)) { + // With a 1/4 chance, change the lower bound. + // It is possible that it is changed before first use, but there is no + // problem with that. + const int64_t rand_lower_key = + GenerateOneKey(thread, FLAGS_ops_per_thread); + lower_bound_str = Key(rand_lower_key); + lower_bound = Slice(lower_bound_str); + } + + // Record some options to op_logs + op_logs += "total_order_seek: "; + op_logs += (ro.total_order_seek ? "1 " : "0 "); + op_logs += "auto_prefix_mode: "; + op_logs += (ro.auto_prefix_mode ? "1 " : "0 "); + if (ro.iterate_upper_bound != nullptr) { + op_logs += "ub: " + upper_bound.ToString(true) + " "; + } + if (ro.iterate_lower_bound != nullptr) { + op_logs += "lb: " + lower_bound.ToString(true) + " "; + } + + // Set up an iterator, perform the same operations without bounds and with + // total order seek, and compare the results. This is to identify bugs + // related to bounds, prefix extractor, or reseeking. Sometimes we are + // comparing iterators with the same set-up, and it doesn't hurt to check + // them to be equal. + // + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions cmp_ro; + cmp_ro.timestamp = ro.timestamp; + cmp_ro.iter_start_ts = ro.iter_start_ts; + cmp_ro.snapshot = snapshot_guard.snapshot(); + cmp_ro.total_order_seek = true; + + ColumnFamilyHandle* const cmp_cfh = + GetControlCfh(thread, rand_column_families[0]); + assert(cmp_cfh); + + std::unique_ptr<Iterator> cmp_iter(db_->NewIterator(cmp_ro, cmp_cfh)); + + bool diverged = false; + + Slice key(key_str); + + const bool support_seek_first_or_last = expect_total_order; + + LastIterateOp last_op; + if (support_seek_first_or_last && thread->rand.OneIn(100)) { + iter->SeekToFirst(); + cmp_iter->SeekToFirst(); + last_op = kLastOpSeekToFirst; + op_logs += "STF "; + } else if (support_seek_first_or_last && thread->rand.OneIn(100)) { + iter->SeekToLast(); + cmp_iter->SeekToLast(); + last_op = kLastOpSeekToLast; + op_logs += "STL "; + } else if (thread->rand.OneIn(8)) { + iter->SeekForPrev(key); + cmp_iter->SeekForPrev(key); + last_op = kLastOpSeekForPrev; + op_logs += "SFP " + key.ToString(true) + " "; + } else { + iter->Seek(key); + cmp_iter->Seek(key); + last_op = kLastOpSeek; + op_logs += "S " + key.ToString(true) + " "; + } + + VerifyIterator(thread, cmp_cfh, ro, iter.get(), cmp_iter.get(), last_op, + key, op_logs, &diverged); + + const bool no_reverse = + (FLAGS_memtablerep == "prefix_hash" && !expect_total_order); + for (uint64_t i = 0; i < FLAGS_num_iterations && iter->Valid(); ++i) { + if (no_reverse || thread->rand.OneIn(2)) { + iter->Next(); + if (!diverged) { + assert(cmp_iter->Valid()); + cmp_iter->Next(); + } + op_logs += "N"; + } else { + iter->Prev(); + if (!diverged) { + assert(cmp_iter->Valid()); + cmp_iter->Prev(); + } + op_logs += "P"; + } + + last_op = kLastOpNextOrPrev; + + VerifyIterator(thread, cmp_cfh, ro, iter.get(), cmp_iter.get(), last_op, + key, op_logs, &diverged); + } + + thread->stats.AddIterations(1); + + op_logs += "; "; + } + + return Status::OK(); +} + +#ifndef ROCKSDB_LITE +// Test the return status of GetLiveFiles. +Status StressTest::VerifyGetLiveFiles() const { + std::vector<std::string> live_file; + uint64_t manifest_size = 0; + return db_->GetLiveFiles(live_file, &manifest_size); +} + +// Test the return status of GetSortedWalFiles. +Status StressTest::VerifyGetSortedWalFiles() const { + VectorLogPtr log_ptr; + return db_->GetSortedWalFiles(log_ptr); +} + +// Test the return status of GetCurrentWalFile. +Status StressTest::VerifyGetCurrentWalFile() const { + std::unique_ptr<LogFile> cur_wal_file; + return db_->GetCurrentWalFile(&cur_wal_file); +} +#endif // !ROCKSDB_LITE + +// Compare the two iterator, iter and cmp_iter are in the same position, +// unless iter might be made invalidate or undefined because of +// upper or lower bounds, or prefix extractor. +// Will flag failure if the verification fails. +// diverged = true if the two iterator is already diverged. +// True if verification passed, false if not. +void StressTest::VerifyIterator(ThreadState* thread, + ColumnFamilyHandle* cmp_cfh, + const ReadOptions& ro, Iterator* iter, + Iterator* cmp_iter, LastIterateOp op, + const Slice& seek_key, + const std::string& op_logs, bool* diverged) { + assert(diverged); + + if (*diverged) { + return; + } + + if (ro.iter_start_ts != nullptr) { + assert(FLAGS_user_timestamp_size > 0); + // We currently do not verify iterator when dumping history of internal + // keys. + *diverged = true; + return; + } + + if (op == kLastOpSeekToFirst && ro.iterate_lower_bound != nullptr) { + // SeekToFirst() with lower bound is not well defined. + *diverged = true; + return; + } else if (op == kLastOpSeekToLast && ro.iterate_upper_bound != nullptr) { + // SeekToLast() with higher bound is not well defined. + *diverged = true; + return; + } else if (op == kLastOpSeek && ro.iterate_lower_bound != nullptr && + (options_.comparator->CompareWithoutTimestamp( + *ro.iterate_lower_bound, /*a_has_ts=*/false, seek_key, + /*b_has_ts=*/false) >= 0 || + (ro.iterate_upper_bound != nullptr && + options_.comparator->CompareWithoutTimestamp( + *ro.iterate_lower_bound, /*a_has_ts=*/false, + *ro.iterate_upper_bound, /*b_has_ts*/ false) >= 0))) { + // Lower bound behavior is not well defined if it is larger than + // seek key or upper bound. Disable the check for now. + *diverged = true; + return; + } else if (op == kLastOpSeekForPrev && ro.iterate_upper_bound != nullptr && + (options_.comparator->CompareWithoutTimestamp( + *ro.iterate_upper_bound, /*a_has_ts=*/false, seek_key, + /*b_has_ts=*/false) <= 0 || + (ro.iterate_lower_bound != nullptr && + options_.comparator->CompareWithoutTimestamp( + *ro.iterate_lower_bound, /*a_has_ts=*/false, + *ro.iterate_upper_bound, /*b_has_ts=*/false) >= 0))) { + // Uppder bound behavior is not well defined if it is smaller than + // seek key or lower bound. Disable the check for now. + *diverged = true; + return; + } + + const SliceTransform* pe = (ro.total_order_seek || ro.auto_prefix_mode) + ? nullptr + : options_.prefix_extractor.get(); + const Comparator* cmp = options_.comparator; + + if (iter->Valid() && !cmp_iter->Valid()) { + if (pe != nullptr) { + if (!pe->InDomain(seek_key)) { + // Prefix seek a non-in-domain key is undefined. Skip checking for + // this scenario. + *diverged = true; + return; + } else if (!pe->InDomain(iter->key())) { + // out of range is iterator key is not in domain anymore. + *diverged = true; + return; + } else if (pe->Transform(iter->key()) != pe->Transform(seek_key)) { + *diverged = true; + return; + } + } + fprintf(stderr, + "Control interator is invalid but iterator has key %s " + "%s\n", + iter->key().ToString(true).c_str(), op_logs.c_str()); + + *diverged = true; + } else if (cmp_iter->Valid()) { + // Iterator is not valid. It can be legimate if it has already been + // out of upper or lower bound, or filtered out by prefix iterator. + const Slice& total_order_key = cmp_iter->key(); + + if (pe != nullptr) { + if (!pe->InDomain(seek_key)) { + // Prefix seek a non-in-domain key is undefined. Skip checking for + // this scenario. + *diverged = true; + return; + } + + if (!pe->InDomain(total_order_key) || + pe->Transform(total_order_key) != pe->Transform(seek_key)) { + // If the prefix is exhausted, the only thing needs to check + // is the iterator isn't return a position in prefix. + // Either way, checking can stop from here. + *diverged = true; + if (!iter->Valid() || !pe->InDomain(iter->key()) || + pe->Transform(iter->key()) != pe->Transform(seek_key)) { + return; + } + fprintf(stderr, + "Iterator stays in prefix but contol doesn't" + " iterator key %s control iterator key %s %s\n", + iter->key().ToString(true).c_str(), + cmp_iter->key().ToString(true).c_str(), op_logs.c_str()); + } + } + // Check upper or lower bounds. + if (!*diverged) { + if ((iter->Valid() && iter->key() != cmp_iter->key()) || + (!iter->Valid() && + (ro.iterate_upper_bound == nullptr || + cmp->CompareWithoutTimestamp(total_order_key, /*a_has_ts=*/false, + *ro.iterate_upper_bound, + /*b_has_ts=*/false) < 0) && + (ro.iterate_lower_bound == nullptr || + cmp->CompareWithoutTimestamp(total_order_key, /*a_has_ts=*/false, + *ro.iterate_lower_bound, + /*b_has_ts=*/false) > 0))) { + fprintf(stderr, + "Iterator diverged from control iterator which" + " has value %s %s\n", + total_order_key.ToString(true).c_str(), op_logs.c_str()); + if (iter->Valid()) { + fprintf(stderr, "iterator has value %s\n", + iter->key().ToString(true).c_str()); + } else { + fprintf(stderr, "iterator is not valid\n"); + } + *diverged = true; + } + } + } + + if (!*diverged && iter->Valid()) { + const WideColumns expected_columns = + GenerateExpectedWideColumns(GetValueBase(iter->value()), iter->value()); + if (iter->columns() != expected_columns) { + fprintf(stderr, "Value and columns inconsistent for iterator: %s\n", + DebugString(iter->value(), iter->columns(), expected_columns) + .c_str()); + + *diverged = true; + } + } + + if (*diverged) { + fprintf(stderr, "Control CF %s\n", cmp_cfh->GetName().c_str()); + thread->stats.AddErrors(1); + // Fail fast to preserve the DB state. + thread->shared->SetVerificationFailure(); + } +} + +#ifdef ROCKSDB_LITE +Status StressTest::TestBackupRestore( + ThreadState* /* thread */, + const std::vector<int>& /* rand_column_families */, + const std::vector<int64_t>& /* rand_keys */) { + assert(false); + fprintf(stderr, + "RocksDB lite does not support " + "TestBackupRestore\n"); + std::terminate(); +} + +Status StressTest::TestCheckpoint( + ThreadState* /* thread */, + const std::vector<int>& /* rand_column_families */, + const std::vector<int64_t>& /* rand_keys */) { + assert(false); + fprintf(stderr, + "RocksDB lite does not support " + "TestCheckpoint\n"); + std::terminate(); +} + +void StressTest::TestCompactFiles(ThreadState* /* thread */, + ColumnFamilyHandle* /* column_family */) { + assert(false); + fprintf(stderr, + "RocksDB lite does not support " + "CompactFiles\n"); + std::terminate(); +} +#else // ROCKSDB_LITE +Status StressTest::TestBackupRestore( + ThreadState* thread, const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) { + std::vector<std::unique_ptr<MutexLock>> locks; + if (ShouldAcquireMutexOnKey()) { + for (int rand_column_family : rand_column_families) { + // `rand_keys[0]` on each chosen CF will be verified. + locks.emplace_back(new MutexLock( + thread->shared->GetMutexForKey(rand_column_family, rand_keys[0]))); + } + } + + const std::string backup_dir = + FLAGS_db + "/.backup" + std::to_string(thread->tid); + const std::string restore_dir = + FLAGS_db + "/.restore" + std::to_string(thread->tid); + BackupEngineOptions backup_opts(backup_dir); + // For debugging, get info_log from live options + backup_opts.info_log = db_->GetDBOptions().info_log.get(); + if (thread->rand.OneIn(10)) { + backup_opts.share_table_files = false; + } else { + backup_opts.share_table_files = true; + if (thread->rand.OneIn(5)) { + backup_opts.share_files_with_checksum = false; + } else { + backup_opts.share_files_with_checksum = true; + if (thread->rand.OneIn(2)) { + // old + backup_opts.share_files_with_checksum_naming = + BackupEngineOptions::kLegacyCrc32cAndFileSize; + } else { + // new + backup_opts.share_files_with_checksum_naming = + BackupEngineOptions::kUseDbSessionId; + } + if (thread->rand.OneIn(2)) { + backup_opts.share_files_with_checksum_naming = + backup_opts.share_files_with_checksum_naming | + BackupEngineOptions::kFlagIncludeFileSize; + } + } + } + if (thread->rand.OneIn(2)) { + backup_opts.schema_version = 1; + } else { + backup_opts.schema_version = 2; + } + BackupEngine* backup_engine = nullptr; + std::string from = "a backup/restore operation"; + Status s = BackupEngine::Open(db_stress_env, backup_opts, &backup_engine); + if (!s.ok()) { + from = "BackupEngine::Open"; + } + if (s.ok()) { + if (backup_opts.schema_version >= 2 && thread->rand.OneIn(2)) { + TEST_BackupMetaSchemaOptions test_opts; + test_opts.crc32c_checksums = thread->rand.OneIn(2) == 0; + test_opts.file_sizes = thread->rand.OneIn(2) == 0; + TEST_SetBackupMetaSchemaOptions(backup_engine, test_opts); + } + CreateBackupOptions create_opts; + if (FLAGS_disable_wal) { + // The verification can only work when latest value of `key` is backed up, + // which requires flushing in case of WAL disabled. + // + // Note this triggers a flush with a key lock held. Meanwhile, operations + // like flush/compaction may attempt to grab key locks like in + // `DbStressCompactionFilter`. The philosophy around preventing deadlock + // is the background operation key lock acquisition only tries but does + // not wait for the lock. So here in the foreground it is OK to hold the + // lock and wait on a background operation (flush). + create_opts.flush_before_backup = true; + } + s = backup_engine->CreateNewBackup(create_opts, db_); + if (!s.ok()) { + from = "BackupEngine::CreateNewBackup"; + } + } + if (s.ok()) { + delete backup_engine; + backup_engine = nullptr; + s = BackupEngine::Open(db_stress_env, backup_opts, &backup_engine); + if (!s.ok()) { + from = "BackupEngine::Open (again)"; + } + } + std::vector<BackupInfo> backup_info; + // If inplace_not_restore, we verify the backup by opening it as a + // read-only DB. If !inplace_not_restore, we restore it to a temporary + // directory for verification. + bool inplace_not_restore = thread->rand.OneIn(3); + if (s.ok()) { + backup_engine->GetBackupInfo(&backup_info, + /*include_file_details*/ inplace_not_restore); + if (backup_info.empty()) { + s = Status::NotFound("no backups found"); + from = "BackupEngine::GetBackupInfo"; + } + } + if (s.ok() && thread->rand.OneIn(2)) { + s = backup_engine->VerifyBackup( + backup_info.front().backup_id, + thread->rand.OneIn(2) /* verify_with_checksum */); + if (!s.ok()) { + from = "BackupEngine::VerifyBackup"; + } + } + const bool allow_persistent = thread->tid == 0; // not too many + bool from_latest = false; + int count = static_cast<int>(backup_info.size()); + if (s.ok() && !inplace_not_restore) { + if (count > 1) { + s = backup_engine->RestoreDBFromBackup( + RestoreOptions(), backup_info[thread->rand.Uniform(count)].backup_id, + restore_dir /* db_dir */, restore_dir /* wal_dir */); + if (!s.ok()) { + from = "BackupEngine::RestoreDBFromBackup"; + } + } else { + from_latest = true; + s = backup_engine->RestoreDBFromLatestBackup(RestoreOptions(), + restore_dir /* db_dir */, + restore_dir /* wal_dir */); + if (!s.ok()) { + from = "BackupEngine::RestoreDBFromLatestBackup"; + } + } + } + if (s.ok() && !inplace_not_restore) { + // Purge early if restoring, to ensure the restored directory doesn't + // have some secret dependency on the backup directory. + uint32_t to_keep = 0; + if (allow_persistent) { + // allow one thread to keep up to 2 backups + to_keep = thread->rand.Uniform(3); + } + s = backup_engine->PurgeOldBackups(to_keep); + if (!s.ok()) { + from = "BackupEngine::PurgeOldBackups"; + } + } + DB* restored_db = nullptr; + std::vector<ColumnFamilyHandle*> restored_cf_handles; + // Not yet implemented: opening restored BlobDB or TransactionDB + if (s.ok() && !FLAGS_use_txn && !FLAGS_use_blob_db) { + Options restore_options(options_); + restore_options.best_efforts_recovery = false; + restore_options.listeners.clear(); + // Avoid dangling/shared file descriptors, for reliable destroy + restore_options.sst_file_manager = nullptr; + std::vector<ColumnFamilyDescriptor> cf_descriptors; + // TODO(ajkr): `column_family_names_` is not safe to access here when + // `clear_column_family_one_in != 0`. But we can't easily switch to + // `ListColumnFamilies` to get names because it won't necessarily give + // the same order as `column_family_names_`. + assert(FLAGS_clear_column_family_one_in == 0); + for (auto name : column_family_names_) { + cf_descriptors.emplace_back(name, ColumnFamilyOptions(restore_options)); + } + if (inplace_not_restore) { + BackupInfo& info = backup_info[thread->rand.Uniform(count)]; + restore_options.env = info.env_for_open.get(); + s = DB::OpenForReadOnly(DBOptions(restore_options), info.name_for_open, + cf_descriptors, &restored_cf_handles, + &restored_db); + if (!s.ok()) { + from = "DB::OpenForReadOnly in backup/restore"; + } + } else { + s = DB::Open(DBOptions(restore_options), restore_dir, cf_descriptors, + &restored_cf_handles, &restored_db); + if (!s.ok()) { + from = "DB::Open in backup/restore"; + } + } + } + // Note the column families chosen by `rand_column_families` cannot be + // dropped while the locks for `rand_keys` are held. So we should not have + // to worry about accessing those column families throughout this function. + // + // For simplicity, currently only verifies existence/non-existence of a + // single key + for (size_t i = 0; restored_db && s.ok() && i < rand_column_families.size(); + ++i) { + std::string key_str = Key(rand_keys[0]); + Slice key = key_str; + std::string restored_value; + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions read_opts; + std::string ts_str; + Slice ts; + if (FLAGS_user_timestamp_size > 0) { + ts_str = GetNowNanos(); + ts = ts_str; + read_opts.timestamp = &ts; + } + Status get_status = restored_db->Get( + read_opts, restored_cf_handles[rand_column_families[i]], key, + &restored_value); + bool exists = thread->shared->Exists(rand_column_families[i], rand_keys[0]); + if (get_status.ok()) { + if (!exists && from_latest && ShouldAcquireMutexOnKey()) { + std::ostringstream oss; + oss << "0x" << key.ToString(true) + << " exists in restore but not in original db"; + s = Status::Corruption(oss.str()); + } + } else if (get_status.IsNotFound()) { + if (exists && from_latest && ShouldAcquireMutexOnKey()) { + std::ostringstream oss; + oss << "0x" << key.ToString(true) + << " exists in original db but not in restore"; + s = Status::Corruption(oss.str()); + } + } else { + s = get_status; + if (!s.ok()) { + from = "DB::Get in backup/restore"; + } + } + } + if (restored_db != nullptr) { + for (auto* cf_handle : restored_cf_handles) { + restored_db->DestroyColumnFamilyHandle(cf_handle); + } + delete restored_db; + restored_db = nullptr; + } + if (s.ok() && inplace_not_restore) { + // Purge late if inplace open read-only + uint32_t to_keep = 0; + if (allow_persistent) { + // allow one thread to keep up to 2 backups + to_keep = thread->rand.Uniform(3); + } + s = backup_engine->PurgeOldBackups(to_keep); + if (!s.ok()) { + from = "BackupEngine::PurgeOldBackups"; + } + } + if (backup_engine != nullptr) { + delete backup_engine; + backup_engine = nullptr; + } + if (s.ok()) { + // Preserve directories on failure, or allowed persistent backup + if (!allow_persistent) { + s = DestroyDir(db_stress_env, backup_dir); + if (!s.ok()) { + from = "Destroy backup dir"; + } + } + } + if (s.ok()) { + s = DestroyDir(db_stress_env, restore_dir); + if (!s.ok()) { + from = "Destroy restore dir"; + } + } + if (!s.ok()) { + fprintf(stderr, "Failure in %s with: %s\n", from.c_str(), + s.ToString().c_str()); + } + return s; +} + +Status StressTest::TestApproximateSize( + ThreadState* thread, uint64_t iteration, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) { + // rand_keys likely only has one key. Just use the first one. + assert(!rand_keys.empty()); + assert(!rand_column_families.empty()); + int64_t key1 = rand_keys[0]; + int64_t key2; + if (thread->rand.OneIn(2)) { + // Two totally random keys. This tends to cover large ranges. + key2 = GenerateOneKey(thread, iteration); + if (key2 < key1) { + std::swap(key1, key2); + } + } else { + // Unless users pass a very large FLAGS_max_key, it we should not worry + // about overflow. It is for testing, so we skip the overflow checking + // for simplicity. + key2 = key1 + static_cast<int64_t>(thread->rand.Uniform(1000)); + } + std::string key1_str = Key(key1); + std::string key2_str = Key(key2); + Range range{Slice(key1_str), Slice(key2_str)}; + SizeApproximationOptions sao; + sao.include_memtables = thread->rand.OneIn(2); + if (sao.include_memtables) { + sao.include_files = thread->rand.OneIn(2); + } + if (thread->rand.OneIn(2)) { + if (thread->rand.OneIn(2)) { + sao.files_size_error_margin = 0.0; + } else { + sao.files_size_error_margin = + static_cast<double>(thread->rand.Uniform(3)); + } + } + uint64_t result; + return db_->GetApproximateSizes( + sao, column_families_[rand_column_families[0]], &range, 1, &result); +} + +Status StressTest::TestCheckpoint(ThreadState* thread, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) { + std::vector<std::unique_ptr<MutexLock>> locks; + if (ShouldAcquireMutexOnKey()) { + for (int rand_column_family : rand_column_families) { + // `rand_keys[0]` on each chosen CF will be verified. + locks.emplace_back(new MutexLock( + thread->shared->GetMutexForKey(rand_column_family, rand_keys[0]))); + } + } + + std::string checkpoint_dir = + FLAGS_db + "/.checkpoint" + std::to_string(thread->tid); + Options tmp_opts(options_); + tmp_opts.listeners.clear(); + tmp_opts.env = db_stress_env; + + DestroyDB(checkpoint_dir, tmp_opts); + + if (db_stress_env->FileExists(checkpoint_dir).ok()) { + // If the directory might still exist, try to delete the files one by one. + // Likely a trash file is still there. + Status my_s = DestroyDir(db_stress_env, checkpoint_dir); + if (!my_s.ok()) { + fprintf(stderr, "Fail to destory directory before checkpoint: %s", + my_s.ToString().c_str()); + } + } + + Checkpoint* checkpoint = nullptr; + Status s = Checkpoint::Create(db_, &checkpoint); + if (s.ok()) { + s = checkpoint->CreateCheckpoint(checkpoint_dir); + if (!s.ok()) { + fprintf(stderr, "Fail to create checkpoint to %s\n", + checkpoint_dir.c_str()); + std::vector<std::string> files; + Status my_s = db_stress_env->GetChildren(checkpoint_dir, &files); + if (my_s.ok()) { + for (const auto& f : files) { + fprintf(stderr, " %s\n", f.c_str()); + } + } else { + fprintf(stderr, "Fail to get files under the directory to %s\n", + my_s.ToString().c_str()); + } + } + } + delete checkpoint; + checkpoint = nullptr; + std::vector<ColumnFamilyHandle*> cf_handles; + DB* checkpoint_db = nullptr; + if (s.ok()) { + Options options(options_); + options.best_efforts_recovery = false; + options.listeners.clear(); + // Avoid race condition in trash handling after delete checkpoint_db + options.sst_file_manager.reset(); + std::vector<ColumnFamilyDescriptor> cf_descs; + // TODO(ajkr): `column_family_names_` is not safe to access here when + // `clear_column_family_one_in != 0`. But we can't easily switch to + // `ListColumnFamilies` to get names because it won't necessarily give + // the same order as `column_family_names_`. + assert(FLAGS_clear_column_family_one_in == 0); + if (FLAGS_clear_column_family_one_in == 0) { + for (const auto& name : column_family_names_) { + cf_descs.emplace_back(name, ColumnFamilyOptions(options)); + } + s = DB::OpenForReadOnly(DBOptions(options), checkpoint_dir, cf_descs, + &cf_handles, &checkpoint_db); + } + } + if (checkpoint_db != nullptr) { + // Note the column families chosen by `rand_column_families` cannot be + // dropped while the locks for `rand_keys` are held. So we should not have + // to worry about accessing those column families throughout this function. + for (size_t i = 0; s.ok() && i < rand_column_families.size(); ++i) { + std::string key_str = Key(rand_keys[0]); + Slice key = key_str; + std::string ts_str; + Slice ts; + ReadOptions read_opts; + if (FLAGS_user_timestamp_size > 0) { + ts_str = GetNowNanos(); + ts = ts_str; + read_opts.timestamp = &ts; + } + std::string value; + Status get_status = checkpoint_db->Get( + read_opts, cf_handles[rand_column_families[i]], key, &value); + bool exists = + thread->shared->Exists(rand_column_families[i], rand_keys[0]); + if (get_status.ok()) { + if (!exists && ShouldAcquireMutexOnKey()) { + std::ostringstream oss; + oss << "0x" << key.ToString(true) << " exists in checkpoint " + << checkpoint_dir << " but not in original db"; + s = Status::Corruption(oss.str()); + } + } else if (get_status.IsNotFound()) { + if (exists && ShouldAcquireMutexOnKey()) { + std::ostringstream oss; + oss << "0x" << key.ToString(true) + << " exists in original db but not in checkpoint " + << checkpoint_dir; + s = Status::Corruption(oss.str()); + } + } else { + s = get_status; + } + } + for (auto cfh : cf_handles) { + delete cfh; + } + cf_handles.clear(); + delete checkpoint_db; + checkpoint_db = nullptr; + } + + if (!s.ok()) { + fprintf(stderr, "A checkpoint operation failed with: %s\n", + s.ToString().c_str()); + } else { + DestroyDB(checkpoint_dir, tmp_opts); + } + return s; +} + +void StressTest::TestGetProperty(ThreadState* thread) const { + std::unordered_set<std::string> levelPropertyNames = { + DB::Properties::kAggregatedTablePropertiesAtLevel, + DB::Properties::kCompressionRatioAtLevelPrefix, + DB::Properties::kNumFilesAtLevelPrefix, + }; + std::unordered_set<std::string> unknownPropertyNames = { + DB::Properties::kEstimateOldestKeyTime, + DB::Properties::kOptionsStatistics, + DB::Properties:: + kLiveSstFilesSizeAtTemperature, // similar to levelPropertyNames, it + // requires a number suffix + }; + unknownPropertyNames.insert(levelPropertyNames.begin(), + levelPropertyNames.end()); + + std::unordered_set<std::string> blobCachePropertyNames = { + DB::Properties::kBlobCacheCapacity, + DB::Properties::kBlobCacheUsage, + DB::Properties::kBlobCachePinnedUsage, + }; + if (db_->GetOptions().blob_cache == nullptr) { + unknownPropertyNames.insert(blobCachePropertyNames.begin(), + blobCachePropertyNames.end()); + } + + std::string prop; + for (const auto& ppt_name_and_info : InternalStats::ppt_name_to_info) { + bool res = db_->GetProperty(ppt_name_and_info.first, &prop); + if (unknownPropertyNames.find(ppt_name_and_info.first) == + unknownPropertyNames.end()) { + if (!res) { + fprintf(stderr, "Failed to get DB property: %s\n", + ppt_name_and_info.first.c_str()); + thread->shared->SetVerificationFailure(); + } + if (ppt_name_and_info.second.handle_int != nullptr) { + uint64_t prop_int; + if (!db_->GetIntProperty(ppt_name_and_info.first, &prop_int)) { + fprintf(stderr, "Failed to get Int property: %s\n", + ppt_name_and_info.first.c_str()); + thread->shared->SetVerificationFailure(); + } + } + if (ppt_name_and_info.second.handle_map != nullptr) { + std::map<std::string, std::string> prop_map; + if (!db_->GetMapProperty(ppt_name_and_info.first, &prop_map)) { + fprintf(stderr, "Failed to get Map property: %s\n", + ppt_name_and_info.first.c_str()); + thread->shared->SetVerificationFailure(); + } + } + } + } + + ROCKSDB_NAMESPACE::ColumnFamilyMetaData cf_meta_data; + db_->GetColumnFamilyMetaData(&cf_meta_data); + int level_size = static_cast<int>(cf_meta_data.levels.size()); + for (int level = 0; level < level_size; level++) { + for (const auto& ppt_name : levelPropertyNames) { + bool res = db_->GetProperty(ppt_name + std::to_string(level), &prop); + if (!res) { + fprintf(stderr, "Failed to get DB property: %s\n", + (ppt_name + std::to_string(level)).c_str()); + thread->shared->SetVerificationFailure(); + } + } + } + + // Test for an invalid property name + if (thread->rand.OneIn(100)) { + if (db_->GetProperty("rocksdb.invalid_property_name", &prop)) { + fprintf(stderr, "Failed to return false for invalid property name\n"); + thread->shared->SetVerificationFailure(); + } + } +} + +void StressTest::TestCompactFiles(ThreadState* thread, + ColumnFamilyHandle* column_family) { + ROCKSDB_NAMESPACE::ColumnFamilyMetaData cf_meta_data; + db_->GetColumnFamilyMetaData(column_family, &cf_meta_data); + + if (cf_meta_data.levels.empty()) { + return; + } + + // Randomly compact up to three consecutive files from a level + const int kMaxRetry = 3; + for (int attempt = 0; attempt < kMaxRetry; ++attempt) { + size_t random_level = + thread->rand.Uniform(static_cast<int>(cf_meta_data.levels.size())); + + const auto& files = cf_meta_data.levels[random_level].files; + if (files.size() > 0) { + size_t random_file_index = + thread->rand.Uniform(static_cast<int>(files.size())); + if (files[random_file_index].being_compacted) { + // Retry as the selected file is currently being compacted + continue; + } + + std::vector<std::string> input_files; + input_files.push_back(files[random_file_index].name); + if (random_file_index > 0 && + !files[random_file_index - 1].being_compacted) { + input_files.push_back(files[random_file_index - 1].name); + } + if (random_file_index + 1 < files.size() && + !files[random_file_index + 1].being_compacted) { + input_files.push_back(files[random_file_index + 1].name); + } + + size_t output_level = + std::min(random_level + 1, cf_meta_data.levels.size() - 1); + auto s = db_->CompactFiles(CompactionOptions(), column_family, + input_files, static_cast<int>(output_level)); + if (!s.ok()) { + fprintf(stdout, "Unable to perform CompactFiles(): %s\n", + s.ToString().c_str()); + thread->stats.AddNumCompactFilesFailed(1); + } else { + thread->stats.AddNumCompactFilesSucceed(1); + } + break; + } + } +} +#endif // ROCKSDB_LITE + +Status StressTest::TestFlush(const std::vector<int>& rand_column_families) { + FlushOptions flush_opts; + if (FLAGS_atomic_flush) { + return db_->Flush(flush_opts, column_families_); + } + std::vector<ColumnFamilyHandle*> cfhs; + std::for_each(rand_column_families.begin(), rand_column_families.end(), + [this, &cfhs](int k) { cfhs.push_back(column_families_[k]); }); + return db_->Flush(flush_opts, cfhs); +} + +Status StressTest::TestPauseBackground(ThreadState* thread) { + Status status = db_->PauseBackgroundWork(); + if (!status.ok()) { + return status; + } + // To avoid stalling/deadlocking ourself in this thread, just + // sleep here during pause and let other threads do db operations. + // Sleep up to ~16 seconds (2**24 microseconds), but very skewed + // toward short pause. (1 chance in 25 of pausing >= 1s; + // 1 chance in 625 of pausing full 16s.) + int pwr2_micros = + std::min(thread->rand.Uniform(25), thread->rand.Uniform(25)); + clock_->SleepForMicroseconds(1 << pwr2_micros); + return db_->ContinueBackgroundWork(); +} + +void StressTest::TestAcquireSnapshot(ThreadState* thread, + int rand_column_family, + const std::string& keystr, uint64_t i) { + Slice key = keystr; + ColumnFamilyHandle* column_family = column_families_[rand_column_family]; + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions ropt; +#ifndef ROCKSDB_LITE + auto db_impl = static_cast_with_check<DBImpl>(db_->GetRootDB()); + const bool ww_snapshot = thread->rand.OneIn(10); + const Snapshot* snapshot = + ww_snapshot ? db_impl->GetSnapshotForWriteConflictBoundary() + : db_->GetSnapshot(); +#else + const Snapshot* snapshot = db_->GetSnapshot(); +#endif // !ROCKSDB_LITE + ropt.snapshot = snapshot; + + // Ideally, we want snapshot taking and timestamp generation to be atomic + // here, so that the snapshot corresponds to the timestamp. However, it is + // not possible with current GetSnapshot() API. + std::string ts_str; + Slice ts; + if (FLAGS_user_timestamp_size > 0) { + ts_str = GetNowNanos(); + ts = ts_str; + ropt.timestamp = &ts; + } + + std::string value_at; + // When taking a snapshot, we also read a key from that snapshot. We + // will later read the same key before releasing the snapshot and + // verify that the results are the same. + auto status_at = db_->Get(ropt, column_family, key, &value_at); + std::vector<bool>* key_vec = nullptr; + + if (FLAGS_compare_full_db_state_snapshot && (thread->tid == 0)) { + key_vec = new std::vector<bool>(FLAGS_max_key); + // When `prefix_extractor` is set, seeking to beginning and scanning + // across prefixes are only supported with `total_order_seek` set. + ropt.total_order_seek = true; + std::unique_ptr<Iterator> iterator(db_->NewIterator(ropt)); + for (iterator->SeekToFirst(); iterator->Valid(); iterator->Next()) { + uint64_t key_val; + if (GetIntVal(iterator->key().ToString(), &key_val)) { + (*key_vec)[key_val] = true; + } + } + } + + ThreadState::SnapshotState snap_state = {snapshot, + rand_column_family, + column_family->GetName(), + keystr, + status_at, + value_at, + key_vec, + ts_str}; + uint64_t hold_for = FLAGS_snapshot_hold_ops; + if (FLAGS_long_running_snapshots) { + // Hold 10% of snapshots for 10x more + if (thread->rand.OneIn(10)) { + assert(hold_for < std::numeric_limits<uint64_t>::max() / 10); + hold_for *= 10; + // Hold 1% of snapshots for 100x more + if (thread->rand.OneIn(10)) { + assert(hold_for < std::numeric_limits<uint64_t>::max() / 10); + hold_for *= 10; + } + } + } + uint64_t release_at = std::min(FLAGS_ops_per_thread - 1, i + hold_for); + thread->snapshot_queue.emplace(release_at, snap_state); +} + +Status StressTest::MaybeReleaseSnapshots(ThreadState* thread, uint64_t i) { + while (!thread->snapshot_queue.empty() && + i >= thread->snapshot_queue.front().first) { + auto snap_state = thread->snapshot_queue.front().second; + assert(snap_state.snapshot); + // Note: this is unsafe as the cf might be dropped concurrently. But + // it is ok since unclean cf drop is cunnrently not supported by write + // prepared transactions. + Status s = AssertSame(db_, column_families_[snap_state.cf_at], snap_state); + db_->ReleaseSnapshot(snap_state.snapshot); + delete snap_state.key_vec; + thread->snapshot_queue.pop(); + if (!s.ok()) { + return s; + } + } + return Status::OK(); +} + +void StressTest::TestCompactRange(ThreadState* thread, int64_t rand_key, + const Slice& start_key, + ColumnFamilyHandle* column_family) { + int64_t end_key_num; + if (std::numeric_limits<int64_t>::max() - rand_key < + FLAGS_compact_range_width) { + end_key_num = std::numeric_limits<int64_t>::max(); + } else { + end_key_num = FLAGS_compact_range_width + rand_key; + } + std::string end_key_buf = Key(end_key_num); + Slice end_key(end_key_buf); + + CompactRangeOptions cro; + cro.exclusive_manual_compaction = static_cast<bool>(thread->rand.Next() % 2); + cro.change_level = static_cast<bool>(thread->rand.Next() % 2); + std::vector<BottommostLevelCompaction> bottom_level_styles = { + BottommostLevelCompaction::kSkip, + BottommostLevelCompaction::kIfHaveCompactionFilter, + BottommostLevelCompaction::kForce, + BottommostLevelCompaction::kForceOptimized}; + cro.bottommost_level_compaction = + bottom_level_styles[thread->rand.Next() % + static_cast<uint32_t>(bottom_level_styles.size())]; + cro.allow_write_stall = static_cast<bool>(thread->rand.Next() % 2); + cro.max_subcompactions = static_cast<uint32_t>(thread->rand.Next() % 4); + std::vector<BlobGarbageCollectionPolicy> blob_gc_policies = { + BlobGarbageCollectionPolicy::kForce, + BlobGarbageCollectionPolicy::kDisable, + BlobGarbageCollectionPolicy::kUseDefault}; + cro.blob_garbage_collection_policy = + blob_gc_policies[thread->rand.Next() % + static_cast<uint32_t>(blob_gc_policies.size())]; + cro.blob_garbage_collection_age_cutoff = + static_cast<double>(thread->rand.Next() % 100) / 100.0; + + const Snapshot* pre_snapshot = nullptr; + uint32_t pre_hash = 0; + if (thread->rand.OneIn(2)) { + // Do some validation by declaring a snapshot and compare the data before + // and after the compaction + pre_snapshot = db_->GetSnapshot(); + pre_hash = + GetRangeHash(thread, pre_snapshot, column_family, start_key, end_key); + } + + Status status = db_->CompactRange(cro, column_family, &start_key, &end_key); + + if (!status.ok()) { + fprintf(stdout, "Unable to perform CompactRange(): %s\n", + status.ToString().c_str()); + } + + if (pre_snapshot != nullptr) { + uint32_t post_hash = + GetRangeHash(thread, pre_snapshot, column_family, start_key, end_key); + if (pre_hash != post_hash) { + fprintf(stderr, + "Data hash different before and after compact range " + "start_key %s end_key %s\n", + start_key.ToString(true).c_str(), end_key.ToString(true).c_str()); + thread->stats.AddErrors(1); + // Fail fast to preserve the DB state. + thread->shared->SetVerificationFailure(); + } + db_->ReleaseSnapshot(pre_snapshot); + } +} + +uint32_t StressTest::GetRangeHash(ThreadState* thread, const Snapshot* snapshot, + ColumnFamilyHandle* column_family, + const Slice& start_key, + const Slice& end_key) { + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions ro; + ro.snapshot = snapshot; + ro.total_order_seek = true; + std::string ts_str; + Slice ts; + if (FLAGS_user_timestamp_size > 0) { + ts_str = GetNowNanos(); + ts = ts_str; + ro.timestamp = &ts; + } + + std::unique_ptr<Iterator> it(db_->NewIterator(ro, column_family)); + + constexpr char kCrcCalculatorSepearator = ';'; + + uint32_t crc = 0; + + for (it->Seek(start_key); + it->Valid() && options_.comparator->Compare(it->key(), end_key) <= 0; + it->Next()) { + crc = crc32c::Extend(crc, it->key().data(), it->key().size()); + crc = crc32c::Extend(crc, &kCrcCalculatorSepearator, sizeof(char)); + crc = crc32c::Extend(crc, it->value().data(), it->value().size()); + crc = crc32c::Extend(crc, &kCrcCalculatorSepearator, sizeof(char)); + + for (const auto& column : it->columns()) { + crc = crc32c::Extend(crc, column.name().data(), column.name().size()); + crc = crc32c::Extend(crc, &kCrcCalculatorSepearator, sizeof(char)); + crc = crc32c::Extend(crc, column.value().data(), column.value().size()); + crc = crc32c::Extend(crc, &kCrcCalculatorSepearator, sizeof(char)); + } + } + + if (!it->status().ok()) { + fprintf(stderr, "Iterator non-OK when calculating range CRC: %s\n", + it->status().ToString().c_str()); + thread->stats.AddErrors(1); + // Fail fast to preserve the DB state. + thread->shared->SetVerificationFailure(); + } + + return crc; +} + +void StressTest::PrintEnv() const { + fprintf(stdout, "RocksDB version : %d.%d\n", kMajorVersion, + kMinorVersion); + fprintf(stdout, "Format version : %d\n", FLAGS_format_version); + fprintf(stdout, "TransactionDB : %s\n", + FLAGS_use_txn ? "true" : "false"); + + if (FLAGS_use_txn) { +#ifndef ROCKSDB_LITE + fprintf(stdout, "Two write queues: : %s\n", + FLAGS_two_write_queues ? "true" : "false"); + fprintf(stdout, "Write policy : %d\n", + static_cast<int>(FLAGS_txn_write_policy)); + if (static_cast<uint64_t>(TxnDBWritePolicy::WRITE_PREPARED) == + FLAGS_txn_write_policy || + static_cast<uint64_t>(TxnDBWritePolicy::WRITE_UNPREPARED) == + FLAGS_txn_write_policy) { + fprintf(stdout, "Snapshot cache bits : %d\n", + static_cast<int>(FLAGS_wp_snapshot_cache_bits)); + fprintf(stdout, "Commit cache bits : %d\n", + static_cast<int>(FLAGS_wp_commit_cache_bits)); + } + fprintf(stdout, "last cwb for recovery : %s\n", + FLAGS_use_only_the_last_commit_time_batch_for_recovery ? "true" + : "false"); +#endif // !ROCKSDB_LITE + } + +#ifndef ROCKSDB_LITE + fprintf(stdout, "Stacked BlobDB : %s\n", + FLAGS_use_blob_db ? "true" : "false"); +#endif // !ROCKSDB_LITE + fprintf(stdout, "Read only mode : %s\n", + FLAGS_read_only ? "true" : "false"); + fprintf(stdout, "Atomic flush : %s\n", + FLAGS_atomic_flush ? "true" : "false"); + fprintf(stdout, "Manual WAL flush : %s\n", + FLAGS_manual_wal_flush_one_in > 0 ? "true" : "false"); + fprintf(stdout, "Column families : %d\n", FLAGS_column_families); + if (!FLAGS_test_batches_snapshots) { + fprintf(stdout, "Clear CFs one in : %d\n", + FLAGS_clear_column_family_one_in); + } + fprintf(stdout, "Number of threads : %d\n", FLAGS_threads); + fprintf(stdout, "Ops per thread : %lu\n", + (unsigned long)FLAGS_ops_per_thread); + std::string ttl_state("unused"); + if (FLAGS_ttl > 0) { + ttl_state = std::to_string(FLAGS_ttl); + } + fprintf(stdout, "Time to live(sec) : %s\n", ttl_state.c_str()); + fprintf(stdout, "Read percentage : %d%%\n", FLAGS_readpercent); + fprintf(stdout, "Prefix percentage : %d%%\n", FLAGS_prefixpercent); + fprintf(stdout, "Write percentage : %d%%\n", FLAGS_writepercent); + fprintf(stdout, "Delete percentage : %d%%\n", FLAGS_delpercent); + fprintf(stdout, "Delete range percentage : %d%%\n", FLAGS_delrangepercent); + fprintf(stdout, "No overwrite percentage : %d%%\n", + FLAGS_nooverwritepercent); + fprintf(stdout, "Iterate percentage : %d%%\n", FLAGS_iterpercent); + fprintf(stdout, "Custom ops percentage : %d%%\n", FLAGS_customopspercent); + fprintf(stdout, "DB-write-buffer-size : %" PRIu64 "\n", + FLAGS_db_write_buffer_size); + fprintf(stdout, "Write-buffer-size : %d\n", FLAGS_write_buffer_size); + fprintf(stdout, "Iterations : %lu\n", + (unsigned long)FLAGS_num_iterations); + fprintf(stdout, "Max key : %lu\n", + (unsigned long)FLAGS_max_key); + fprintf(stdout, "Ratio #ops/#keys : %f\n", + (1.0 * FLAGS_ops_per_thread * FLAGS_threads) / FLAGS_max_key); + fprintf(stdout, "Num times DB reopens : %d\n", FLAGS_reopen); + fprintf(stdout, "Batches/snapshots : %d\n", + FLAGS_test_batches_snapshots); + fprintf(stdout, "Do update in place : %d\n", FLAGS_in_place_update); + fprintf(stdout, "Num keys per lock : %d\n", + 1 << FLAGS_log2_keys_per_lock); + std::string compression = CompressionTypeToString(compression_type_e); + fprintf(stdout, "Compression : %s\n", compression.c_str()); + std::string bottommost_compression = + CompressionTypeToString(bottommost_compression_type_e); + fprintf(stdout, "Bottommost Compression : %s\n", + bottommost_compression.c_str()); + std::string checksum = ChecksumTypeToString(checksum_type_e); + fprintf(stdout, "Checksum type : %s\n", checksum.c_str()); + fprintf(stdout, "File checksum impl : %s\n", + FLAGS_file_checksum_impl.c_str()); + fprintf(stdout, "Bloom bits / key : %s\n", + FormatDoubleParam(FLAGS_bloom_bits).c_str()); + fprintf(stdout, "Max subcompactions : %" PRIu64 "\n", + FLAGS_subcompactions); + fprintf(stdout, "Use MultiGet : %s\n", + FLAGS_use_multiget ? "true" : "false"); + + const char* memtablerep = ""; + switch (FLAGS_rep_factory) { + case kSkipList: + memtablerep = "skip_list"; + break; + case kHashSkipList: + memtablerep = "prefix_hash"; + break; + case kVectorRep: + memtablerep = "vector"; + break; + } + + fprintf(stdout, "Memtablerep : %s\n", memtablerep); + +#ifndef NDEBUG + KillPoint* kp = KillPoint::GetInstance(); + fprintf(stdout, "Test kill odd : %d\n", kp->rocksdb_kill_odds); + if (!kp->rocksdb_kill_exclude_prefixes.empty()) { + fprintf(stdout, "Skipping kill points prefixes:\n"); + for (auto& p : kp->rocksdb_kill_exclude_prefixes) { + fprintf(stdout, " %s\n", p.c_str()); + } + } +#endif + fprintf(stdout, "Periodic Compaction Secs : %" PRIu64 "\n", + FLAGS_periodic_compaction_seconds); + fprintf(stdout, "Compaction TTL : %" PRIu64 "\n", + FLAGS_compaction_ttl); + const char* compaction_pri = ""; + switch (FLAGS_compaction_pri) { + case kByCompensatedSize: + compaction_pri = "kByCompensatedSize"; + break; + case kOldestLargestSeqFirst: + compaction_pri = "kOldestLargestSeqFirst"; + break; + case kOldestSmallestSeqFirst: + compaction_pri = "kOldestSmallestSeqFirst"; + break; + case kMinOverlappingRatio: + compaction_pri = "kMinOverlappingRatio"; + break; + case kRoundRobin: + compaction_pri = "kRoundRobin"; + break; + } + fprintf(stdout, "Compaction Pri : %s\n", compaction_pri); + fprintf(stdout, "Background Purge : %d\n", + static_cast<int>(FLAGS_avoid_unnecessary_blocking_io)); + fprintf(stdout, "Write DB ID to manifest : %d\n", + static_cast<int>(FLAGS_write_dbid_to_manifest)); + fprintf(stdout, "Max Write Batch Group Size: %" PRIu64 "\n", + FLAGS_max_write_batch_group_size_bytes); + fprintf(stdout, "Use dynamic level : %d\n", + static_cast<int>(FLAGS_level_compaction_dynamic_level_bytes)); + fprintf(stdout, "Read fault one in : %d\n", FLAGS_read_fault_one_in); + fprintf(stdout, "Write fault one in : %d\n", FLAGS_write_fault_one_in); + fprintf(stdout, "Open metadata write fault one in:\n"); + fprintf(stdout, " %d\n", + FLAGS_open_metadata_write_fault_one_in); + fprintf(stdout, "Sync fault injection : %d\n", + FLAGS_sync_fault_injection); + fprintf(stdout, "Best efforts recovery : %d\n", + static_cast<int>(FLAGS_best_efforts_recovery)); + fprintf(stdout, "Fail if OPTIONS file error: %d\n", + static_cast<int>(FLAGS_fail_if_options_file_error)); + fprintf(stdout, "User timestamp size bytes : %d\n", + static_cast<int>(FLAGS_user_timestamp_size)); + fprintf(stdout, "WAL compression : %s\n", + FLAGS_wal_compression.c_str()); + fprintf(stdout, "Try verify sst unique id : %d\n", + static_cast<int>(FLAGS_verify_sst_unique_id_in_manifest)); + + fprintf(stdout, "------------------------------------------------\n"); +} + +void StressTest::Open(SharedState* shared) { + assert(db_ == nullptr); +#ifndef ROCKSDB_LITE + assert(txn_db_ == nullptr); +#else + (void)shared; +#endif + if (!InitializeOptionsFromFile(options_)) { + InitializeOptionsFromFlags(cache_, compressed_cache_, filter_policy_, + options_); + } + InitializeOptionsGeneral(cache_, compressed_cache_, filter_policy_, options_); + + if (FLAGS_prefix_size == 0 && FLAGS_rep_factory == kHashSkipList) { + fprintf(stderr, + "prefeix_size cannot be zero if memtablerep == prefix_hash\n"); + exit(1); + } + if (FLAGS_prefix_size != 0 && FLAGS_rep_factory != kHashSkipList) { + fprintf(stderr, + "WARNING: prefix_size is non-zero but " + "memtablerep != prefix_hash\n"); + } + + if ((options_.enable_blob_files || options_.enable_blob_garbage_collection || + FLAGS_allow_setting_blob_options_dynamically) && + FLAGS_best_efforts_recovery) { + fprintf(stderr, + "Integrated BlobDB is currently incompatible with best-effort " + "recovery\n"); + exit(1); + } + + fprintf(stdout, + "Integrated BlobDB: blob files enabled %d, min blob size %" PRIu64 + ", blob file size %" PRIu64 + ", blob compression type %s, blob GC enabled %d, cutoff %f, force " + "threshold %f, blob compaction readahead size %" PRIu64 + ", blob file starting level %d\n", + options_.enable_blob_files, options_.min_blob_size, + options_.blob_file_size, + CompressionTypeToString(options_.blob_compression_type).c_str(), + options_.enable_blob_garbage_collection, + options_.blob_garbage_collection_age_cutoff, + options_.blob_garbage_collection_force_threshold, + options_.blob_compaction_readahead_size, + options_.blob_file_starting_level); + + if (FLAGS_use_blob_cache) { + fprintf(stdout, + "Integrated BlobDB: blob cache enabled" + ", block and blob caches shared: %d", + FLAGS_use_shared_block_and_blob_cache); + if (!FLAGS_use_shared_block_and_blob_cache) { + fprintf(stdout, + ", blob cache size %" PRIu64 ", blob cache num shard bits: %d", + FLAGS_blob_cache_size, FLAGS_blob_cache_numshardbits); + } + fprintf(stdout, ", blob cache prepopulated: %d\n", + FLAGS_prepopulate_blob_cache); + } else { + fprintf(stdout, "Integrated BlobDB: blob cache disabled\n"); + } + + fprintf(stdout, "DB path: [%s]\n", FLAGS_db.c_str()); + + Status s; + + if (FLAGS_ttl == -1) { + std::vector<std::string> existing_column_families; + s = DB::ListColumnFamilies(DBOptions(options_), FLAGS_db, + &existing_column_families); // ignore errors + if (!s.ok()) { + // DB doesn't exist + assert(existing_column_families.empty()); + assert(column_family_names_.empty()); + column_family_names_.push_back(kDefaultColumnFamilyName); + } else if (column_family_names_.empty()) { + // this is the first call to the function Open() + column_family_names_ = existing_column_families; + } else { + // this is a reopen. just assert that existing column_family_names are + // equivalent to what we remember + auto sorted_cfn = column_family_names_; + std::sort(sorted_cfn.begin(), sorted_cfn.end()); + std::sort(existing_column_families.begin(), + existing_column_families.end()); + if (sorted_cfn != existing_column_families) { + fprintf(stderr, "Expected column families differ from the existing:\n"); + fprintf(stderr, "Expected: {"); + for (auto cf : sorted_cfn) { + fprintf(stderr, "%s ", cf.c_str()); + } + fprintf(stderr, "}\n"); + fprintf(stderr, "Existing: {"); + for (auto cf : existing_column_families) { + fprintf(stderr, "%s ", cf.c_str()); + } + fprintf(stderr, "}\n"); + } + assert(sorted_cfn == existing_column_families); + } + std::vector<ColumnFamilyDescriptor> cf_descriptors; + for (auto name : column_family_names_) { + if (name != kDefaultColumnFamilyName) { + new_column_family_name_ = + std::max(new_column_family_name_.load(), std::stoi(name) + 1); + } + cf_descriptors.emplace_back(name, ColumnFamilyOptions(options_)); + } + while (cf_descriptors.size() < (size_t)FLAGS_column_families) { + std::string name = std::to_string(new_column_family_name_.load()); + new_column_family_name_++; + cf_descriptors.emplace_back(name, ColumnFamilyOptions(options_)); + column_family_names_.push_back(name); + } + + options_.listeners.clear(); +#ifndef ROCKSDB_LITE + options_.listeners.emplace_back(new DbStressListener( + FLAGS_db, options_.db_paths, cf_descriptors, db_stress_listener_env)); +#endif // !ROCKSDB_LITE + RegisterAdditionalListeners(); + + if (!FLAGS_use_txn) { + // Determine whether we need to ingest file metadata write failures + // during DB reopen. If it does, enable it. + // Only ingest metadata error if it is reopening, as initial open + // failure doesn't need to be handled. + // TODO cover transaction DB is not covered in this fault test too. + bool ingest_meta_error = false; + bool ingest_write_error = false; + bool ingest_read_error = false; + if ((FLAGS_open_metadata_write_fault_one_in || + FLAGS_open_write_fault_one_in || FLAGS_open_read_fault_one_in) && + fault_fs_guard + ->FileExists(FLAGS_db + "/CURRENT", IOOptions(), nullptr) + .ok()) { + if (!FLAGS_sync) { + // When DB Stress is not sync mode, we expect all WAL writes to + // WAL is durable. Buffering unsynced writes will cause false + // positive in crash tests. Before we figure out a way to + // solve it, skip WAL from failure injection. + fault_fs_guard->SetSkipDirectWritableTypes({kWalFile}); + } + ingest_meta_error = FLAGS_open_metadata_write_fault_one_in; + ingest_write_error = FLAGS_open_write_fault_one_in; + ingest_read_error = FLAGS_open_read_fault_one_in; + if (ingest_meta_error) { + fault_fs_guard->EnableMetadataWriteErrorInjection(); + fault_fs_guard->SetRandomMetadataWriteError( + FLAGS_open_metadata_write_fault_one_in); + } + if (ingest_write_error) { + fault_fs_guard->SetFilesystemDirectWritable(false); + fault_fs_guard->EnableWriteErrorInjection(); + fault_fs_guard->SetRandomWriteError( + static_cast<uint32_t>(FLAGS_seed), FLAGS_open_write_fault_one_in, + IOStatus::IOError("Injected Open Error"), + /*inject_for_all_file_types=*/true, /*types=*/{}); + } + if (ingest_read_error) { + fault_fs_guard->SetRandomReadError(FLAGS_open_read_fault_one_in); + } + } + while (true) { +#ifndef ROCKSDB_LITE + // StackableDB-based BlobDB + if (FLAGS_use_blob_db) { + blob_db::BlobDBOptions blob_db_options; + blob_db_options.min_blob_size = FLAGS_blob_db_min_blob_size; + blob_db_options.bytes_per_sync = FLAGS_blob_db_bytes_per_sync; + blob_db_options.blob_file_size = FLAGS_blob_db_file_size; + blob_db_options.enable_garbage_collection = FLAGS_blob_db_enable_gc; + blob_db_options.garbage_collection_cutoff = FLAGS_blob_db_gc_cutoff; + + blob_db::BlobDB* blob_db = nullptr; + s = blob_db::BlobDB::Open(options_, blob_db_options, FLAGS_db, + cf_descriptors, &column_families_, + &blob_db); + if (s.ok()) { + db_ = blob_db; + } + } else +#endif // !ROCKSDB_LITE + { + if (db_preload_finished_.load() && FLAGS_read_only) { + s = DB::OpenForReadOnly(DBOptions(options_), FLAGS_db, + cf_descriptors, &column_families_, &db_); + } else { + s = DB::Open(DBOptions(options_), FLAGS_db, cf_descriptors, + &column_families_, &db_); + } + } + + if (ingest_meta_error || ingest_write_error || ingest_read_error) { + fault_fs_guard->SetFilesystemDirectWritable(true); + fault_fs_guard->DisableMetadataWriteErrorInjection(); + fault_fs_guard->DisableWriteErrorInjection(); + fault_fs_guard->SetSkipDirectWritableTypes({}); + fault_fs_guard->SetRandomReadError(0); + if (s.ok()) { + // Ingested errors might happen in background compactions. We + // wait for all compactions to finish to make sure DB is in + // clean state before executing queries. + s = static_cast_with_check<DBImpl>(db_->GetRootDB()) + ->WaitForCompact(true /* wait_unscheduled */); + if (!s.ok()) { + for (auto cf : column_families_) { + delete cf; + } + column_families_.clear(); + delete db_; + db_ = nullptr; + } + } + if (!s.ok()) { + // After failure to opening a DB due to IO error, retry should + // successfully open the DB with correct data if no IO error shows + // up. + ingest_meta_error = false; + ingest_write_error = false; + ingest_read_error = false; + + Random rand(static_cast<uint32_t>(FLAGS_seed)); + if (rand.OneIn(2)) { + fault_fs_guard->DeleteFilesCreatedAfterLastDirSync(IOOptions(), + nullptr); + } + if (rand.OneIn(3)) { + fault_fs_guard->DropUnsyncedFileData(); + } else if (rand.OneIn(2)) { + fault_fs_guard->DropRandomUnsyncedFileData(&rand); + } + continue; + } + } + break; + } + } else { +#ifndef ROCKSDB_LITE + TransactionDBOptions txn_db_options; + assert(FLAGS_txn_write_policy <= TxnDBWritePolicy::WRITE_UNPREPARED); + txn_db_options.write_policy = + static_cast<TxnDBWritePolicy>(FLAGS_txn_write_policy); + if (FLAGS_unordered_write) { + assert(txn_db_options.write_policy == TxnDBWritePolicy::WRITE_PREPARED); + options_.unordered_write = true; + options_.two_write_queues = true; + txn_db_options.skip_concurrency_control = true; + } else { + options_.two_write_queues = FLAGS_two_write_queues; + } + txn_db_options.wp_snapshot_cache_bits = + static_cast<size_t>(FLAGS_wp_snapshot_cache_bits); + txn_db_options.wp_commit_cache_bits = + static_cast<size_t>(FLAGS_wp_commit_cache_bits); + PrepareTxnDbOptions(shared, txn_db_options); + s = TransactionDB::Open(options_, txn_db_options, FLAGS_db, + cf_descriptors, &column_families_, &txn_db_); + if (!s.ok()) { + fprintf(stderr, "Error in opening the TransactionDB [%s]\n", + s.ToString().c_str()); + fflush(stderr); + } + assert(s.ok()); + + // Do not swap the order of the following. + { + db_ = txn_db_; + db_aptr_.store(txn_db_, std::memory_order_release); + } +#endif + } + if (!s.ok()) { + fprintf(stderr, "Error in opening the DB [%s]\n", s.ToString().c_str()); + fflush(stderr); + } + assert(s.ok()); + assert(column_families_.size() == + static_cast<size_t>(FLAGS_column_families)); + + // Secondary instance does not support write-prepared/write-unprepared + // transactions, thus just disable secondary instance if we use + // transaction. + if (s.ok() && FLAGS_test_secondary && !FLAGS_use_txn) { +#ifndef ROCKSDB_LITE + Options tmp_opts; + // TODO(yanqin) support max_open_files != -1 for secondary instance. + tmp_opts.max_open_files = -1; + tmp_opts.env = db_stress_env; + const std::string& secondary_path = FLAGS_secondaries_base; + s = DB::OpenAsSecondary(tmp_opts, FLAGS_db, secondary_path, + cf_descriptors, &cmp_cfhs_, &cmp_db_); + assert(s.ok()); + assert(cmp_cfhs_.size() == static_cast<size_t>(FLAGS_column_families)); +#else + fprintf(stderr, "Secondary is not supported in RocksDBLite\n"); + exit(1); +#endif // !ROCKSDB_LITE + } + } else { +#ifndef ROCKSDB_LITE + DBWithTTL* db_with_ttl; + s = DBWithTTL::Open(options_, FLAGS_db, &db_with_ttl, FLAGS_ttl); + db_ = db_with_ttl; +#else + fprintf(stderr, "TTL is not supported in RocksDBLite\n"); + exit(1); +#endif + } + + if (FLAGS_preserve_unverified_changes) { + // Up until now, no live file should have become obsolete due to these + // options. After `DisableFileDeletions()` we can reenable auto compactions + // since, even if live files become obsolete, they won't be deleted. + assert(options_.avoid_flush_during_recovery); + assert(options_.disable_auto_compactions); + if (s.ok()) { + s = db_->DisableFileDeletions(); + } + if (s.ok()) { + s = db_->EnableAutoCompaction(column_families_); + } + } + + if (!s.ok()) { + fprintf(stderr, "open error: %s\n", s.ToString().c_str()); + exit(1); + } +} + +void StressTest::Reopen(ThreadState* thread) { +#ifndef ROCKSDB_LITE + // BG jobs in WritePrepared must be canceled first because i) they can access + // the db via a callbac ii) they hold on to a snapshot and the upcoming + // ::Close would complain about it. + const bool write_prepared = FLAGS_use_txn && FLAGS_txn_write_policy != 0; + bool bg_canceled __attribute__((unused)) = false; + if (write_prepared || thread->rand.OneIn(2)) { + const bool wait = + write_prepared || static_cast<bool>(thread->rand.OneIn(2)); + CancelAllBackgroundWork(db_, wait); + bg_canceled = wait; + } + assert(!write_prepared || bg_canceled); +#else + (void)thread; +#endif + + for (auto cf : column_families_) { + delete cf; + } + column_families_.clear(); + +#ifndef ROCKSDB_LITE + if (thread->rand.OneIn(2)) { + Status s = db_->Close(); + if (!s.ok()) { + fprintf(stderr, "Non-ok close status: %s\n", s.ToString().c_str()); + fflush(stderr); + } + assert(s.ok()); + } +#endif + delete db_; + db_ = nullptr; +#ifndef ROCKSDB_LITE + txn_db_ = nullptr; +#endif + + num_times_reopened_++; + auto now = clock_->NowMicros(); + fprintf(stdout, "%s Reopening database for the %dth time\n", + clock_->TimeToString(now / 1000000).c_str(), num_times_reopened_); + Open(thread->shared); + + if ((FLAGS_sync_fault_injection || FLAGS_disable_wal || + FLAGS_manual_wal_flush_one_in > 0) && + IsStateTracked()) { + Status s = thread->shared->SaveAtAndAfter(db_); + if (!s.ok()) { + fprintf(stderr, "Error enabling history tracing: %s\n", + s.ToString().c_str()); + exit(1); + } + } +} + +bool StressTest::MaybeUseOlderTimestampForPointLookup(ThreadState* thread, + std::string& ts_str, + Slice& ts_slice, + ReadOptions& read_opts) { + if (FLAGS_user_timestamp_size == 0) { + return false; + } + + assert(thread); + if (!thread->rand.OneInOpt(3)) { + return false; + } + + const SharedState* const shared = thread->shared; + assert(shared); + const uint64_t start_ts = shared->GetStartTimestamp(); + + uint64_t now = db_stress_env->NowNanos(); + + assert(now > start_ts); + uint64_t time_diff = now - start_ts; + uint64_t ts = start_ts + (thread->rand.Next64() % time_diff); + ts_str.clear(); + PutFixed64(&ts_str, ts); + ts_slice = ts_str; + read_opts.timestamp = &ts_slice; + return true; +} + +void StressTest::MaybeUseOlderTimestampForRangeScan(ThreadState* thread, + std::string& ts_str, + Slice& ts_slice, + ReadOptions& read_opts) { + if (FLAGS_user_timestamp_size == 0) { + return; + } + + assert(thread); + if (!thread->rand.OneInOpt(3)) { + return; + } + + const Slice* const saved_ts = read_opts.timestamp; + assert(saved_ts != nullptr); + + const SharedState* const shared = thread->shared; + assert(shared); + const uint64_t start_ts = shared->GetStartTimestamp(); + + uint64_t now = db_stress_env->NowNanos(); + + assert(now > start_ts); + uint64_t time_diff = now - start_ts; + uint64_t ts = start_ts + (thread->rand.Next64() % time_diff); + ts_str.clear(); + PutFixed64(&ts_str, ts); + ts_slice = ts_str; + read_opts.timestamp = &ts_slice; + + // TODO (yanqin): support Merge with iter_start_ts + if (!thread->rand.OneInOpt(3) || FLAGS_use_merge || FLAGS_use_full_merge_v1) { + return; + } + + ts_str.clear(); + PutFixed64(&ts_str, start_ts); + ts_slice = ts_str; + read_opts.iter_start_ts = &ts_slice; + read_opts.timestamp = saved_ts; +} + +void CheckAndSetOptionsForUserTimestamp(Options& options) { + assert(FLAGS_user_timestamp_size > 0); + const Comparator* const cmp = test::BytewiseComparatorWithU64TsWrapper(); + assert(cmp); + if (FLAGS_user_timestamp_size != cmp->timestamp_size()) { + fprintf(stderr, + "Only -user_timestamp_size=%d is supported in stress test.\n", + static_cast<int>(cmp->timestamp_size())); + exit(1); + } + if (FLAGS_use_txn) { + fprintf(stderr, "TransactionDB does not support timestamp yet.\n"); + exit(1); + } +#ifndef ROCKSDB_LITE + if (FLAGS_enable_blob_files || FLAGS_use_blob_db) { + fprintf(stderr, "BlobDB not supported with timestamp.\n"); + exit(1); + } +#endif // !ROCKSDB_LITE + if (FLAGS_test_cf_consistency || FLAGS_test_batches_snapshots) { + fprintf(stderr, + "Due to per-key ts-seq ordering constraint, only the (default) " + "non-batched test is supported with timestamp.\n"); + exit(1); + } + if (FLAGS_ingest_external_file_one_in > 0) { + fprintf(stderr, "Bulk loading may not support timestamp yet.\n"); + exit(1); + } + options.comparator = cmp; +} + +bool InitializeOptionsFromFile(Options& options) { +#ifndef ROCKSDB_LITE + DBOptions db_options; + std::vector<ColumnFamilyDescriptor> cf_descriptors; + if (!FLAGS_options_file.empty()) { + Status s = LoadOptionsFromFile(FLAGS_options_file, db_stress_env, + &db_options, &cf_descriptors); + if (!s.ok()) { + fprintf(stderr, "Unable to load options file %s --- %s\n", + FLAGS_options_file.c_str(), s.ToString().c_str()); + exit(1); + } + db_options.env = new DbStressEnvWrapper(db_stress_env); + options = Options(db_options, cf_descriptors[0].options); + return true; + } +#else + (void)options; + fprintf(stderr, "--options_file not supported in lite mode\n"); + exit(1); +#endif //! ROCKSDB_LITE + return false; +} + +void InitializeOptionsFromFlags( + const std::shared_ptr<Cache>& cache, + const std::shared_ptr<Cache>& block_cache_compressed, + const std::shared_ptr<const FilterPolicy>& filter_policy, + Options& options) { + BlockBasedTableOptions block_based_options; + block_based_options.block_cache = cache; + block_based_options.cache_index_and_filter_blocks = + FLAGS_cache_index_and_filter_blocks; + block_based_options.metadata_cache_options.top_level_index_pinning = + static_cast<PinningTier>(FLAGS_top_level_index_pinning); + block_based_options.metadata_cache_options.partition_pinning = + static_cast<PinningTier>(FLAGS_partition_pinning); + block_based_options.metadata_cache_options.unpartitioned_pinning = + static_cast<PinningTier>(FLAGS_unpartitioned_pinning); + block_based_options.block_cache_compressed = block_cache_compressed; + block_based_options.checksum = checksum_type_e; + block_based_options.block_size = FLAGS_block_size; + block_based_options.cache_usage_options.options_overrides.insert( + {CacheEntryRole::kCompressionDictionaryBuildingBuffer, + {/*.charged = */ FLAGS_charge_compression_dictionary_building_buffer + ? CacheEntryRoleOptions::Decision::kEnabled + : CacheEntryRoleOptions::Decision::kDisabled}}); + block_based_options.cache_usage_options.options_overrides.insert( + {CacheEntryRole::kFilterConstruction, + {/*.charged = */ FLAGS_charge_filter_construction + ? CacheEntryRoleOptions::Decision::kEnabled + : CacheEntryRoleOptions::Decision::kDisabled}}); + block_based_options.cache_usage_options.options_overrides.insert( + {CacheEntryRole::kBlockBasedTableReader, + {/*.charged = */ FLAGS_charge_table_reader + ? CacheEntryRoleOptions::Decision::kEnabled + : CacheEntryRoleOptions::Decision::kDisabled}}); + block_based_options.cache_usage_options.options_overrides.insert( + {CacheEntryRole::kFileMetadata, + {/*.charged = */ FLAGS_charge_file_metadata + ? CacheEntryRoleOptions::Decision::kEnabled + : CacheEntryRoleOptions::Decision::kDisabled}}); + block_based_options.cache_usage_options.options_overrides.insert( + {CacheEntryRole::kBlobCache, + {/*.charged = */ FLAGS_charge_blob_cache + ? CacheEntryRoleOptions::Decision::kEnabled + : CacheEntryRoleOptions::Decision::kDisabled}}); + block_based_options.format_version = + static_cast<uint32_t>(FLAGS_format_version); + block_based_options.index_block_restart_interval = + static_cast<int32_t>(FLAGS_index_block_restart_interval); + block_based_options.filter_policy = filter_policy; + block_based_options.partition_filters = FLAGS_partition_filters; + block_based_options.optimize_filters_for_memory = + FLAGS_optimize_filters_for_memory; + block_based_options.detect_filter_construct_corruption = + FLAGS_detect_filter_construct_corruption; + block_based_options.index_type = + static_cast<BlockBasedTableOptions::IndexType>(FLAGS_index_type); + block_based_options.data_block_index_type = + static_cast<BlockBasedTableOptions::DataBlockIndexType>( + FLAGS_data_block_index_type); + block_based_options.prepopulate_block_cache = + static_cast<BlockBasedTableOptions::PrepopulateBlockCache>( + FLAGS_prepopulate_block_cache); + block_based_options.initial_auto_readahead_size = + FLAGS_initial_auto_readahead_size; + block_based_options.max_auto_readahead_size = FLAGS_max_auto_readahead_size; + block_based_options.num_file_reads_for_auto_readahead = + FLAGS_num_file_reads_for_auto_readahead; + options.table_factory.reset(NewBlockBasedTableFactory(block_based_options)); + options.db_write_buffer_size = FLAGS_db_write_buffer_size; + options.write_buffer_size = FLAGS_write_buffer_size; + options.max_write_buffer_number = FLAGS_max_write_buffer_number; + options.min_write_buffer_number_to_merge = + FLAGS_min_write_buffer_number_to_merge; + options.max_write_buffer_number_to_maintain = + FLAGS_max_write_buffer_number_to_maintain; + options.max_write_buffer_size_to_maintain = + FLAGS_max_write_buffer_size_to_maintain; + options.memtable_prefix_bloom_size_ratio = + FLAGS_memtable_prefix_bloom_size_ratio; + options.memtable_whole_key_filtering = FLAGS_memtable_whole_key_filtering; + options.disable_auto_compactions = FLAGS_disable_auto_compactions; + options.max_background_compactions = FLAGS_max_background_compactions; + options.max_background_flushes = FLAGS_max_background_flushes; + options.compaction_style = + static_cast<ROCKSDB_NAMESPACE::CompactionStyle>(FLAGS_compaction_style); + options.compaction_pri = + static_cast<ROCKSDB_NAMESPACE::CompactionPri>(FLAGS_compaction_pri); + options.num_levels = FLAGS_num_levels; + if (FLAGS_prefix_size >= 0) { + options.prefix_extractor.reset(NewFixedPrefixTransform(FLAGS_prefix_size)); + } + options.max_open_files = FLAGS_open_files; + options.statistics = dbstats; + options.env = db_stress_env; + options.use_fsync = FLAGS_use_fsync; + options.compaction_readahead_size = FLAGS_compaction_readahead_size; + options.allow_mmap_reads = FLAGS_mmap_read; + options.allow_mmap_writes = FLAGS_mmap_write; + options.use_direct_reads = FLAGS_use_direct_reads; + options.use_direct_io_for_flush_and_compaction = + FLAGS_use_direct_io_for_flush_and_compaction; + options.recycle_log_file_num = + static_cast<size_t>(FLAGS_recycle_log_file_num); + options.target_file_size_base = FLAGS_target_file_size_base; + options.target_file_size_multiplier = FLAGS_target_file_size_multiplier; + options.max_bytes_for_level_base = FLAGS_max_bytes_for_level_base; + options.max_bytes_for_level_multiplier = FLAGS_max_bytes_for_level_multiplier; + options.level0_stop_writes_trigger = FLAGS_level0_stop_writes_trigger; + options.level0_slowdown_writes_trigger = FLAGS_level0_slowdown_writes_trigger; + options.level0_file_num_compaction_trigger = + FLAGS_level0_file_num_compaction_trigger; + options.compression = compression_type_e; + options.bottommost_compression = bottommost_compression_type_e; + options.compression_opts.max_dict_bytes = FLAGS_compression_max_dict_bytes; + options.compression_opts.zstd_max_train_bytes = + FLAGS_compression_zstd_max_train_bytes; + options.compression_opts.parallel_threads = + FLAGS_compression_parallel_threads; + options.compression_opts.max_dict_buffer_bytes = + FLAGS_compression_max_dict_buffer_bytes; + if (ZSTD_FinalizeDictionarySupported()) { + options.compression_opts.use_zstd_dict_trainer = + FLAGS_compression_use_zstd_dict_trainer; + } else if (!FLAGS_compression_use_zstd_dict_trainer) { + fprintf( + stderr, + "WARNING: use_zstd_dict_trainer is false but zstd finalizeDictionary " + "cannot be used because ZSTD 1.4.5+ is not linked with the binary." + " zstd dictionary trainer will be used.\n"); + } + options.max_manifest_file_size = FLAGS_max_manifest_file_size; + options.inplace_update_support = FLAGS_in_place_update; + options.max_subcompactions = static_cast<uint32_t>(FLAGS_subcompactions); + options.allow_concurrent_memtable_write = + FLAGS_allow_concurrent_memtable_write; + options.experimental_mempurge_threshold = + FLAGS_experimental_mempurge_threshold; + options.periodic_compaction_seconds = FLAGS_periodic_compaction_seconds; + options.stats_dump_period_sec = + static_cast<unsigned int>(FLAGS_stats_dump_period_sec); + options.ttl = FLAGS_compaction_ttl; + options.enable_pipelined_write = FLAGS_enable_pipelined_write; + options.enable_write_thread_adaptive_yield = + FLAGS_enable_write_thread_adaptive_yield; + options.compaction_options_universal.size_ratio = FLAGS_universal_size_ratio; + options.compaction_options_universal.min_merge_width = + FLAGS_universal_min_merge_width; + options.compaction_options_universal.max_merge_width = + FLAGS_universal_max_merge_width; + options.compaction_options_universal.max_size_amplification_percent = + FLAGS_universal_max_size_amplification_percent; + options.atomic_flush = FLAGS_atomic_flush; + options.manual_wal_flush = FLAGS_manual_wal_flush_one_in > 0 ? true : false; + options.avoid_unnecessary_blocking_io = FLAGS_avoid_unnecessary_blocking_io; + options.write_dbid_to_manifest = FLAGS_write_dbid_to_manifest; + options.avoid_flush_during_recovery = FLAGS_avoid_flush_during_recovery; + options.max_write_batch_group_size_bytes = + FLAGS_max_write_batch_group_size_bytes; + options.level_compaction_dynamic_level_bytes = + FLAGS_level_compaction_dynamic_level_bytes; + options.track_and_verify_wals_in_manifest = true; + options.verify_sst_unique_id_in_manifest = + FLAGS_verify_sst_unique_id_in_manifest; + options.memtable_protection_bytes_per_key = + FLAGS_memtable_protection_bytes_per_key; + + // Integrated BlobDB + options.enable_blob_files = FLAGS_enable_blob_files; + options.min_blob_size = FLAGS_min_blob_size; + options.blob_file_size = FLAGS_blob_file_size; + options.blob_compression_type = + StringToCompressionType(FLAGS_blob_compression_type.c_str()); + options.enable_blob_garbage_collection = FLAGS_enable_blob_garbage_collection; + options.blob_garbage_collection_age_cutoff = + FLAGS_blob_garbage_collection_age_cutoff; + options.blob_garbage_collection_force_threshold = + FLAGS_blob_garbage_collection_force_threshold; + options.blob_compaction_readahead_size = FLAGS_blob_compaction_readahead_size; + options.blob_file_starting_level = FLAGS_blob_file_starting_level; + + if (FLAGS_use_blob_cache) { + if (FLAGS_use_shared_block_and_blob_cache) { + options.blob_cache = cache; + } else { + if (FLAGS_blob_cache_size > 0) { + LRUCacheOptions co; + co.capacity = FLAGS_blob_cache_size; + co.num_shard_bits = FLAGS_blob_cache_numshardbits; + options.blob_cache = NewLRUCache(co); + } else { + fprintf(stderr, + "Unable to create a standalone blob cache if blob_cache_size " + "<= 0.\n"); + exit(1); + } + } + switch (FLAGS_prepopulate_blob_cache) { + case 0: + options.prepopulate_blob_cache = PrepopulateBlobCache::kDisable; + break; + case 1: + options.prepopulate_blob_cache = PrepopulateBlobCache::kFlushOnly; + break; + default: + fprintf(stderr, "Unknown prepopulate blob cache mode\n"); + exit(1); + } + } + + options.wal_compression = + StringToCompressionType(FLAGS_wal_compression.c_str()); + + if (FLAGS_enable_tiered_storage) { + options.bottommost_temperature = Temperature::kCold; + } + options.preclude_last_level_data_seconds = + FLAGS_preclude_last_level_data_seconds; + options.preserve_internal_time_seconds = FLAGS_preserve_internal_time_seconds; + + switch (FLAGS_rep_factory) { + case kSkipList: + // no need to do anything + break; +#ifndef ROCKSDB_LITE + case kHashSkipList: + options.memtable_factory.reset(NewHashSkipListRepFactory(10000)); + break; + case kVectorRep: + options.memtable_factory.reset(new VectorRepFactory()); + break; +#else + default: + fprintf(stderr, + "RocksdbLite only supports skip list mem table. Skip " + "--rep_factory\n"); +#endif // ROCKSDB_LITE + } + + if (FLAGS_use_full_merge_v1) { + options.merge_operator = MergeOperators::CreateDeprecatedPutOperator(); + } else { + options.merge_operator = MergeOperators::CreatePutOperator(); + } + + if (FLAGS_enable_compaction_filter) { + options.compaction_filter_factory = + std::make_shared<DbStressCompactionFilterFactory>(); + } + + options.best_efforts_recovery = FLAGS_best_efforts_recovery; + options.paranoid_file_checks = FLAGS_paranoid_file_checks; + options.fail_if_options_file_error = FLAGS_fail_if_options_file_error; + + if (FLAGS_user_timestamp_size > 0) { + CheckAndSetOptionsForUserTimestamp(options); + } + + options.allow_data_in_errors = FLAGS_allow_data_in_errors; +} + +void InitializeOptionsGeneral( + const std::shared_ptr<Cache>& cache, + const std::shared_ptr<Cache>& block_cache_compressed, + const std::shared_ptr<const FilterPolicy>& filter_policy, + Options& options) { + options.create_missing_column_families = true; + options.create_if_missing = true; + + if (!options.statistics) { + options.statistics = dbstats; + } + + if (options.env == Options().env) { + options.env = db_stress_env; + } + + assert(options.table_factory); + auto table_options = + options.table_factory->GetOptions<BlockBasedTableOptions>(); + if (table_options) { + if (FLAGS_cache_size > 0) { + table_options->block_cache = cache; + } + if (!table_options->block_cache_compressed && + FLAGS_compressed_cache_size > 0) { + table_options->block_cache_compressed = block_cache_compressed; + } + if (!table_options->filter_policy) { + table_options->filter_policy = filter_policy; + } + } + + // TODO: row_cache, thread-pool IO priority, CPU priority. + + if (!options.rate_limiter) { + if (FLAGS_rate_limiter_bytes_per_sec > 0) { + options.rate_limiter.reset(NewGenericRateLimiter( + FLAGS_rate_limiter_bytes_per_sec, 1000 /* refill_period_us */, + 10 /* fairness */, + FLAGS_rate_limit_bg_reads ? RateLimiter::Mode::kReadsOnly + : RateLimiter::Mode::kWritesOnly)); + } + } + + if (!options.file_checksum_gen_factory) { + options.file_checksum_gen_factory = + GetFileChecksumImpl(FLAGS_file_checksum_impl); + } + + if (FLAGS_sst_file_manager_bytes_per_sec > 0 || + FLAGS_sst_file_manager_bytes_per_truncate > 0) { + Status status; + options.sst_file_manager.reset(NewSstFileManager( + db_stress_env, options.info_log, "" /* trash_dir */, + static_cast<int64_t>(FLAGS_sst_file_manager_bytes_per_sec), + true /* delete_existing_trash */, &status, + 0.25 /* max_trash_db_ratio */, + FLAGS_sst_file_manager_bytes_per_truncate)); + if (!status.ok()) { + fprintf(stderr, "SstFileManager creation failed: %s\n", + status.ToString().c_str()); + exit(1); + } + } + + if (FLAGS_preserve_unverified_changes) { + if (!options.avoid_flush_during_recovery) { + fprintf(stderr, + "WARNING: flipping `avoid_flush_during_recovery` to true for " + "`preserve_unverified_changes` to keep all files\n"); + options.avoid_flush_during_recovery = true; + } + // Together with `avoid_flush_during_recovery == true`, this will prevent + // live files from becoming obsolete and deleted between `DB::Open()` and + // `DisableFileDeletions()` due to flush or compaction. We do not need to + // warn the user since we will reenable compaction soon. + options.disable_auto_compactions = true; + } + + options.table_properties_collector_factories.emplace_back( + std::make_shared<DbStressTablePropertiesCollectorFactory>()); +} + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_test_base.h b/src/rocksdb/db_stress_tool/db_stress_test_base.h new file mode 100644 index 000000000..81fbbe24b --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_test_base.h @@ -0,0 +1,337 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#pragma once + +#include "db_stress_tool/db_stress_common.h" +#include "db_stress_tool/db_stress_shared_state.h" + +namespace ROCKSDB_NAMESPACE { +class SystemClock; +class Transaction; +class TransactionDB; +struct TransactionDBOptions; + +class StressTest { + public: + StressTest(); + + virtual ~StressTest(); + + std::shared_ptr<Cache> NewCache(size_t capacity, int32_t num_shard_bits); + + static std::vector<std::string> GetBlobCompressionTags(); + + bool BuildOptionsTable(); + + void InitDb(SharedState*); + // The initialization work is split into two parts to avoid a circular + // dependency with `SharedState`. + virtual void FinishInitDb(SharedState*); + void TrackExpectedState(SharedState* shared); + void OperateDb(ThreadState* thread); + virtual void VerifyDb(ThreadState* thread) const = 0; + virtual void ContinuouslyVerifyDb(ThreadState* /*thread*/) const = 0; + void PrintStatistics(); + + protected: + Status AssertSame(DB* db, ColumnFamilyHandle* cf, + ThreadState::SnapshotState& snap_state); + + // Currently PreloadDb has to be single-threaded. + void PreloadDbAndReopenAsReadOnly(int64_t number_of_keys, + SharedState* shared); + + Status SetOptions(ThreadState* thread); + +#ifndef ROCKSDB_LITE + // For transactionsDB, there can be txns prepared but not yet committeed + // right before previous stress run crash. + // They will be recovered and processed through + // ProcessRecoveredPreparedTxnsHelper on the start of current stress run. + void ProcessRecoveredPreparedTxns(SharedState* shared); + + // Default implementation will first update ExpectedState to be + // `SharedState::UNKNOWN` for each keys in `txn` and then randomly + // commit or rollback `txn`. + virtual void ProcessRecoveredPreparedTxnsHelper(Transaction* txn, + SharedState* shared); + + Status NewTxn(WriteOptions& write_opts, Transaction** txn); + + Status CommitTxn(Transaction* txn, ThreadState* thread = nullptr); + + Status RollbackTxn(Transaction* txn); +#endif + + virtual void MaybeClearOneColumnFamily(ThreadState* /* thread */) {} + + virtual bool ShouldAcquireMutexOnKey() const { return false; } + + // Returns true if DB state is tracked by the stress test. + virtual bool IsStateTracked() const = 0; + + virtual std::vector<int> GenerateColumnFamilies( + const int /* num_column_families */, int rand_column_family) const { + return {rand_column_family}; + } + + virtual std::vector<int64_t> GenerateKeys(int64_t rand_key) const { + return {rand_key}; + } + + virtual Status TestGet(ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) = 0; + + virtual std::vector<Status> TestMultiGet( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) = 0; + + virtual Status TestPrefixScan(ThreadState* thread, + const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) = 0; + + virtual Status TestPut(ThreadState* thread, WriteOptions& write_opts, + const ReadOptions& read_opts, + const std::vector<int>& cf_ids, + const std::vector<int64_t>& keys, + char (&value)[100]) = 0; + + virtual Status TestDelete(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) = 0; + + virtual Status TestDeleteRange(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) = 0; + + virtual void TestIngestExternalFile( + ThreadState* thread, const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) = 0; + + // Issue compact range, starting with start_key, whose integer value + // is rand_key. + virtual void TestCompactRange(ThreadState* thread, int64_t rand_key, + const Slice& start_key, + ColumnFamilyHandle* column_family); + + // Calculate a hash value for all keys in range [start_key, end_key] + // at a certain snapshot. + uint32_t GetRangeHash(ThreadState* thread, const Snapshot* snapshot, + ColumnFamilyHandle* column_family, + const Slice& start_key, const Slice& end_key); + + // Return a column family handle that mirrors what is pointed by + // `column_family_id`, which will be used to validate data to be correct. + // By default, the column family itself will be returned. + virtual ColumnFamilyHandle* GetControlCfh(ThreadState* /* thread*/, + int column_family_id) { + return column_families_[column_family_id]; + } + +#ifndef ROCKSDB_LITE + // Generated a list of keys that close to boundaries of SST keys. + // If there isn't any SST file in the DB, return empty list. + std::vector<std::string> GetWhiteBoxKeys(ThreadState* thread, DB* db, + ColumnFamilyHandle* cfh, + size_t num_keys); +#else // !ROCKSDB_LITE + std::vector<std::string> GetWhiteBoxKeys(ThreadState*, DB*, + ColumnFamilyHandle*, size_t) { + // Not supported in LITE mode. + return {}; + } +#endif // !ROCKSDB_LITE + + // Given a key K, this creates an iterator which scans to K and then + // does a random sequence of Next/Prev operations. + virtual Status TestIterate(ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys); + + virtual Status TestIterateAgainstExpected( + ThreadState* /* thread */, const ReadOptions& /* read_opts */, + const std::vector<int>& /* rand_column_families */, + const std::vector<int64_t>& /* rand_keys */) { + return Status::NotSupported(); + } + + // Enum used by VerifyIterator() to identify the mode to validate. + enum LastIterateOp { + kLastOpSeek, + kLastOpSeekForPrev, + kLastOpNextOrPrev, + kLastOpSeekToFirst, + kLastOpSeekToLast + }; + + // Compare the two iterator, iter and cmp_iter are in the same position, + // unless iter might be made invalidate or undefined because of + // upper or lower bounds, or prefix extractor. + // Will flag failure if the verification fails. + // diverged = true if the two iterator is already diverged. + // True if verification passed, false if not. + // op_logs is the information to print when validation fails. + void VerifyIterator(ThreadState* thread, ColumnFamilyHandle* cmp_cfh, + const ReadOptions& ro, Iterator* iter, Iterator* cmp_iter, + LastIterateOp op, const Slice& seek_key, + const std::string& op_logs, bool* diverged); + + virtual Status TestBackupRestore(ThreadState* thread, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys); + + virtual Status TestCheckpoint(ThreadState* thread, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys); + + void TestCompactFiles(ThreadState* thread, ColumnFamilyHandle* column_family); + + Status TestFlush(const std::vector<int>& rand_column_families); + + Status TestPauseBackground(ThreadState* thread); + + void TestAcquireSnapshot(ThreadState* thread, int rand_column_family, + const std::string& keystr, uint64_t i); + + Status MaybeReleaseSnapshots(ThreadState* thread, uint64_t i); +#ifndef ROCKSDB_LITE + Status VerifyGetLiveFiles() const; + Status VerifyGetSortedWalFiles() const; + Status VerifyGetCurrentWalFile() const; + void TestGetProperty(ThreadState* thread) const; + + virtual Status TestApproximateSize( + ThreadState* thread, uint64_t iteration, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys); +#endif // !ROCKSDB_LITE + + virtual Status TestCustomOperations( + ThreadState* /*thread*/, + const std::vector<int>& /*rand_column_families*/) { + return Status::NotSupported("TestCustomOperations() must be overridden"); + } + + void VerificationAbort(SharedState* shared, std::string msg, Status s) const; + + void VerificationAbort(SharedState* shared, std::string msg, int cf, + int64_t key) const; + + void VerificationAbort(SharedState* shared, std::string msg, int cf, + int64_t key, Slice value_from_db, + Slice value_from_expected) const; + + void VerificationAbort(SharedState* shared, int cf, int64_t key, + const Slice& value, const WideColumns& columns, + const WideColumns& expected_columns) const; + + static std::string DebugString(const Slice& value, const WideColumns& columns, + const WideColumns& expected_columns); + + void PrintEnv() const; + + void Open(SharedState* shared); + + void Reopen(ThreadState* thread); + + virtual void RegisterAdditionalListeners() {} + +#ifndef ROCKSDB_LITE + virtual void PrepareTxnDbOptions(SharedState* /*shared*/, + TransactionDBOptions& /*txn_db_opts*/) {} +#endif + + // Returns whether the timestamp of read_opts is updated. + bool MaybeUseOlderTimestampForPointLookup(ThreadState* thread, + std::string& ts_str, + Slice& ts_slice, + ReadOptions& read_opts); + + void MaybeUseOlderTimestampForRangeScan(ThreadState* thread, + std::string& ts_str, Slice& ts_slice, + ReadOptions& read_opts); + + std::shared_ptr<Cache> cache_; + std::shared_ptr<Cache> compressed_cache_; + std::shared_ptr<const FilterPolicy> filter_policy_; + DB* db_; +#ifndef ROCKSDB_LITE + TransactionDB* txn_db_; +#endif + + // Currently only used in MultiOpsTxnsStressTest + std::atomic<DB*> db_aptr_; + + Options options_; + SystemClock* clock_; + std::vector<ColumnFamilyHandle*> column_families_; + std::vector<std::string> column_family_names_; + std::atomic<int> new_column_family_name_; + int num_times_reopened_; + std::unordered_map<std::string, std::vector<std::string>> options_table_; + std::vector<std::string> options_index_; + std::atomic<bool> db_preload_finished_; + + // Fields used for continuous verification from another thread + DB* cmp_db_; + std::vector<ColumnFamilyHandle*> cmp_cfhs_; + bool is_db_stopped_; +}; + +// Load options from OPTIONS file and populate `options`. +extern bool InitializeOptionsFromFile(Options& options); + +// Initialize `options` using command line arguments. +// When this function is called, `cache`, `block_cache_compressed`, +// `filter_policy` have all been initialized. Therefore, we just pass them as +// input arguments. +extern void InitializeOptionsFromFlags( + const std::shared_ptr<Cache>& cache, + const std::shared_ptr<Cache>& block_cache_compressed, + const std::shared_ptr<const FilterPolicy>& filter_policy, Options& options); + +// Initialize `options` on which `InitializeOptionsFromFile()` and +// `InitializeOptionsFromFlags()` have both been called already. +// There are two cases. +// Case 1: OPTIONS file is not specified. Command line arguments have been used +// to initialize `options`. InitializeOptionsGeneral() will use +// `cache`, `block_cache_compressed` and `filter_policy` to initialize +// corresponding fields of `options`. InitializeOptionsGeneral() will +// also set up other fields of `options` so that stress test can run. +// Examples include `create_if_missing` and +// `create_missing_column_families`, etc. +// Case 2: OPTIONS file is specified. It is possible that, after loading from +// the given OPTIONS files, some shared object fields are still not +// initialized because they are not set in the OPTIONS file. In this +// case, if command line arguments indicate that the user wants to set +// up such shared objects, e.g. block cache, compressed block cache, +// row cache, filter policy, then InitializeOptionsGeneral() will honor +// the user's choice, thus passing `cache`, `block_cache_compressed`, +// `filter_policy` as input arguments. +// +// InitializeOptionsGeneral() must not overwrite fields of `options` loaded +// from OPTIONS file. +extern void InitializeOptionsGeneral( + const std::shared_ptr<Cache>& cache, + const std::shared_ptr<Cache>& block_cache_compressed, + const std::shared_ptr<const FilterPolicy>& filter_policy, Options& options); + +// If no OPTIONS file is specified, set up `options` so that we can test +// user-defined timestamp which requires `-user_timestamp_size=8`. +// This function also checks for known (currently) incompatible features with +// user-defined timestamp. +extern void CheckAndSetOptionsForUserTimestamp(Options& options); + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/db_stress_tool.cc b/src/rocksdb/db_stress_tool/db_stress_tool.cc new file mode 100644 index 000000000..6c5e952db --- /dev/null +++ b/src/rocksdb/db_stress_tool/db_stress_tool.cc @@ -0,0 +1,365 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. +// +// The test uses an array to compare against values written to the database. +// Keys written to the array are in 1:1 correspondence to the actual values in +// the database according to the formula in the function GenerateValue. + +// Space is reserved in the array from 0 to FLAGS_max_key and values are +// randomly written/deleted/read from those positions. During verification we +// compare all the positions in the array. To shorten/elongate the running +// time, you could change the settings: FLAGS_max_key, FLAGS_ops_per_thread, +// (sometimes also FLAGS_threads). +// +// NOTE that if FLAGS_test_batches_snapshots is set, the test will have +// different behavior. See comment of the flag for details. + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" +#include "db_stress_tool/db_stress_driver.h" +#include "rocksdb/convenience.h" +#include "utilities/fault_injection_fs.h" + +namespace ROCKSDB_NAMESPACE { +namespace { +static std::shared_ptr<ROCKSDB_NAMESPACE::Env> env_guard; +static std::shared_ptr<ROCKSDB_NAMESPACE::DbStressEnvWrapper> env_wrapper_guard; +static std::shared_ptr<ROCKSDB_NAMESPACE::DbStressEnvWrapper> + dbsl_env_wrapper_guard; +static std::shared_ptr<CompositeEnvWrapper> fault_env_guard; +} // namespace + +KeyGenContext key_gen_ctx; + +int db_stress_tool(int argc, char** argv) { + SetUsageMessage(std::string("\nUSAGE:\n") + std::string(argv[0]) + + " [OPTIONS]..."); + ParseCommandLineFlags(&argc, &argv, true); + + SanitizeDoubleParam(&FLAGS_bloom_bits); + SanitizeDoubleParam(&FLAGS_memtable_prefix_bloom_size_ratio); + SanitizeDoubleParam(&FLAGS_max_bytes_for_level_multiplier); + +#ifndef NDEBUG + if (FLAGS_mock_direct_io) { + SetupSyncPointsToMockDirectIO(); + } +#endif + if (FLAGS_statistics) { + dbstats = ROCKSDB_NAMESPACE::CreateDBStatistics(); + if (FLAGS_test_secondary) { + dbstats_secondaries = ROCKSDB_NAMESPACE::CreateDBStatistics(); + } + } + compression_type_e = StringToCompressionType(FLAGS_compression_type.c_str()); + bottommost_compression_type_e = + StringToCompressionType(FLAGS_bottommost_compression_type.c_str()); + checksum_type_e = StringToChecksumType(FLAGS_checksum_type.c_str()); + + Env* raw_env; + + int env_opts = !FLAGS_env_uri.empty() + !FLAGS_fs_uri.empty(); + if (env_opts > 1) { + fprintf(stderr, "Error: --env_uri and --fs_uri are mutually exclusive\n"); + exit(1); + } + + Status s = Env::CreateFromUri(ConfigOptions(), FLAGS_env_uri, FLAGS_fs_uri, + &raw_env, &env_guard); + if (!s.ok()) { + fprintf(stderr, "Error Creating Env URI: %s: %s\n", FLAGS_env_uri.c_str(), + s.ToString().c_str()); + exit(1); + } + dbsl_env_wrapper_guard = std::make_shared<DbStressEnvWrapper>(raw_env); + db_stress_listener_env = dbsl_env_wrapper_guard.get(); + + if (FLAGS_read_fault_one_in || FLAGS_sync_fault_injection || + FLAGS_write_fault_one_in || FLAGS_open_metadata_write_fault_one_in || + FLAGS_open_write_fault_one_in || FLAGS_open_read_fault_one_in) { + FaultInjectionTestFS* fs = + new FaultInjectionTestFS(raw_env->GetFileSystem()); + fault_fs_guard.reset(fs); + if (FLAGS_write_fault_one_in) { + fault_fs_guard->SetFilesystemDirectWritable(false); + } else { + fault_fs_guard->SetFilesystemDirectWritable(true); + } + fault_env_guard = + std::make_shared<CompositeEnvWrapper>(raw_env, fault_fs_guard); + raw_env = fault_env_guard.get(); + } + + env_wrapper_guard = std::make_shared<DbStressEnvWrapper>(raw_env); + db_stress_env = env_wrapper_guard.get(); + + if (FLAGS_write_fault_one_in) { + // In the write injection case, we need to use the FS interface and returns + // the IOStatus with different error and flags. Therefore, + // DbStressEnvWrapper cannot be used which will swallow the FS + // implementations. We should directly use the raw_env which is the + // CompositeEnvWrapper of env and fault_fs. + db_stress_env = raw_env; + } + + FLAGS_rep_factory = StringToRepFactory(FLAGS_memtablerep.c_str()); + + // The number of background threads should be at least as much the + // max number of concurrent compactions. + db_stress_env->SetBackgroundThreads(FLAGS_max_background_compactions, + ROCKSDB_NAMESPACE::Env::Priority::LOW); + db_stress_env->SetBackgroundThreads(FLAGS_num_bottom_pri_threads, + ROCKSDB_NAMESPACE::Env::Priority::BOTTOM); + if (FLAGS_prefixpercent > 0 && FLAGS_prefix_size < 0) { + fprintf(stderr, + "Error: prefixpercent is non-zero while prefix_size is " + "not positive!\n"); + exit(1); + } + if (FLAGS_test_batches_snapshots && FLAGS_prefix_size <= 0) { + fprintf(stderr, + "Error: please specify prefix_size for " + "test_batches_snapshots test!\n"); + exit(1); + } + if (FLAGS_memtable_prefix_bloom_size_ratio > 0.0 && FLAGS_prefix_size < 0 && + !FLAGS_memtable_whole_key_filtering) { + fprintf(stderr, + "Error: please specify positive prefix_size or enable whole key " + "filtering in order to use memtable_prefix_bloom_size_ratio\n"); + exit(1); + } + if ((FLAGS_readpercent + FLAGS_prefixpercent + FLAGS_writepercent + + FLAGS_delpercent + FLAGS_delrangepercent + FLAGS_iterpercent + + FLAGS_customopspercent) != 100) { + fprintf( + stderr, + "Error: " + "Read(-readpercent=%d)+Prefix(-prefixpercent=%d)+Write(-writepercent=%" + "d)+Delete(-delpercent=%d)+DeleteRange(-delrangepercent=%d)" + "+Iterate(-iterpercent=%d)+CustomOps(-customopspercent=%d) percents != " + "100!\n", + FLAGS_readpercent, FLAGS_prefixpercent, FLAGS_writepercent, + FLAGS_delpercent, FLAGS_delrangepercent, FLAGS_iterpercent, + FLAGS_customopspercent); + exit(1); + } + if (FLAGS_disable_wal == 1 && FLAGS_reopen > 0) { + fprintf(stderr, "Error: Db cannot reopen safely with disable_wal set!\n"); + exit(1); + } + if ((unsigned)FLAGS_reopen >= FLAGS_ops_per_thread) { + fprintf(stderr, + "Error: #DB-reopens should be < ops_per_thread\n" + "Provided reopens = %d and ops_per_thread = %lu\n", + FLAGS_reopen, (unsigned long)FLAGS_ops_per_thread); + exit(1); + } + if (FLAGS_test_batches_snapshots && FLAGS_delrangepercent > 0) { + fprintf(stderr, + "Error: nonzero delrangepercent unsupported in " + "test_batches_snapshots mode\n"); + exit(1); + } + if (FLAGS_active_width > FLAGS_max_key) { + fprintf(stderr, "Error: active_width can be at most max_key\n"); + exit(1); + } else if (FLAGS_active_width == 0) { + FLAGS_active_width = FLAGS_max_key; + } + if (FLAGS_value_size_mult * kRandomValueMaxFactor > kValueMaxLen) { + fprintf(stderr, "Error: value_size_mult can be at most %d\n", + kValueMaxLen / kRandomValueMaxFactor); + exit(1); + } + if (FLAGS_use_merge && FLAGS_nooverwritepercent == 100) { + fprintf( + stderr, + "Error: nooverwritepercent must not be 100 when using merge operands"); + exit(1); + } + if (FLAGS_ingest_external_file_one_in > 0 && + FLAGS_nooverwritepercent == 100) { + fprintf( + stderr, + "Error: nooverwritepercent must not be 100 when using file ingestion"); + exit(1); + } + if (FLAGS_clear_column_family_one_in > 0 && FLAGS_backup_one_in > 0) { + fprintf(stderr, + "Error: clear_column_family_one_in must be 0 when using backup\n"); + exit(1); + } + if (FLAGS_test_cf_consistency && FLAGS_disable_wal) { + FLAGS_atomic_flush = true; + } + + if (FLAGS_read_only) { + if (FLAGS_writepercent != 0 || FLAGS_delpercent != 0 || + FLAGS_delrangepercent != 0) { + fprintf(stderr, "Error: updates are not supported in read only mode\n"); + exit(1); + } else if (FLAGS_checkpoint_one_in > 0 && + FLAGS_clear_column_family_one_in > 0) { + fprintf(stdout, + "Warn: checkpoint won't be validated since column families may " + "be dropped.\n"); + } + } + + // Choose a location for the test database if none given with --db=<path> + if (FLAGS_db.empty()) { + std::string default_db_path; + db_stress_env->GetTestDirectory(&default_db_path); + default_db_path += "/dbstress"; + FLAGS_db = default_db_path; + } + + if ((FLAGS_test_secondary || FLAGS_continuous_verification_interval > 0) && + FLAGS_secondaries_base.empty()) { + std::string default_secondaries_path; + db_stress_env->GetTestDirectory(&default_secondaries_path); + default_secondaries_path += "/dbstress_secondaries"; + s = db_stress_env->CreateDirIfMissing(default_secondaries_path); + if (!s.ok()) { + fprintf(stderr, "Failed to create directory %s: %s\n", + default_secondaries_path.c_str(), s.ToString().c_str()); + exit(1); + } + FLAGS_secondaries_base = default_secondaries_path; + } + + if (FLAGS_best_efforts_recovery && !FLAGS_skip_verifydb && + !FLAGS_disable_wal) { + fprintf(stderr, + "With best-efforts recovery, either skip_verifydb or disable_wal " + "should be set to true.\n"); + exit(1); + } + if (FLAGS_skip_verifydb) { + if (FLAGS_verify_db_one_in > 0) { + fprintf(stderr, + "Must set -verify_db_one_in=0 if skip_verifydb is true.\n"); + exit(1); + } + if (FLAGS_continuous_verification_interval > 0) { + fprintf(stderr, + "Must set -continuous_verification_interval=0 if skip_verifydb " + "is true.\n"); + exit(1); + } + } + if (FLAGS_enable_compaction_filter && + (FLAGS_acquire_snapshot_one_in > 0 || FLAGS_compact_range_one_in > 0 || + FLAGS_iterpercent > 0 || FLAGS_test_batches_snapshots || + FLAGS_test_cf_consistency)) { + fprintf( + stderr, + "Error: acquire_snapshot_one_in, compact_range_one_in, iterpercent, " + "test_batches_snapshots must all be 0 when using compaction filter\n"); + exit(1); + } + if (FLAGS_test_multi_ops_txns) { + CheckAndSetOptionsForMultiOpsTxnStressTest(); + } + + if (FLAGS_create_timestamped_snapshot_one_in > 0) { + if (!FLAGS_use_txn) { + fprintf(stderr, "timestamped snapshot supported only in TransactionDB\n"); + exit(1); + } else if (FLAGS_txn_write_policy != 0) { + fprintf(stderr, + "timestamped snapshot supported only in write-committed\n"); + exit(1); + } + } + + if (FLAGS_preserve_unverified_changes && FLAGS_reopen != 0) { + fprintf(stderr, + "Reopen DB is incompatible with preserving unverified changes\n"); + exit(1); + } + + if (FLAGS_use_txn && FLAGS_sync_fault_injection && + FLAGS_txn_write_policy != 0) { + fprintf(stderr, + "For TransactionDB, correctness testing with unsync data loss is " + "currently compatible with only write committed policy\n"); + exit(1); + } + + if (FLAGS_use_put_entity_one_in > 0 && + (FLAGS_ingest_external_file_one_in > 0 || FLAGS_use_merge || + FLAGS_use_full_merge_v1 || FLAGS_use_txn || FLAGS_test_multi_ops_txns || + FLAGS_user_timestamp_size > 0)) { + fprintf(stderr, + "PutEntity is currently incompatible with SstFileWriter, Merge," + " transactions, and user-defined timestamps\n"); + exit(1); + } + +#ifndef NDEBUG + KillPoint* kp = KillPoint::GetInstance(); + kp->rocksdb_kill_odds = FLAGS_kill_random_test; + kp->rocksdb_kill_exclude_prefixes = SplitString(FLAGS_kill_exclude_prefixes); +#endif + + unsigned int levels = FLAGS_max_key_len; + std::vector<std::string> weights; + uint64_t scale_factor = FLAGS_key_window_scale_factor; + key_gen_ctx.window = scale_factor * 100; + if (!FLAGS_key_len_percent_dist.empty()) { + weights = SplitString(FLAGS_key_len_percent_dist); + if (weights.size() != levels) { + fprintf(stderr, + "Number of weights in key_len_dist should be equal to" + " max_key_len"); + exit(1); + } + + uint64_t total_weight = 0; + for (std::string& weight : weights) { + uint64_t val = std::stoull(weight); + key_gen_ctx.weights.emplace_back(val * scale_factor); + total_weight += val; + } + if (total_weight != 100) { + fprintf(stderr, "Sum of all weights in key_len_dist should be 100"); + exit(1); + } + } else { + uint64_t keys_per_level = key_gen_ctx.window / levels; + for (unsigned int level = 0; level + 1 < levels; ++level) { + key_gen_ctx.weights.emplace_back(keys_per_level); + } + key_gen_ctx.weights.emplace_back(key_gen_ctx.window - + keys_per_level * (levels - 1)); + } + + std::unique_ptr<ROCKSDB_NAMESPACE::StressTest> stress; + if (FLAGS_test_cf_consistency) { + stress.reset(CreateCfConsistencyStressTest()); + } else if (FLAGS_test_batches_snapshots) { + stress.reset(CreateBatchedOpsStressTest()); + } else if (FLAGS_test_multi_ops_txns) { + stress.reset(CreateMultiOpsTxnsStressTest()); + } else { + stress.reset(CreateNonBatchedOpsStressTest()); + } + // Initialize the Zipfian pre-calculated array + InitializeHotKeyGenerator(FLAGS_hot_key_alpha); + if (RunStressTest(stress.get())) { + return 0; + } else { + return 1; + } +} + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/expected_state.cc b/src/rocksdb/db_stress_tool/expected_state.cc new file mode 100644 index 000000000..d08403b76 --- /dev/null +++ b/src/rocksdb/db_stress_tool/expected_state.cc @@ -0,0 +1,761 @@ +// Copyright (c) 2021-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). + +#ifdef GFLAGS + +#include "db_stress_tool/expected_state.h" + +#include "db/wide/wide_column_serialization.h" +#include "db_stress_tool/db_stress_common.h" +#include "db_stress_tool/db_stress_shared_state.h" +#include "rocksdb/trace_reader_writer.h" +#include "rocksdb/trace_record_result.h" + +namespace ROCKSDB_NAMESPACE { + +ExpectedState::ExpectedState(size_t max_key, size_t num_column_families) + : max_key_(max_key), + num_column_families_(num_column_families), + values_(nullptr) {} + +void ExpectedState::ClearColumnFamily(int cf) { + std::fill(&Value(cf, 0 /* key */), &Value(cf + 1, 0 /* key */), + SharedState::DELETION_SENTINEL); +} + +void ExpectedState::Put(int cf, int64_t key, uint32_t value_base, + bool pending) { + if (!pending) { + // prevent expected-value update from reordering before Write + std::atomic_thread_fence(std::memory_order_release); + } + Value(cf, key).store(pending ? SharedState::UNKNOWN_SENTINEL : value_base, + std::memory_order_relaxed); + if (pending) { + // prevent Write from reordering before expected-value update + std::atomic_thread_fence(std::memory_order_release); + } +} + +uint32_t ExpectedState::Get(int cf, int64_t key) const { + return Value(cf, key); +} + +bool ExpectedState::Delete(int cf, int64_t key, bool pending) { + if (Value(cf, key) == SharedState::DELETION_SENTINEL) { + return false; + } + Put(cf, key, SharedState::DELETION_SENTINEL, pending); + return true; +} + +bool ExpectedState::SingleDelete(int cf, int64_t key, bool pending) { + return Delete(cf, key, pending); +} + +int ExpectedState::DeleteRange(int cf, int64_t begin_key, int64_t end_key, + bool pending) { + int covered = 0; + for (int64_t key = begin_key; key < end_key; ++key) { + if (Delete(cf, key, pending)) { + ++covered; + } + } + return covered; +} + +bool ExpectedState::Exists(int cf, int64_t key) { + // UNKNOWN_SENTINEL counts as exists. That assures a key for which overwrite + // is disallowed can't be accidentally added a second time, in which case + // SingleDelete wouldn't be able to properly delete the key. It does allow + // the case where a SingleDelete might be added which covers nothing, but + // that's not a correctness issue. + uint32_t expected_value = Value(cf, key).load(); + return expected_value != SharedState::DELETION_SENTINEL; +} + +void ExpectedState::Reset() { + for (size_t i = 0; i < num_column_families_; ++i) { + for (size_t j = 0; j < max_key_; ++j) { + Value(static_cast<int>(i), j) + .store(SharedState::DELETION_SENTINEL, std::memory_order_relaxed); + } + } +} + +FileExpectedState::FileExpectedState(std::string expected_state_file_path, + size_t max_key, size_t num_column_families) + : ExpectedState(max_key, num_column_families), + expected_state_file_path_(expected_state_file_path) {} + +Status FileExpectedState::Open(bool create) { + size_t expected_values_size = GetValuesLen(); + + Env* default_env = Env::Default(); + + Status status; + if (create) { + std::unique_ptr<WritableFile> wfile; + const EnvOptions soptions; + status = default_env->NewWritableFile(expected_state_file_path_, &wfile, + soptions); + if (status.ok()) { + std::string buf(expected_values_size, '\0'); + status = wfile->Append(buf); + } + } + if (status.ok()) { + status = default_env->NewMemoryMappedFileBuffer( + expected_state_file_path_, &expected_state_mmap_buffer_); + } + if (status.ok()) { + assert(expected_state_mmap_buffer_->GetLen() == expected_values_size); + values_ = static_cast<std::atomic<uint32_t>*>( + expected_state_mmap_buffer_->GetBase()); + assert(values_ != nullptr); + if (create) { + Reset(); + } + } else { + assert(values_ == nullptr); + } + return status; +} + +AnonExpectedState::AnonExpectedState(size_t max_key, size_t num_column_families) + : ExpectedState(max_key, num_column_families) {} + +#ifndef NDEBUG +Status AnonExpectedState::Open(bool create) { +#else +Status AnonExpectedState::Open(bool /* create */) { +#endif + // AnonExpectedState only supports being freshly created. + assert(create); + values_allocation_.reset( + new std::atomic<uint32_t>[GetValuesLen() / + sizeof(std::atomic<uint32_t>)]); + values_ = &values_allocation_[0]; + Reset(); + return Status::OK(); +} + +ExpectedStateManager::ExpectedStateManager(size_t max_key, + size_t num_column_families) + : max_key_(max_key), + num_column_families_(num_column_families), + latest_(nullptr) {} + +ExpectedStateManager::~ExpectedStateManager() {} + +const std::string FileExpectedStateManager::kLatestBasename = "LATEST"; +const std::string FileExpectedStateManager::kStateFilenameSuffix = ".state"; +const std::string FileExpectedStateManager::kTraceFilenameSuffix = ".trace"; +const std::string FileExpectedStateManager::kTempFilenamePrefix = "."; +const std::string FileExpectedStateManager::kTempFilenameSuffix = ".tmp"; + +FileExpectedStateManager::FileExpectedStateManager( + size_t max_key, size_t num_column_families, + std::string expected_state_dir_path) + : ExpectedStateManager(max_key, num_column_families), + expected_state_dir_path_(std::move(expected_state_dir_path)) { + assert(!expected_state_dir_path_.empty()); +} + +Status FileExpectedStateManager::Open() { + // Before doing anything, sync directory state with ours. That is, determine + // `saved_seqno_`, and create any necessary missing files. + std::vector<std::string> expected_state_dir_children; + Status s = Env::Default()->GetChildren(expected_state_dir_path_, + &expected_state_dir_children); + bool found_trace = false; + if (s.ok()) { + for (size_t i = 0; i < expected_state_dir_children.size(); ++i) { + const auto& filename = expected_state_dir_children[i]; + if (filename.size() >= kStateFilenameSuffix.size() && + filename.rfind(kStateFilenameSuffix) == + filename.size() - kStateFilenameSuffix.size() && + filename.rfind(kLatestBasename, 0) == std::string::npos) { + SequenceNumber found_seqno = ParseUint64( + filename.substr(0, filename.size() - kStateFilenameSuffix.size())); + if (saved_seqno_ == kMaxSequenceNumber || found_seqno > saved_seqno_) { + saved_seqno_ = found_seqno; + } + } + } + // Check if crash happened after creating state file but before creating + // trace file. + if (saved_seqno_ != kMaxSequenceNumber) { + std::string saved_seqno_trace_path = GetPathForFilename( + std::to_string(saved_seqno_) + kTraceFilenameSuffix); + Status exists_status = Env::Default()->FileExists(saved_seqno_trace_path); + if (exists_status.ok()) { + found_trace = true; + } else if (exists_status.IsNotFound()) { + found_trace = false; + } else { + s = exists_status; + } + } + } + if (s.ok() && saved_seqno_ != kMaxSequenceNumber && !found_trace) { + // Create an empty trace file so later logic does not need to distinguish + // missing vs. empty trace file. + std::unique_ptr<WritableFile> wfile; + const EnvOptions soptions; + std::string saved_seqno_trace_path = + GetPathForFilename(std::to_string(saved_seqno_) + kTraceFilenameSuffix); + s = Env::Default()->NewWritableFile(saved_seqno_trace_path, &wfile, + soptions); + } + + if (s.ok()) { + s = Clean(); + } + + std::string expected_state_file_path = + GetPathForFilename(kLatestBasename + kStateFilenameSuffix); + bool found = false; + if (s.ok()) { + Status exists_status = Env::Default()->FileExists(expected_state_file_path); + if (exists_status.ok()) { + found = true; + } else if (exists_status.IsNotFound()) { + found = false; + } else { + s = exists_status; + } + } + + if (!found) { + // Initialize the file in a temp path and then rename it. That way, in case + // this process is killed during setup, `Clean()` will take care of removing + // the incomplete expected values file. + std::string temp_expected_state_file_path = + GetTempPathForFilename(kLatestBasename + kStateFilenameSuffix); + FileExpectedState temp_expected_state(temp_expected_state_file_path, + max_key_, num_column_families_); + if (s.ok()) { + s = temp_expected_state.Open(true /* create */); + } + if (s.ok()) { + s = Env::Default()->RenameFile(temp_expected_state_file_path, + expected_state_file_path); + } + } + + if (s.ok()) { + latest_.reset(new FileExpectedState(std::move(expected_state_file_path), + max_key_, num_column_families_)); + s = latest_->Open(false /* create */); + } + return s; +} + +#ifndef ROCKSDB_LITE +Status FileExpectedStateManager::SaveAtAndAfter(DB* db) { + SequenceNumber seqno = db->GetLatestSequenceNumber(); + + std::string state_filename = std::to_string(seqno) + kStateFilenameSuffix; + std::string state_file_temp_path = GetTempPathForFilename(state_filename); + std::string state_file_path = GetPathForFilename(state_filename); + + std::string latest_file_path = + GetPathForFilename(kLatestBasename + kStateFilenameSuffix); + + std::string trace_filename = std::to_string(seqno) + kTraceFilenameSuffix; + std::string trace_file_path = GetPathForFilename(trace_filename); + + // Populate a tempfile and then rename it to atomically create "<seqno>.state" + // with contents from "LATEST.state" + Status s = CopyFile(FileSystem::Default(), latest_file_path, + state_file_temp_path, 0 /* size */, false /* use_fsync */, + nullptr /* io_tracer */, Temperature::kUnknown); + if (s.ok()) { + s = FileSystem::Default()->RenameFile(state_file_temp_path, state_file_path, + IOOptions(), nullptr /* dbg */); + } + SequenceNumber old_saved_seqno = 0; + if (s.ok()) { + old_saved_seqno = saved_seqno_; + saved_seqno_ = seqno; + } + + // If there is a crash now, i.e., after "<seqno>.state" was created but before + // "<seqno>.trace" is created, it will be treated as if "<seqno>.trace" were + // present but empty. + + // Create "<seqno>.trace" directly. It is initially empty so no need for + // tempfile. + std::unique_ptr<TraceWriter> trace_writer; + if (s.ok()) { + EnvOptions soptions; + // Disable buffering so traces will not get stuck in application buffer. + soptions.writable_file_max_buffer_size = 0; + s = NewFileTraceWriter(Env::Default(), soptions, trace_file_path, + &trace_writer); + } + if (s.ok()) { + TraceOptions trace_opts; + trace_opts.filter |= kTraceFilterGet; + trace_opts.filter |= kTraceFilterMultiGet; + trace_opts.filter |= kTraceFilterIteratorSeek; + trace_opts.filter |= kTraceFilterIteratorSeekForPrev; + trace_opts.preserve_write_order = true; + s = db->StartTrace(trace_opts, std::move(trace_writer)); + } + + // Delete old state/trace files. Deletion order does not matter since we only + // delete after successfully saving new files, so old files will never be used + // again, even if we crash. + if (s.ok() && old_saved_seqno != kMaxSequenceNumber && + old_saved_seqno != saved_seqno_) { + s = Env::Default()->DeleteFile(GetPathForFilename( + std::to_string(old_saved_seqno) + kStateFilenameSuffix)); + } + if (s.ok() && old_saved_seqno != kMaxSequenceNumber && + old_saved_seqno != saved_seqno_) { + s = Env::Default()->DeleteFile(GetPathForFilename( + std::to_string(old_saved_seqno) + kTraceFilenameSuffix)); + } + return s; +} +#else // ROCKSDB_LITE +Status FileExpectedStateManager::SaveAtAndAfter(DB* /* db */) { + return Status::NotSupported(); +} +#endif // ROCKSDB_LITE + +bool FileExpectedStateManager::HasHistory() { + return saved_seqno_ != kMaxSequenceNumber; +} + +#ifndef ROCKSDB_LITE + +namespace { + +// An `ExpectedStateTraceRecordHandler` applies a configurable number of +// write operation trace records to the configured expected state. It is used in +// `FileExpectedStateManager::Restore()` to sync the expected state with the +// DB's post-recovery state. +class ExpectedStateTraceRecordHandler : public TraceRecord::Handler, + public WriteBatch::Handler { + public: + ExpectedStateTraceRecordHandler(uint64_t max_write_ops, ExpectedState* state) + : max_write_ops_(max_write_ops), + state_(state), + buffered_writes_(nullptr) {} + + ~ExpectedStateTraceRecordHandler() { assert(IsDone()); } + + // True if we have already reached the limit on write operations to apply. + bool IsDone() { return num_write_ops_ == max_write_ops_; } + + Status Handle(const WriteQueryTraceRecord& record, + std::unique_ptr<TraceRecordResult>* /* result */) override { + if (IsDone()) { + return Status::OK(); + } + WriteBatch batch(record.GetWriteBatchRep().ToString()); + return batch.Iterate(this); + } + + // Ignore reads. + Status Handle(const GetQueryTraceRecord& /* record */, + std::unique_ptr<TraceRecordResult>* /* result */) override { + return Status::OK(); + } + + // Ignore reads. + Status Handle(const IteratorSeekQueryTraceRecord& /* record */, + std::unique_ptr<TraceRecordResult>* /* result */) override { + return Status::OK(); + } + + // Ignore reads. + Status Handle(const MultiGetQueryTraceRecord& /* record */, + std::unique_ptr<TraceRecordResult>* /* result */) override { + return Status::OK(); + } + + // Below are the WriteBatch::Handler overrides. We could use a separate + // object, but it's convenient and works to share state with the + // `TraceRecord::Handler`. + + Status PutCF(uint32_t column_family_id, const Slice& key_with_ts, + const Slice& value) override { + Slice key = + StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size); + uint64_t key_id; + if (!GetIntVal(key.ToString(), &key_id)) { + return Status::Corruption("unable to parse key", key.ToString()); + } + uint32_t value_id = GetValueBase(value); + + bool should_buffer_write = !(buffered_writes_ == nullptr); + if (should_buffer_write) { + return WriteBatchInternal::Put(buffered_writes_.get(), column_family_id, + key, value); + } + + state_->Put(column_family_id, static_cast<int64_t>(key_id), value_id, + false /* pending */); + ++num_write_ops_; + return Status::OK(); + } + + Status PutEntityCF(uint32_t column_family_id, const Slice& key_with_ts, + const Slice& entity) override { + Slice key = + StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size); + + uint64_t key_id = 0; + if (!GetIntVal(key.ToString(), &key_id)) { + return Status::Corruption("Unable to parse key", key.ToString()); + } + + Slice entity_copy = entity; + WideColumns columns; + if (!WideColumnSerialization::Deserialize(entity_copy, columns).ok()) { + return Status::Corruption("Unable to deserialize entity", + entity.ToString(/* hex */ true)); + } + + if (columns.empty() || columns[0].name() != kDefaultWideColumnName) { + return Status::Corruption("Cannot find default column in entity", + entity.ToString(/* hex */ true)); + } + + const Slice& value_of_default = columns[0].value(); + + const uint32_t value_base = GetValueBase(value_of_default); + + if (columns != GenerateExpectedWideColumns(value_base, value_of_default)) { + return Status::Corruption("Wide columns in entity inconsistent", + entity.ToString(/* hex */ true)); + } + + if (buffered_writes_) { + return WriteBatchInternal::PutEntity(buffered_writes_.get(), + column_family_id, key, columns); + } + + state_->Put(column_family_id, static_cast<int64_t>(key_id), value_base, + false /* pending */); + + ++num_write_ops_; + + return Status::OK(); + } + + Status DeleteCF(uint32_t column_family_id, + const Slice& key_with_ts) override { + Slice key = + StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size); + uint64_t key_id; + if (!GetIntVal(key.ToString(), &key_id)) { + return Status::Corruption("unable to parse key", key.ToString()); + } + + bool should_buffer_write = !(buffered_writes_ == nullptr); + if (should_buffer_write) { + return WriteBatchInternal::Delete(buffered_writes_.get(), + column_family_id, key); + } + + state_->Delete(column_family_id, static_cast<int64_t>(key_id), + false /* pending */); + ++num_write_ops_; + return Status::OK(); + } + + Status SingleDeleteCF(uint32_t column_family_id, + const Slice& key_with_ts) override { + bool should_buffer_write = !(buffered_writes_ == nullptr); + if (should_buffer_write) { + Slice key = + StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size); + Slice ts = + ExtractTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size); + std::array<Slice, 2> key_with_ts_arr{{key, ts}}; + return WriteBatchInternal::SingleDelete( + buffered_writes_.get(), column_family_id, + SliceParts(key_with_ts_arr.data(), 2)); + } + + return DeleteCF(column_family_id, key_with_ts); + } + + Status DeleteRangeCF(uint32_t column_family_id, + const Slice& begin_key_with_ts, + const Slice& end_key_with_ts) override { + Slice begin_key = + StripTimestampFromUserKey(begin_key_with_ts, FLAGS_user_timestamp_size); + Slice end_key = + StripTimestampFromUserKey(end_key_with_ts, FLAGS_user_timestamp_size); + uint64_t begin_key_id, end_key_id; + if (!GetIntVal(begin_key.ToString(), &begin_key_id)) { + return Status::Corruption("unable to parse begin key", + begin_key.ToString()); + } + if (!GetIntVal(end_key.ToString(), &end_key_id)) { + return Status::Corruption("unable to parse end key", end_key.ToString()); + } + + bool should_buffer_write = !(buffered_writes_ == nullptr); + if (should_buffer_write) { + return WriteBatchInternal::DeleteRange( + buffered_writes_.get(), column_family_id, begin_key, end_key); + } + + state_->DeleteRange(column_family_id, static_cast<int64_t>(begin_key_id), + static_cast<int64_t>(end_key_id), false /* pending */); + ++num_write_ops_; + return Status::OK(); + } + + Status MergeCF(uint32_t column_family_id, const Slice& key_with_ts, + const Slice& value) override { + Slice key = + StripTimestampFromUserKey(key_with_ts, FLAGS_user_timestamp_size); + + bool should_buffer_write = !(buffered_writes_ == nullptr); + if (should_buffer_write) { + return WriteBatchInternal::Merge(buffered_writes_.get(), column_family_id, + key, value); + } + + return PutCF(column_family_id, key, value); + } + + Status MarkBeginPrepare(bool = false) override { + assert(!buffered_writes_); + buffered_writes_.reset(new WriteBatch()); + return Status::OK(); + } + + Status MarkEndPrepare(const Slice& xid) override { + assert(buffered_writes_); + std::string xid_str = xid.ToString(); + assert(xid_to_buffered_writes_.find(xid_str) == + xid_to_buffered_writes_.end()); + + xid_to_buffered_writes_[xid_str].swap(buffered_writes_); + + buffered_writes_.reset(); + + return Status::OK(); + } + + Status MarkCommit(const Slice& xid) override { + std::string xid_str = xid.ToString(); + assert(xid_to_buffered_writes_.find(xid_str) != + xid_to_buffered_writes_.end()); + assert(xid_to_buffered_writes_.at(xid_str)); + + Status s = xid_to_buffered_writes_.at(xid_str)->Iterate(this); + xid_to_buffered_writes_.erase(xid_str); + + return s; + } + + Status MarkRollback(const Slice& xid) override { + std::string xid_str = xid.ToString(); + assert(xid_to_buffered_writes_.find(xid_str) != + xid_to_buffered_writes_.end()); + assert(xid_to_buffered_writes_.at(xid_str)); + xid_to_buffered_writes_.erase(xid_str); + + return Status::OK(); + } + + private: + uint64_t num_write_ops_ = 0; + uint64_t max_write_ops_; + ExpectedState* state_; + std::unordered_map<std::string, std::unique_ptr<WriteBatch>> + xid_to_buffered_writes_; + std::unique_ptr<WriteBatch> buffered_writes_; +}; + +} // anonymous namespace + +Status FileExpectedStateManager::Restore(DB* db) { + assert(HasHistory()); + SequenceNumber seqno = db->GetLatestSequenceNumber(); + if (seqno < saved_seqno_) { + return Status::Corruption("DB is older than any restorable expected state"); + } + + std::string state_filename = + std::to_string(saved_seqno_) + kStateFilenameSuffix; + std::string state_file_path = GetPathForFilename(state_filename); + + std::string latest_file_temp_path = + GetTempPathForFilename(kLatestBasename + kStateFilenameSuffix); + std::string latest_file_path = + GetPathForFilename(kLatestBasename + kStateFilenameSuffix); + + std::string trace_filename = + std::to_string(saved_seqno_) + kTraceFilenameSuffix; + std::string trace_file_path = GetPathForFilename(trace_filename); + + std::unique_ptr<TraceReader> trace_reader; + Status s = NewFileTraceReader(Env::Default(), EnvOptions(), trace_file_path, + &trace_reader); + + if (s.ok()) { + // We are going to replay on top of "`seqno`.state" to create a new + // "LATEST.state". Start off by creating a tempfile so we can later make the + // new "LATEST.state" appear atomically using `RenameFile()`. + s = CopyFile(FileSystem::Default(), state_file_path, latest_file_temp_path, + 0 /* size */, false /* use_fsync */, nullptr /* io_tracer */, + Temperature::kUnknown); + } + + { + std::unique_ptr<Replayer> replayer; + std::unique_ptr<ExpectedState> state; + std::unique_ptr<ExpectedStateTraceRecordHandler> handler; + if (s.ok()) { + state.reset(new FileExpectedState(latest_file_temp_path, max_key_, + num_column_families_)); + s = state->Open(false /* create */); + } + if (s.ok()) { + handler.reset(new ExpectedStateTraceRecordHandler(seqno - saved_seqno_, + state.get())); + // TODO(ajkr): An API limitation requires we provide `handles` although + // they will be unused since we only use the replayer for reading records. + // Just give a default CFH for now to satisfy the requirement. + s = db->NewDefaultReplayer({db->DefaultColumnFamily()} /* handles */, + std::move(trace_reader), &replayer); + } + + if (s.ok()) { + s = replayer->Prepare(); + } + for (;;) { + std::unique_ptr<TraceRecord> record; + s = replayer->Next(&record); + if (!s.ok()) { + break; + } + std::unique_ptr<TraceRecordResult> res; + record->Accept(handler.get(), &res); + } + if (s.IsCorruption() && handler->IsDone()) { + // There could be a corruption reading the tail record of the trace due to + // `db_stress` crashing while writing it. It shouldn't matter as long as + // we already found all the write ops we need to catch up the expected + // state. + s = Status::OK(); + } + if (s.IsIncomplete()) { + // OK because `Status::Incomplete` is expected upon finishing all the + // trace records. + s = Status::OK(); + } + } + + if (s.ok()) { + s = FileSystem::Default()->RenameFile(latest_file_temp_path, + latest_file_path, IOOptions(), + nullptr /* dbg */); + } + if (s.ok()) { + latest_.reset(new FileExpectedState(latest_file_path, max_key_, + num_column_families_)); + s = latest_->Open(false /* create */); + } + + // Delete old state/trace files. We must delete the state file first. + // Otherwise, a crash-recovery immediately after deleting the trace file could + // lead to `Restore()` unable to replay to `seqno`. + if (s.ok()) { + s = Env::Default()->DeleteFile(state_file_path); + } + if (s.ok()) { + saved_seqno_ = kMaxSequenceNumber; + s = Env::Default()->DeleteFile(trace_file_path); + } + return s; +} +#else // ROCKSDB_LITE +Status FileExpectedStateManager::Restore(DB* /* db */) { + return Status::NotSupported(); +} +#endif // ROCKSDB_LITE + +Status FileExpectedStateManager::Clean() { + std::vector<std::string> expected_state_dir_children; + Status s = Env::Default()->GetChildren(expected_state_dir_path_, + &expected_state_dir_children); + // An incomplete `Open()` or incomplete `SaveAtAndAfter()` could have left + // behind invalid temporary files. An incomplete `SaveAtAndAfter()` could have + // also left behind stale state/trace files. An incomplete `Restore()` could + // have left behind stale trace files. + for (size_t i = 0; s.ok() && i < expected_state_dir_children.size(); ++i) { + const auto& filename = expected_state_dir_children[i]; + if (filename.rfind(kTempFilenamePrefix, 0 /* pos */) == 0 && + filename.size() >= kTempFilenameSuffix.size() && + filename.rfind(kTempFilenameSuffix) == + filename.size() - kTempFilenameSuffix.size()) { + // Delete all temp files. + s = Env::Default()->DeleteFile(GetPathForFilename(filename)); + } else if (filename.size() >= kStateFilenameSuffix.size() && + filename.rfind(kStateFilenameSuffix) == + filename.size() - kStateFilenameSuffix.size() && + filename.rfind(kLatestBasename, 0) == std::string::npos && + ParseUint64(filename.substr( + 0, filename.size() - kStateFilenameSuffix.size())) < + saved_seqno_) { + assert(saved_seqno_ != kMaxSequenceNumber); + // Delete stale state files. + s = Env::Default()->DeleteFile(GetPathForFilename(filename)); + } else if (filename.size() >= kTraceFilenameSuffix.size() && + filename.rfind(kTraceFilenameSuffix) == + filename.size() - kTraceFilenameSuffix.size() && + ParseUint64(filename.substr( + 0, filename.size() - kTraceFilenameSuffix.size())) < + saved_seqno_) { + // Delete stale trace files. + s = Env::Default()->DeleteFile(GetPathForFilename(filename)); + } + } + return s; +} + +std::string FileExpectedStateManager::GetTempPathForFilename( + const std::string& filename) { + assert(!expected_state_dir_path_.empty()); + std::string expected_state_dir_path_slash = + expected_state_dir_path_.back() == '/' ? expected_state_dir_path_ + : expected_state_dir_path_ + "/"; + return expected_state_dir_path_slash + kTempFilenamePrefix + filename + + kTempFilenameSuffix; +} + +std::string FileExpectedStateManager::GetPathForFilename( + const std::string& filename) { + assert(!expected_state_dir_path_.empty()); + std::string expected_state_dir_path_slash = + expected_state_dir_path_.back() == '/' ? expected_state_dir_path_ + : expected_state_dir_path_ + "/"; + return expected_state_dir_path_slash + filename; +} + +AnonExpectedStateManager::AnonExpectedStateManager(size_t max_key, + size_t num_column_families) + : ExpectedStateManager(max_key, num_column_families) {} + +Status AnonExpectedStateManager::Open() { + latest_.reset(new AnonExpectedState(max_key_, num_column_families_)); + return latest_->Open(true /* create */); +} + +} // namespace ROCKSDB_NAMESPACE + +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/expected_state.h b/src/rocksdb/db_stress_tool/expected_state.h new file mode 100644 index 000000000..41d747e76 --- /dev/null +++ b/src/rocksdb/db_stress_tool/expected_state.h @@ -0,0 +1,287 @@ +// Copyright (c) 2021-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). + +#ifdef GFLAGS + +#pragma once + +#include <stdint.h> + +#include <atomic> +#include <memory> + +#include "db/dbformat.h" +#include "file/file_util.h" +#include "rocksdb/db.h" +#include "rocksdb/env.h" +#include "rocksdb/file_system.h" +#include "rocksdb/rocksdb_namespace.h" +#include "rocksdb/types.h" +#include "util/string_util.h" + +namespace ROCKSDB_NAMESPACE { + +// An `ExpectedState` provides read/write access to expected values for every +// key. +class ExpectedState { + public: + explicit ExpectedState(size_t max_key, size_t num_column_families); + + virtual ~ExpectedState() {} + + // Requires external locking preventing concurrent execution with any other + // member function. + virtual Status Open(bool create) = 0; + + // Requires external locking covering all keys in `cf`. + void ClearColumnFamily(int cf); + + // @param pending True if the update may have started but is not yet + // guaranteed finished. This is useful for crash-recovery testing when the + // process may crash before updating the expected values array. + // + // Requires external locking covering `key` in `cf`. + void Put(int cf, int64_t key, uint32_t value_base, bool pending); + + // Requires external locking covering `key` in `cf`. + uint32_t Get(int cf, int64_t key) const; + + // @param pending See comment above Put() + // Returns true if the key was not yet deleted. + // + // Requires external locking covering `key` in `cf`. + bool Delete(int cf, int64_t key, bool pending); + + // @param pending See comment above Put() + // Returns true if the key was not yet deleted. + // + // Requires external locking covering `key` in `cf`. + bool SingleDelete(int cf, int64_t key, bool pending); + + // @param pending See comment above Put() + // Returns number of keys deleted by the call. + // + // Requires external locking covering keys in `[begin_key, end_key)` in `cf`. + int DeleteRange(int cf, int64_t begin_key, int64_t end_key, bool pending); + + // Requires external locking covering `key` in `cf`. + bool Exists(int cf, int64_t key); + + private: + // Requires external locking covering `key` in `cf`. + std::atomic<uint32_t>& Value(int cf, int64_t key) const { + return values_[cf * max_key_ + key]; + } + + const size_t max_key_; + const size_t num_column_families_; + + protected: + size_t GetValuesLen() const { + return sizeof(std::atomic<uint32_t>) * num_column_families_ * max_key_; + } + + // Requires external locking preventing concurrent execution with any other + // member function. + void Reset(); + + std::atomic<uint32_t>* values_; +}; + +// A `FileExpectedState` implements `ExpectedState` backed by a file. +class FileExpectedState : public ExpectedState { + public: + explicit FileExpectedState(std::string expected_state_file_path, + size_t max_key, size_t num_column_families); + + // Requires external locking preventing concurrent execution with any other + // member function. + Status Open(bool create) override; + + private: + const std::string expected_state_file_path_; + std::unique_ptr<MemoryMappedFileBuffer> expected_state_mmap_buffer_; +}; + +// An `AnonExpectedState` implements `ExpectedState` backed by a memory +// allocation. +class AnonExpectedState : public ExpectedState { + public: + explicit AnonExpectedState(size_t max_key, size_t num_column_families); + + // Requires external locking preventing concurrent execution with any other + // member function. + Status Open(bool create) override; + + private: + std::unique_ptr<std::atomic<uint32_t>[]> values_allocation_; +}; + +// An `ExpectedStateManager` manages data about the expected state of the +// database. It exposes operations for reading and modifying the latest +// expected state. +class ExpectedStateManager { + public: + explicit ExpectedStateManager(size_t max_key, size_t num_column_families); + + virtual ~ExpectedStateManager(); + + // Requires external locking preventing concurrent execution with any other + // member function. + virtual Status Open() = 0; + + // Saves expected values for the current state of `db` and begins tracking + // changes. Following a successful `SaveAtAndAfter()`, `Restore()` can be + // called on the same DB, as long as its state does not roll back to before + // its current state. + // + // Requires external locking preventing concurrent execution with any other + // member function. Furthermore, `db` must not be mutated while this function + // is executing. + virtual Status SaveAtAndAfter(DB* db) = 0; + + // Returns true if at least one state of historical expected values can be + // restored. + // + // Requires external locking preventing concurrent execution with any other + // member function. + virtual bool HasHistory() = 0; + + // Restores expected values according to the current state of `db`. See + // `SaveAtAndAfter()` for conditions where this can be called. + // + // Requires external locking preventing concurrent execution with any other + // member function. Furthermore, `db` must not be mutated while this function + // is executing. + virtual Status Restore(DB* db) = 0; + + // Requires external locking covering all keys in `cf`. + void ClearColumnFamily(int cf) { return latest_->ClearColumnFamily(cf); } + + // @param pending True if the update may have started but is not yet + // guaranteed finished. This is useful for crash-recovery testing when the + // process may crash before updating the expected values array. + // + // Requires external locking covering `key` in `cf`. + void Put(int cf, int64_t key, uint32_t value_base, bool pending) { + return latest_->Put(cf, key, value_base, pending); + } + + // Requires external locking covering `key` in `cf`. + uint32_t Get(int cf, int64_t key) const { return latest_->Get(cf, key); } + + // @param pending See comment above Put() + // Returns true if the key was not yet deleted. + // + // Requires external locking covering `key` in `cf`. + bool Delete(int cf, int64_t key, bool pending) { + return latest_->Delete(cf, key, pending); + } + + // @param pending See comment above Put() + // Returns true if the key was not yet deleted. + // + // Requires external locking covering `key` in `cf`. + bool SingleDelete(int cf, int64_t key, bool pending) { + return latest_->SingleDelete(cf, key, pending); + } + + // @param pending See comment above Put() + // Returns number of keys deleted by the call. + // + // Requires external locking covering keys in `[begin_key, end_key)` in `cf`. + int DeleteRange(int cf, int64_t begin_key, int64_t end_key, bool pending) { + return latest_->DeleteRange(cf, begin_key, end_key, pending); + } + + // Requires external locking covering `key` in `cf`. + bool Exists(int cf, int64_t key) { return latest_->Exists(cf, key); } + + protected: + const size_t max_key_; + const size_t num_column_families_; + std::unique_ptr<ExpectedState> latest_; +}; + +// A `FileExpectedStateManager` implements an `ExpectedStateManager` backed by +// a directory of files containing data about the expected state of the +// database. +class FileExpectedStateManager : public ExpectedStateManager { + public: + explicit FileExpectedStateManager(size_t max_key, size_t num_column_families, + std::string expected_state_dir_path); + + // Requires external locking preventing concurrent execution with any other + // member function. + Status Open() override; + + // See `ExpectedStateManager::SaveAtAndAfter()` API doc. + // + // This implementation makes a copy of "LATEST.state" into + // "<current seqno>.state", and starts a trace in "<current seqno>.trace". + // Due to using external files, a following `Restore()` can happen even + // from a different process. + Status SaveAtAndAfter(DB* db) override; + + // See `ExpectedStateManager::HasHistory()` API doc. + bool HasHistory() override; + + // See `ExpectedStateManager::Restore()` API doc. + // + // Say `db->GetLatestSequenceNumber()` was `a` last time `SaveAtAndAfter()` + // was called and now it is `b`. Then this function replays `b - a` write + // operations from "`a`.trace" onto "`a`.state", and then copies the resulting + // file into "LATEST.state". + Status Restore(DB* db) override; + + private: + // Requires external locking preventing concurrent execution with any other + // member function. + Status Clean(); + + std::string GetTempPathForFilename(const std::string& filename); + std::string GetPathForFilename(const std::string& filename); + + static const std::string kLatestBasename; + static const std::string kStateFilenameSuffix; + static const std::string kTraceFilenameSuffix; + static const std::string kTempFilenamePrefix; + static const std::string kTempFilenameSuffix; + + const std::string expected_state_dir_path_; + SequenceNumber saved_seqno_ = kMaxSequenceNumber; +}; + +// An `AnonExpectedStateManager` implements an `ExpectedStateManager` backed by +// a memory allocation containing data about the expected state of the database. +class AnonExpectedStateManager : public ExpectedStateManager { + public: + explicit AnonExpectedStateManager(size_t max_key, size_t num_column_families); + + // See `ExpectedStateManager::SaveAtAndAfter()` API doc. + // + // This implementation returns `Status::NotSupported` since we do not + // currently have a need to keep history of expected state within a process. + Status SaveAtAndAfter(DB* /* db */) override { + return Status::NotSupported(); + } + + // See `ExpectedStateManager::HasHistory()` API doc. + bool HasHistory() override { return false; } + + // See `ExpectedStateManager::Restore()` API doc. + // + // This implementation returns `Status::NotSupported` since we do not + // currently have a need to keep history of expected state within a process. + Status Restore(DB* /* db */) override { return Status::NotSupported(); } + + // Requires external locking preventing concurrent execution with any other + // member function. + Status Open() override; +}; + +} // namespace ROCKSDB_NAMESPACE + +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/multi_ops_txns_stress.cc b/src/rocksdb/db_stress_tool/multi_ops_txns_stress.cc new file mode 100644 index 000000000..7db5e8942 --- /dev/null +++ b/src/rocksdb/db_stress_tool/multi_ops_txns_stress.cc @@ -0,0 +1,1808 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#include "db_stress_tool/multi_ops_txns_stress.h" + +#include "rocksdb/utilities/write_batch_with_index.h" +#include "util/defer.h" +#include "utilities/fault_injection_fs.h" +#include "utilities/transactions/write_prepared_txn_db.h" + +namespace ROCKSDB_NAMESPACE { + +// The description of A and C can be found in multi_ops_txns_stress.h +DEFINE_int32(lb_a, 0, "(Inclusive) lower bound of A"); +DEFINE_int32(ub_a, 1000, "(Exclusive) upper bound of A"); +DEFINE_int32(lb_c, 0, "(Inclusive) lower bound of C"); +DEFINE_int32(ub_c, 1000, "(Exclusive) upper bound of C"); + +DEFINE_string(key_spaces_path, "", + "Path to file describing the lower and upper bounds of A and C"); + +DEFINE_int32(delay_snapshot_read_one_in, 0, + "With a chance of 1/N, inject a random delay between taking " + "snapshot and read."); + +DEFINE_int32(rollback_one_in, 0, + "If non-zero, rollback non-read-only transactions with a " + "probability of 1/N."); + +DEFINE_int32(clear_wp_commit_cache_one_in, 0, + "If non-zero, evict all commit entries from commit cache with a " + "probability of 1/N. This options applies to write-prepared and " + "write-unprepared transactions."); + +extern "C" bool rocksdb_write_prepared_TEST_ShouldClearCommitCache(void) { + static Random rand(static_cast<uint32_t>(db_stress_env->NowMicros())); + return FLAGS_clear_wp_commit_cache_one_in > 0 && + rand.OneIn(FLAGS_clear_wp_commit_cache_one_in); +} + +// MultiOpsTxnsStressTest can either operate on a database with pre-populated +// data (possibly from previous ones), or create a new db and preload it with +// data specified via `-lb_a`, `-ub_a`, `-lb_c`, `-ub_c`, etc. Among these, we +// define the test key spaces as two key ranges: [lb_a, ub_a) and [lb_c, ub_c). +// The key spaces specification is persisted in a file whose absolute path can +// be specified via `-key_spaces_path`. +// +// Whether an existing db is used or a new one is created, key_spaces_path will +// be used. In the former case, the test reads the key spaces specification +// from `-key_spaces_path` and decodes [lb_a, ub_a) and [lb_c, ub_c). In the +// latter case, the test writes a key spaces specification to a file at the +// location, and this file will be used by future runs until a new db is +// created. +// +// Create a fresh new database (-destroy_db_initially=1 or there is no database +// in the location specified by -db). See PreloadDb(). +// +// Use an existing, non-empty database. See ScanExistingDb(). +// +// This test is multi-threaded, and thread count can be specified via +// `-threads`. For simplicity, we partition the key ranges and each thread +// operates on a subrange independently. +// Within each subrange, a KeyGenerator object is responsible for key +// generation. A KeyGenerator maintains two sets: set of existing keys within +// [low, high), set of non-existing keys within [low, high). [low, high) is the +// subrange. The test initialization makes sure there is at least one +// non-existing key, otherwise the test will return an error and exit before +// any test thread is spawned. + +void MultiOpsTxnsStressTest::KeyGenerator::FinishInit() { + assert(existing_.empty()); + assert(!existing_uniq_.empty()); + assert(low_ < high_); + for (auto v : existing_uniq_) { + assert(low_ <= v); + assert(high_ > v); + existing_.push_back(v); + } + if (non_existing_uniq_.empty()) { + fprintf( + stderr, + "Cannot allocate key in [%u, %u)\nStart with a new DB or try change " + "the number of threads for testing via -threads=<#threads>\n", + static_cast<unsigned int>(low_), static_cast<unsigned int>(high_)); + fflush(stdout); + fflush(stderr); + assert(false); + } + initialized_ = true; +} + +std::pair<uint32_t, uint32_t> +MultiOpsTxnsStressTest::KeyGenerator::ChooseExisting() { + assert(initialized_); + const size_t N = existing_.size(); + assert(N > 0); + uint32_t rnd = rand_.Uniform(static_cast<int>(N)); + assert(rnd < N); + return std::make_pair(existing_[rnd], rnd); +} + +uint32_t MultiOpsTxnsStressTest::KeyGenerator::Allocate() { + assert(initialized_); + auto it = non_existing_uniq_.begin(); + assert(non_existing_uniq_.end() != it); + uint32_t ret = *it; + // Remove this element from non_existing_. + // Need to call UndoAllocation() if the calling transaction does not commit. + non_existing_uniq_.erase(it); + return ret; +} + +void MultiOpsTxnsStressTest::KeyGenerator::Replace(uint32_t old_val, + uint32_t old_pos, + uint32_t new_val) { + assert(initialized_); + { + auto it = existing_uniq_.find(old_val); + assert(it != existing_uniq_.end()); + existing_uniq_.erase(it); + } + + { + assert(0 == existing_uniq_.count(new_val)); + existing_uniq_.insert(new_val); + existing_[old_pos] = new_val; + } + + { + assert(0 == non_existing_uniq_.count(old_val)); + non_existing_uniq_.insert(old_val); + } +} + +void MultiOpsTxnsStressTest::KeyGenerator::UndoAllocation(uint32_t new_val) { + assert(initialized_); + assert(0 == non_existing_uniq_.count(new_val)); + non_existing_uniq_.insert(new_val); +} + +std::string MultiOpsTxnsStressTest::Record::EncodePrimaryKey(uint32_t a) { + std::string ret; + PutFixed32(&ret, kPrimaryIndexId); + PutFixed32(&ret, a); + + char* const buf = &ret[0]; + std::reverse(buf, buf + sizeof(kPrimaryIndexId)); + std::reverse(buf + sizeof(kPrimaryIndexId), + buf + sizeof(kPrimaryIndexId) + sizeof(a)); + return ret; +} + +std::string MultiOpsTxnsStressTest::Record::EncodeSecondaryKey(uint32_t c) { + std::string ret; + PutFixed32(&ret, kSecondaryIndexId); + PutFixed32(&ret, c); + + char* const buf = &ret[0]; + std::reverse(buf, buf + sizeof(kSecondaryIndexId)); + std::reverse(buf + sizeof(kSecondaryIndexId), + buf + sizeof(kSecondaryIndexId) + sizeof(c)); + return ret; +} + +std::string MultiOpsTxnsStressTest::Record::EncodeSecondaryKey(uint32_t c, + uint32_t a) { + std::string ret; + PutFixed32(&ret, kSecondaryIndexId); + PutFixed32(&ret, c); + PutFixed32(&ret, a); + + char* const buf = &ret[0]; + std::reverse(buf, buf + sizeof(kSecondaryIndexId)); + std::reverse(buf + sizeof(kSecondaryIndexId), + buf + sizeof(kSecondaryIndexId) + sizeof(c)); + std::reverse(buf + sizeof(kSecondaryIndexId) + sizeof(c), + buf + sizeof(kSecondaryIndexId) + sizeof(c) + sizeof(a)); + return ret; +} + +std::tuple<Status, uint32_t, uint32_t> +MultiOpsTxnsStressTest::Record::DecodePrimaryIndexValue( + Slice primary_index_value) { + if (primary_index_value.size() != 8) { + return std::tuple<Status, uint32_t, uint32_t>{Status::Corruption(""), 0, 0}; + } + uint32_t b = 0; + uint32_t c = 0; + if (!GetFixed32(&primary_index_value, &b) || + !GetFixed32(&primary_index_value, &c)) { + assert(false); + return std::tuple<Status, uint32_t, uint32_t>{Status::Corruption(""), 0, 0}; + } + return std::tuple<Status, uint32_t, uint32_t>{Status::OK(), b, c}; +} + +std::pair<Status, uint32_t> +MultiOpsTxnsStressTest::Record::DecodeSecondaryIndexValue( + Slice secondary_index_value) { + if (secondary_index_value.size() != 4) { + return std::make_pair(Status::Corruption(""), 0); + } + uint32_t crc = 0; + bool result __attribute__((unused)) = + GetFixed32(&secondary_index_value, &crc); + assert(result); + return std::make_pair(Status::OK(), crc); +} + +std::pair<std::string, std::string> +MultiOpsTxnsStressTest::Record::EncodePrimaryIndexEntry() const { + std::string primary_index_key = EncodePrimaryKey(); + std::string primary_index_value = EncodePrimaryIndexValue(); + return std::make_pair(primary_index_key, primary_index_value); +} + +std::string MultiOpsTxnsStressTest::Record::EncodePrimaryKey() const { + return EncodePrimaryKey(a_); +} + +std::string MultiOpsTxnsStressTest::Record::EncodePrimaryIndexValue() const { + std::string ret; + PutFixed32(&ret, b_); + PutFixed32(&ret, c_); + return ret; +} + +std::pair<std::string, std::string> +MultiOpsTxnsStressTest::Record::EncodeSecondaryIndexEntry() const { + std::string secondary_index_key = EncodeSecondaryKey(c_, a_); + + // Secondary index value is always 4-byte crc32 of the secondary key + std::string secondary_index_value; + uint32_t crc = + crc32c::Value(secondary_index_key.data(), secondary_index_key.size()); + PutFixed32(&secondary_index_value, crc); + return std::make_pair(std::move(secondary_index_key), secondary_index_value); +} + +std::string MultiOpsTxnsStressTest::Record::EncodeSecondaryKey() const { + return EncodeSecondaryKey(c_, a_); +} + +Status MultiOpsTxnsStressTest::Record::DecodePrimaryIndexEntry( + Slice primary_index_key, Slice primary_index_value) { + if (primary_index_key.size() != 8) { + assert(false); + return Status::Corruption("Primary index key length is not 8"); + } + + uint32_t index_id = 0; + + [[maybe_unused]] bool res = GetFixed32(&primary_index_key, &index_id); + assert(res); + index_id = EndianSwapValue(index_id); + + if (index_id != kPrimaryIndexId) { + std::ostringstream oss; + oss << "Unexpected primary index id: " << index_id; + return Status::Corruption(oss.str()); + } + + res = GetFixed32(&primary_index_key, &a_); + assert(res); + a_ = EndianSwapValue(a_); + assert(primary_index_key.empty()); + + if (primary_index_value.size() != 8) { + return Status::Corruption("Primary index value length is not 8"); + } + GetFixed32(&primary_index_value, &b_); + GetFixed32(&primary_index_value, &c_); + return Status::OK(); +} + +Status MultiOpsTxnsStressTest::Record::DecodeSecondaryIndexEntry( + Slice secondary_index_key, Slice secondary_index_value) { + if (secondary_index_key.size() != 12) { + return Status::Corruption("Secondary index key length is not 12"); + } + uint32_t crc = + crc32c::Value(secondary_index_key.data(), secondary_index_key.size()); + + uint32_t index_id = 0; + + [[maybe_unused]] bool res = GetFixed32(&secondary_index_key, &index_id); + assert(res); + index_id = EndianSwapValue(index_id); + + if (index_id != kSecondaryIndexId) { + std::ostringstream oss; + oss << "Unexpected secondary index id: " << index_id; + return Status::Corruption(oss.str()); + } + + assert(secondary_index_key.size() == 8); + res = GetFixed32(&secondary_index_key, &c_); + assert(res); + c_ = EndianSwapValue(c_); + + assert(secondary_index_key.size() == 4); + res = GetFixed32(&secondary_index_key, &a_); + assert(res); + a_ = EndianSwapValue(a_); + assert(secondary_index_key.empty()); + + if (secondary_index_value.size() != 4) { + return Status::Corruption("Secondary index value length is not 4"); + } + uint32_t val = 0; + GetFixed32(&secondary_index_value, &val); + if (val != crc) { + std::ostringstream oss; + oss << "Secondary index key checksum mismatch, stored: " << val + << ", recomputed: " << crc; + return Status::Corruption(oss.str()); + } + return Status::OK(); +} + +void MultiOpsTxnsStressTest::FinishInitDb(SharedState* shared) { + if (FLAGS_enable_compaction_filter) { + // TODO (yanqin) enable compaction filter + } +#ifndef ROCKSDB_LITE + ProcessRecoveredPreparedTxns(shared); +#endif + + ReopenAndPreloadDbIfNeeded(shared); + // TODO (yanqin) parallelize if key space is large + for (auto& key_gen : key_gen_for_a_) { + assert(key_gen); + key_gen->FinishInit(); + } + // TODO (yanqin) parallelize if key space is large + for (auto& key_gen : key_gen_for_c_) { + assert(key_gen); + key_gen->FinishInit(); + } +} + +void MultiOpsTxnsStressTest::ReopenAndPreloadDbIfNeeded(SharedState* shared) { + (void)shared; +#ifndef ROCKSDB_LITE + bool db_empty = false; + { + std::unique_ptr<Iterator> iter(db_->NewIterator(ReadOptions())); + iter->SeekToFirst(); + if (!iter->Valid()) { + db_empty = true; + } + } + + if (db_empty) { + PreloadDb(shared, FLAGS_threads, FLAGS_lb_a, FLAGS_ub_a, FLAGS_lb_c, + FLAGS_ub_c); + } else { + fprintf(stdout, + "Key ranges will be read from %s.\n-lb_a, -ub_a, -lb_c, -ub_c will " + "be ignored\n", + FLAGS_key_spaces_path.c_str()); + fflush(stdout); + ScanExistingDb(shared, FLAGS_threads); + } +#endif // !ROCKSDB_LITE +} + +// Used for point-lookup transaction +Status MultiOpsTxnsStressTest::TestGet( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& /*rand_column_families*/, + const std::vector<int64_t>& /*rand_keys*/) { + uint32_t a = 0; + uint32_t pos = 0; + std::tie(a, pos) = ChooseExistingA(thread); + return PointLookupTxn(thread, read_opts, a); +} + +// Not used. +std::vector<Status> MultiOpsTxnsStressTest::TestMultiGet( + ThreadState* /*thread*/, const ReadOptions& /*read_opts*/, + const std::vector<int>& /*rand_column_families*/, + const std::vector<int64_t>& /*rand_keys*/) { + return std::vector<Status>{Status::NotSupported()}; +} + +Status MultiOpsTxnsStressTest::TestPrefixScan( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) { + (void)thread; + (void)read_opts; + (void)rand_column_families; + (void)rand_keys; + return Status::OK(); +} + +// Given a key K, this creates an iterator which scans to K and then +// does a random sequence of Next/Prev operations. +Status MultiOpsTxnsStressTest::TestIterate( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& /*rand_column_families*/, + const std::vector<int64_t>& /*rand_keys*/) { + uint32_t c = 0; + uint32_t pos = 0; + std::tie(c, pos) = ChooseExistingC(thread); + return RangeScanTxn(thread, read_opts, c); +} + +// Not intended for use. +Status MultiOpsTxnsStressTest::TestPut(ThreadState* /*thread*/, + WriteOptions& /*write_opts*/, + const ReadOptions& /*read_opts*/, + const std::vector<int>& /*cf_ids*/, + const std::vector<int64_t>& /*keys*/, + char (&value)[100]) { + (void)value; + return Status::NotSupported(); +} + +// Not intended for use. +Status MultiOpsTxnsStressTest::TestDelete( + ThreadState* /*thread*/, WriteOptions& /*write_opts*/, + const std::vector<int>& /*rand_column_families*/, + const std::vector<int64_t>& /*rand_keys*/) { + return Status::NotSupported(); +} + +// Not intended for use. +Status MultiOpsTxnsStressTest::TestDeleteRange( + ThreadState* /*thread*/, WriteOptions& /*write_opts*/, + const std::vector<int>& /*rand_column_families*/, + const std::vector<int64_t>& /*rand_keys*/) { + return Status::NotSupported(); +} + +void MultiOpsTxnsStressTest::TestIngestExternalFile( + ThreadState* thread, const std::vector<int>& rand_column_families, + const std::vector<int64_t>& /*rand_keys*/) { + // TODO (yanqin) + (void)thread; + (void)rand_column_families; +} + +void MultiOpsTxnsStressTest::TestCompactRange( + ThreadState* thread, int64_t /*rand_key*/, const Slice& /*start_key*/, + ColumnFamilyHandle* column_family) { + // TODO (yanqin). + // May use GetRangeHash() for validation before and after DB::CompactRange() + // completes. + (void)thread; + (void)column_family; +} + +Status MultiOpsTxnsStressTest::TestBackupRestore( + ThreadState* thread, const std::vector<int>& rand_column_families, + const std::vector<int64_t>& /*rand_keys*/) { + // TODO (yanqin) + (void)thread; + (void)rand_column_families; + return Status::OK(); +} + +Status MultiOpsTxnsStressTest::TestCheckpoint( + ThreadState* thread, const std::vector<int>& rand_column_families, + const std::vector<int64_t>& /*rand_keys*/) { + // TODO (yanqin) + (void)thread; + (void)rand_column_families; + return Status::OK(); +} + +#ifndef ROCKSDB_LITE +Status MultiOpsTxnsStressTest::TestApproximateSize( + ThreadState* thread, uint64_t iteration, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& /*rand_keys*/) { + // TODO (yanqin) + (void)thread; + (void)iteration; + (void)rand_column_families; + return Status::OK(); +} +#endif // !ROCKSDB_LITE + +Status MultiOpsTxnsStressTest::TestCustomOperations( + ThreadState* thread, const std::vector<int>& rand_column_families) { + (void)rand_column_families; + // Randomly choose from 0, 1, and 2. + // TODO (yanqin) allow user to configure probability of each operation. + uint32_t rand = thread->rand.Uniform(3); + Status s; + if (0 == rand) { + // Update primary key. + uint32_t old_a = 0; + uint32_t pos = 0; + std::tie(old_a, pos) = ChooseExistingA(thread); + uint32_t new_a = GenerateNextA(thread); + s = PrimaryKeyUpdateTxn(thread, old_a, pos, new_a); + } else if (1 == rand) { + // Update secondary key. + uint32_t old_c = 0; + uint32_t pos = 0; + std::tie(old_c, pos) = ChooseExistingC(thread); + uint32_t new_c = GenerateNextC(thread); + s = SecondaryKeyUpdateTxn(thread, old_c, pos, new_c); + } else if (2 == rand) { + // Update primary index value. + uint32_t a = 0; + uint32_t pos = 0; + std::tie(a, pos) = ChooseExistingA(thread); + s = UpdatePrimaryIndexValueTxn(thread, a, /*b_delta=*/1); + } else { + // Should never reach here. + assert(false); + } + + return s; +} + +void MultiOpsTxnsStressTest::RegisterAdditionalListeners() { + options_.listeners.emplace_back(new MultiOpsTxnsStressListener(this)); +} + +#ifndef ROCKSDB_LITE +void MultiOpsTxnsStressTest::PrepareTxnDbOptions( + SharedState* /*shared*/, TransactionDBOptions& txn_db_opts) { + // MultiOpsTxnStressTest uses SingleDelete to delete secondary keys, thus we + // register this callback to let TxnDb know that when rolling back + // a transaction, use only SingleDelete to cancel prior Put from the same + // transaction if applicable. + txn_db_opts.rollback_deletion_type_callback = + [](TransactionDB* /*db*/, ColumnFamilyHandle* /*column_family*/, + const Slice& key) { + Slice ks = key; + uint32_t index_id = 0; + [[maybe_unused]] bool res = GetFixed32(&ks, &index_id); + assert(res); + index_id = EndianSwapValue(index_id); + assert(index_id <= Record::kSecondaryIndexId); + return index_id == Record::kSecondaryIndexId; + }; +} +#endif // !ROCKSDB_LITE + +Status MultiOpsTxnsStressTest::PrimaryKeyUpdateTxn(ThreadState* thread, + uint32_t old_a, + uint32_t old_a_pos, + uint32_t new_a) { +#ifdef ROCKSDB_LITE + (void)thread; + (void)old_a; + (void)old_a_pos; + (void)new_a; + return Status::NotSupported(); +#else + std::string old_pk = Record::EncodePrimaryKey(old_a); + std::string new_pk = Record::EncodePrimaryKey(new_a); + Transaction* txn = nullptr; + WriteOptions wopts; + Status s = NewTxn(wopts, &txn); + if (!s.ok()) { + assert(!txn); + thread->stats.AddErrors(1); + return s; + } + + assert(txn); + txn->SetSnapshotOnNextOperation(/*notifier=*/nullptr); + + const Defer cleanup([new_a, &s, thread, txn, this]() { + if (s.ok()) { + // Two gets, one for existing pk, one for locking potential new pk. + thread->stats.AddGets(/*ngets=*/2, /*nfounds=*/1); + thread->stats.AddDeletes(1); + thread->stats.AddBytesForWrites( + /*nwrites=*/2, + Record::kPrimaryIndexEntrySize + Record::kSecondaryIndexEntrySize); + thread->stats.AddSingleDeletes(1); + return; + } + if (s.IsNotFound()) { + thread->stats.AddGets(/*ngets=*/1, /*nfounds=*/0); + } else if (s.IsBusy() || s.IsIncomplete()) { + // ignore. + // Incomplete also means rollback by application. See the transaction + // implementations. + } else { + thread->stats.AddErrors(1); + } + auto& key_gen = key_gen_for_a_[thread->tid]; + key_gen->UndoAllocation(new_a); + RollbackTxn(txn).PermitUncheckedError(); + }); + + ReadOptions ropts; + ropts.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + std::string value; + s = txn->GetForUpdate(ropts, old_pk, &value); + if (!s.ok()) { + return s; + } + std::string empty_value; + s = txn->GetForUpdate(ropts, new_pk, &empty_value); + if (s.ok()) { + assert(!empty_value.empty()); + s = Status::Busy(); + return s; + } else if (!s.IsNotFound()) { + return s; + } + + auto result = Record::DecodePrimaryIndexValue(value); + s = std::get<0>(result); + if (!s.ok()) { + return s; + } + uint32_t b = std::get<1>(result); + uint32_t c = std::get<2>(result); + + ColumnFamilyHandle* cf = db_->DefaultColumnFamily(); + s = txn->Delete(cf, old_pk, /*assume_tracked=*/true); + if (!s.ok()) { + return s; + } + s = txn->Put(cf, new_pk, value, /*assume_tracked=*/true); + if (!s.ok()) { + return s; + } + + auto* wb = txn->GetWriteBatch(); + assert(wb); + + std::string old_sk = Record::EncodeSecondaryKey(c, old_a); + s = wb->SingleDelete(old_sk); + if (!s.ok()) { + return s; + } + + Record record(new_a, b, c); + std::string new_sk; + std::string new_crc; + std::tie(new_sk, new_crc) = record.EncodeSecondaryIndexEntry(); + s = wb->Put(new_sk, new_crc); + if (!s.ok()) { + return s; + } + + s = txn->Prepare(); + + if (!s.ok()) { + return s; + } + + if (FLAGS_rollback_one_in > 0 && thread->rand.OneIn(FLAGS_rollback_one_in)) { + s = Status::Incomplete(); + return s; + } + + s = WriteToCommitTimeWriteBatch(*txn); + if (!s.ok()) { + return s; + } + + s = CommitAndCreateTimestampedSnapshotIfNeeded(thread, *txn); + + auto& key_gen = key_gen_for_a_.at(thread->tid); + if (s.ok()) { + delete txn; + key_gen->Replace(old_a, old_a_pos, new_a); + } + return s; +#endif // !ROCKSDB_LITE +} + +Status MultiOpsTxnsStressTest::SecondaryKeyUpdateTxn(ThreadState* thread, + uint32_t old_c, + uint32_t old_c_pos, + uint32_t new_c) { +#ifdef ROCKSDB_LITE + (void)thread; + (void)old_c; + (void)old_c_pos; + (void)new_c; + return Status::NotSupported(); +#else + Transaction* txn = nullptr; + WriteOptions wopts; + Status s = NewTxn(wopts, &txn); + if (!s.ok()) { + assert(!txn); + thread->stats.AddErrors(1); + return s; + } + + assert(txn); + + Iterator* it = nullptr; + long iterations = 0; + const Defer cleanup([new_c, &s, thread, &it, txn, this, &iterations]() { + delete it; + if (s.ok()) { + thread->stats.AddIterations(iterations); + thread->stats.AddGets(/*ngets=*/1, /*nfounds=*/1); + thread->stats.AddSingleDeletes(1); + thread->stats.AddBytesForWrites( + /*nwrites=*/2, + Record::kPrimaryIndexEntrySize + Record::kSecondaryIndexEntrySize); + return; + } else if (s.IsBusy() || s.IsTimedOut() || s.IsTryAgain() || + s.IsMergeInProgress() || s.IsIncomplete()) { + // ww-conflict detected, or + // lock cannot be acquired, or + // memtable history is not large enough for conflict checking, or + // Merge operation cannot be resolved, or + // application rollback. + // TODO (yanqin) add stats for other cases? + } else if (s.IsNotFound()) { + // ignore. + } else { + thread->stats.AddErrors(1); + } + auto& key_gen = key_gen_for_c_[thread->tid]; + key_gen->UndoAllocation(new_c); + RollbackTxn(txn).PermitUncheckedError(); + }); + + // TODO (yanqin) try SetSnapshotOnNextOperation(). We currently need to take + // a snapshot here because we will later verify that point lookup in the + // primary index using GetForUpdate() returns the same value for 'c' as the + // iterator. The iterator does not need a snapshot though, because it will be + // assigned the current latest (published) sequence in the db, which will be + // no smaller than the snapshot created here. The GetForUpdate will perform + // ww conflict checking to ensure GetForUpdate() (using the snapshot) sees + // the same data as this iterator. + txn->SetSnapshot(); + std::string old_sk_prefix = Record::EncodeSecondaryKey(old_c); + std::string iter_ub_str = Record::EncodeSecondaryKey(old_c + 1); + Slice iter_ub = iter_ub_str; + ReadOptions ropts; + ropts.snapshot = txn->GetSnapshot(); + ropts.total_order_seek = true; + ropts.iterate_upper_bound = &iter_ub; + ropts.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + it = txn->GetIterator(ropts); + + assert(it); + it->Seek(old_sk_prefix); + if (!it->Valid()) { + s = Status::NotFound(); + return s; + } + auto* wb = txn->GetWriteBatch(); + assert(wb); + + do { + ++iterations; + Record record; + s = record.DecodeSecondaryIndexEntry(it->key(), it->value()); + if (!s.ok()) { + fprintf(stderr, "Cannot decode secondary key (%s => %s): %s\n", + it->key().ToString(true).c_str(), + it->value().ToString(true).c_str(), s.ToString().c_str()); + assert(false); + break; + } + // At this point, record.b is not known yet, thus we need to access + // primary index. + std::string pk = Record::EncodePrimaryKey(record.a_value()); + std::string value; + ReadOptions read_opts; + read_opts.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + read_opts.snapshot = txn->GetSnapshot(); + s = txn->GetForUpdate(read_opts, pk, &value); + if (s.IsBusy() || s.IsTimedOut() || s.IsTryAgain() || + s.IsMergeInProgress()) { + // Write conflict, or cannot acquire lock, or memtable size is not large + // enough, or merge cannot be resolved. + break; + } else if (s.IsNotFound()) { + // We can also fail verification here. + std::ostringstream oss; + auto* dbimpl = static_cast_with_check<DBImpl>(db_->GetRootDB()); + assert(dbimpl); + oss << "snap " << read_opts.snapshot->GetSequenceNumber() + << " (published " << dbimpl->GetLastPublishedSequence() + << "), pk should exist: " << Slice(pk).ToString(true); + fprintf(stderr, "%s\n", oss.str().c_str()); + assert(false); + break; + } + if (!s.ok()) { + std::ostringstream oss; + auto* dbimpl = static_cast_with_check<DBImpl>(db_->GetRootDB()); + assert(dbimpl); + oss << "snap " << read_opts.snapshot->GetSequenceNumber() + << " (published " << dbimpl->GetLastPublishedSequence() << "), " + << s.ToString(); + fprintf(stderr, "%s\n", oss.str().c_str()); + assert(false); + break; + } + auto result = Record::DecodePrimaryIndexValue(value); + s = std::get<0>(result); + if (!s.ok()) { + fprintf(stderr, "Cannot decode primary index value %s: %s\n", + Slice(value).ToString(true).c_str(), s.ToString().c_str()); + assert(false); + break; + } + uint32_t b = std::get<1>(result); + uint32_t c = std::get<2>(result); + if (c != old_c) { + std::ostringstream oss; + auto* dbimpl = static_cast_with_check<DBImpl>(db_->GetRootDB()); + assert(dbimpl); + oss << "snap " << read_opts.snapshot->GetSequenceNumber() + << " (published " << dbimpl->GetLastPublishedSequence() + << "), pk/sk mismatch. pk: (a=" << record.a_value() << ", " + << "c=" << c << "), sk: (c=" << old_c << ")"; + s = Status::Corruption(); + fprintf(stderr, "%s\n", oss.str().c_str()); + assert(false); + break; + } + Record new_rec(record.a_value(), b, new_c); + std::string new_primary_index_value = new_rec.EncodePrimaryIndexValue(); + ColumnFamilyHandle* cf = db_->DefaultColumnFamily(); + s = txn->Put(cf, pk, new_primary_index_value, /*assume_tracked=*/true); + if (!s.ok()) { + break; + } + std::string old_sk = it->key().ToString(/*hex=*/false); + std::string new_sk; + std::string new_crc; + std::tie(new_sk, new_crc) = new_rec.EncodeSecondaryIndexEntry(); + s = wb->SingleDelete(old_sk); + if (!s.ok()) { + break; + } + s = wb->Put(new_sk, new_crc); + if (!s.ok()) { + break; + } + + it->Next(); + } while (it->Valid()); + + if (!s.ok()) { + return s; + } + + s = txn->Prepare(); + + if (!s.ok()) { + return s; + } + + if (FLAGS_rollback_one_in > 0 && thread->rand.OneIn(FLAGS_rollback_one_in)) { + s = Status::Incomplete(); + return s; + } + + s = WriteToCommitTimeWriteBatch(*txn); + if (!s.ok()) { + return s; + } + + s = CommitAndCreateTimestampedSnapshotIfNeeded(thread, *txn); + + if (s.ok()) { + delete txn; + auto& key_gen = key_gen_for_c_.at(thread->tid); + key_gen->Replace(old_c, old_c_pos, new_c); + } + + return s; +#endif // !ROCKSDB_LITE +} + +Status MultiOpsTxnsStressTest::UpdatePrimaryIndexValueTxn(ThreadState* thread, + uint32_t a, + uint32_t b_delta) { +#ifdef ROCKSDB_LITE + (void)thread; + (void)a; + (void)b_delta; + return Status::NotSupported(); +#else + std::string pk_str = Record::EncodePrimaryKey(a); + Transaction* txn = nullptr; + WriteOptions wopts; + Status s = NewTxn(wopts, &txn); + if (!s.ok()) { + assert(!txn); + thread->stats.AddErrors(1); + return s; + } + + assert(txn); + + const Defer cleanup([&s, thread, txn, this]() { + if (s.ok()) { + thread->stats.AddGets(/*ngets=*/1, /*nfounds=*/1); + thread->stats.AddBytesForWrites( + /*nwrites=*/1, /*nbytes=*/Record::kPrimaryIndexEntrySize); + return; + } + if (s.IsNotFound()) { + thread->stats.AddGets(/*ngets=*/1, /*nfounds=*/0); + } else if (s.IsInvalidArgument()) { + // ignored. + } else if (s.IsBusy() || s.IsTimedOut() || s.IsTryAgain() || + s.IsMergeInProgress() || s.IsIncomplete()) { + // ignored. + } else { + thread->stats.AddErrors(1); + } + RollbackTxn(txn).PermitUncheckedError(); + }); + ReadOptions ropts; + ropts.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + std::string value; + s = txn->GetForUpdate(ropts, pk_str, &value); + if (!s.ok()) { + return s; + } + auto result = Record::DecodePrimaryIndexValue(value); + if (!std::get<0>(result).ok()) { + s = std::get<0>(result); + fprintf(stderr, "Cannot decode primary index value %s: %s\n", + Slice(value).ToString(true).c_str(), s.ToString().c_str()); + assert(false); + return s; + } + uint32_t b = std::get<1>(result) + b_delta; + uint32_t c = std::get<2>(result); + Record record(a, b, c); + std::string primary_index_value = record.EncodePrimaryIndexValue(); + ColumnFamilyHandle* cf = db_->DefaultColumnFamily(); + s = txn->Put(cf, pk_str, primary_index_value, /*assume_tracked=*/true); + if (!s.ok()) { + return s; + } + s = txn->Prepare(); + if (!s.ok()) { + return s; + } + + if (FLAGS_rollback_one_in > 0 && thread->rand.OneIn(FLAGS_rollback_one_in)) { + s = Status::Incomplete(); + return s; + } + + s = WriteToCommitTimeWriteBatch(*txn); + if (!s.ok()) { + return s; + } + + s = CommitAndCreateTimestampedSnapshotIfNeeded(thread, *txn); + + if (s.ok()) { + delete txn; + } + return s; +#endif // !ROCKSDB_LITE +} + +Status MultiOpsTxnsStressTest::PointLookupTxn(ThreadState* thread, + ReadOptions ropts, uint32_t a) { +#ifdef ROCKSDB_LITE + (void)thread; + (void)ropts; + (void)a; + return Status::NotSupported(); +#else + std::string pk_str = Record::EncodePrimaryKey(a); + // pk may or may not exist + PinnableSlice value; + + Transaction* txn = nullptr; + WriteOptions wopts; + Status s = NewTxn(wopts, &txn); + if (!s.ok()) { + assert(!txn); + thread->stats.AddErrors(1); + return s; + } + + assert(txn); + + const Defer cleanup([&s, thread, txn, this]() { + if (s.ok()) { + thread->stats.AddGets(/*ngets=*/1, /*nfounds=*/1); + return; + } else if (s.IsNotFound()) { + thread->stats.AddGets(/*ngets=*/1, /*nfounds=*/0); + } else { + thread->stats.AddErrors(1); + } + RollbackTxn(txn).PermitUncheckedError(); + }); + + std::shared_ptr<const Snapshot> snapshot; + SetupSnapshot(thread, ropts, *txn, snapshot); + + if (FLAGS_delay_snapshot_read_one_in > 0 && + thread->rand.OneIn(FLAGS_delay_snapshot_read_one_in)) { + uint64_t delay_ms = thread->rand.Uniform(100) + 1; + db_->GetDBOptions().env->SleepForMicroseconds( + static_cast<int>(delay_ms * 1000)); + } + + s = txn->Get(ropts, db_->DefaultColumnFamily(), pk_str, &value); + if (s.ok()) { + s = txn->Commit(); + } + if (s.ok()) { + delete txn; + } + return s; +#endif // !ROCKSDB_LITE +} + +Status MultiOpsTxnsStressTest::RangeScanTxn(ThreadState* thread, + ReadOptions ropts, uint32_t c) { +#ifdef ROCKSDB_LITE + (void)thread; + (void)ropts; + (void)c; + return Status::NotSupported(); +#else + std::string sk = Record::EncodeSecondaryKey(c); + + Transaction* txn = nullptr; + WriteOptions wopts; + Status s = NewTxn(wopts, &txn); + if (!s.ok()) { + assert(!txn); + thread->stats.AddErrors(1); + return s; + } + + assert(txn); + + const Defer cleanup([&s, thread, txn, this]() { + if (s.ok()) { + thread->stats.AddIterations(1); + return; + } + thread->stats.AddErrors(1); + RollbackTxn(txn).PermitUncheckedError(); + }); + + std::shared_ptr<const Snapshot> snapshot; + SetupSnapshot(thread, ropts, *txn, snapshot); + + if (FLAGS_delay_snapshot_read_one_in > 0 && + thread->rand.OneIn(FLAGS_delay_snapshot_read_one_in)) { + uint64_t delay_ms = thread->rand.Uniform(100) + 1; + db_->GetDBOptions().env->SleepForMicroseconds( + static_cast<int>(delay_ms * 1000)); + } + + std::unique_ptr<Iterator> iter(txn->GetIterator(ropts)); + + constexpr size_t total_nexts = 10; + size_t nexts = 0; + for (iter->Seek(sk); + iter->Valid() && nexts < total_nexts && iter->status().ok(); + iter->Next(), ++nexts) { + } + + if (iter->status().ok()) { + s = txn->Commit(); + } else { + s = iter->status(); + } + + if (s.ok()) { + delete txn; + } + + return s; +#endif // !ROCKSDB_LITE +} + +void MultiOpsTxnsStressTest::VerifyDb(ThreadState* thread) const { + if (thread->shared->HasVerificationFailedYet()) { + return; + } + const Snapshot* const snapshot = db_->GetSnapshot(); + assert(snapshot); + ManagedSnapshot snapshot_guard(db_, snapshot); + + std::ostringstream oss; + oss << "[snap=" << snapshot->GetSequenceNumber() << ","; + + auto* dbimpl = static_cast_with_check<DBImpl>(db_->GetRootDB()); + assert(dbimpl); + + oss << " last_published=" << dbimpl->GetLastPublishedSequence() << "] "; + + if (FLAGS_delay_snapshot_read_one_in > 0 && + thread->rand.OneIn(FLAGS_delay_snapshot_read_one_in)) { + uint64_t delay_ms = thread->rand.Uniform(100) + 1; + db_->GetDBOptions().env->SleepForMicroseconds( + static_cast<int>(delay_ms * 1000)); + } + + // TODO (yanqin) with a probability, we can use either forward or backward + // iterator in subsequent checks. We can also use more advanced features in + // range scan. For now, let's just use simple forward iteration with + // total_order_seek = true. + + // First, iterate primary index. + size_t primary_index_entries_count = 0; + { + std::string iter_ub_str; + PutFixed32(&iter_ub_str, Record::kPrimaryIndexId + 1); + std::reverse(iter_ub_str.begin(), iter_ub_str.end()); + Slice iter_ub = iter_ub_str; + + std::string start_key; + PutFixed32(&start_key, Record::kPrimaryIndexId); + std::reverse(start_key.begin(), start_key.end()); + + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions ropts; + ropts.snapshot = snapshot; + ropts.total_order_seek = true; + ropts.iterate_upper_bound = &iter_ub; + + std::unique_ptr<Iterator> it(db_->NewIterator(ropts)); + for (it->Seek(start_key); it->Valid(); it->Next()) { + Record record; + Status s = record.DecodePrimaryIndexEntry(it->key(), it->value()); + if (!s.ok()) { + oss << "Cannot decode primary index entry " << it->key().ToString(true) + << "=>" << it->value().ToString(true); + VerificationAbort(thread->shared, oss.str(), s); + assert(false); + return; + } + ++primary_index_entries_count; + + // Search secondary index. + uint32_t a = record.a_value(); + uint32_t c = record.c_value(); + char sk_buf[12]; + EncodeFixed32(sk_buf, Record::kSecondaryIndexId); + std::reverse(sk_buf, sk_buf + sizeof(uint32_t)); + EncodeFixed32(sk_buf + sizeof(uint32_t), c); + std::reverse(sk_buf + sizeof(uint32_t), sk_buf + 2 * sizeof(uint32_t)); + EncodeFixed32(sk_buf + 2 * sizeof(uint32_t), a); + std::reverse(sk_buf + 2 * sizeof(uint32_t), sk_buf + sizeof(sk_buf)); + Slice sk(sk_buf, sizeof(sk_buf)); + std::string value; + s = db_->Get(ropts, sk, &value); + if (!s.ok()) { + oss << "Cannot find secondary index entry " << sk.ToString(true); + VerificationAbort(thread->shared, oss.str(), s); + assert(false); + return; + } + } + } + + // Second, iterate secondary index. + size_t secondary_index_entries_count = 0; + { + std::string start_key; + PutFixed32(&start_key, Record::kSecondaryIndexId); + std::reverse(start_key.begin(), start_key.end()); + + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions ropts; + ropts.snapshot = snapshot; + ropts.total_order_seek = true; + + std::unique_ptr<Iterator> it(db_->NewIterator(ropts)); + for (it->Seek(start_key); it->Valid(); it->Next()) { + ++secondary_index_entries_count; + Record record; + Status s = record.DecodeSecondaryIndexEntry(it->key(), it->value()); + if (!s.ok()) { + oss << "Cannot decode secondary index entry " + << it->key().ToString(true) << "=>" << it->value().ToString(true); + VerificationAbort(thread->shared, oss.str(), s); + assert(false); + return; + } + // After decoding secondary index entry, we know a and c. Crc is verified + // in decoding phase. + // + // Form a primary key and search in the primary index. + std::string pk = Record::EncodePrimaryKey(record.a_value()); + std::string value; + s = db_->Get(ropts, pk, &value); + if (!s.ok()) { + oss << "Error searching pk " << Slice(pk).ToString(true) << ". " + << s.ToString() << ". sk " << it->key().ToString(true); + VerificationAbort(thread->shared, oss.str(), s); + assert(false); + return; + } + auto result = Record::DecodePrimaryIndexValue(value); + s = std::get<0>(result); + if (!s.ok()) { + oss << "Error decoding primary index value " + << Slice(value).ToString(true) << ". " << s.ToString(); + VerificationAbort(thread->shared, oss.str(), s); + assert(false); + return; + } + uint32_t c_in_primary = std::get<2>(result); + if (c_in_primary != record.c_value()) { + oss << "Pk/sk mismatch. pk: " << Slice(pk).ToString(true) << "=>" + << Slice(value).ToString(true) << " (a=" << record.a_value() + << ", c=" << c_in_primary << "), sk: " << it->key().ToString(true) + << " (c=" << record.c_value() << ")"; + VerificationAbort(thread->shared, oss.str(), s); + assert(false); + return; + } + } + } + + if (secondary_index_entries_count != primary_index_entries_count) { + oss << "Pk/sk mismatch: primary index has " << primary_index_entries_count + << " entries. Secondary index has " << secondary_index_entries_count + << " entries."; + VerificationAbort(thread->shared, oss.str(), Status::OK()); + assert(false); + return; + } +} + +// VerifyPkSkFast() can be called by MultiOpsTxnsStressListener's callbacks +// which can be called before TransactionDB::Open() returns to caller. +// Therefore, at that time, db_ and txn_db_ may still be nullptr. +// Caller has to make sure that the race condition does not happen. +void MultiOpsTxnsStressTest::VerifyPkSkFast(int job_id) { + DB* const db = db_aptr_.load(std::memory_order_acquire); + if (db == nullptr) { + return; + } + + assert(db_ == db); + assert(db_ != nullptr); + + const Snapshot* const snapshot = db_->GetSnapshot(); + assert(snapshot); + ManagedSnapshot snapshot_guard(db_, snapshot); + + std::ostringstream oss; + auto* dbimpl = static_cast_with_check<DBImpl>(db_->GetRootDB()); + assert(dbimpl); + + oss << "Job " << job_id << ": [" << snapshot->GetSequenceNumber() << "," + << dbimpl->GetLastPublishedSequence() << "] "; + + std::string start_key; + PutFixed32(&start_key, Record::kSecondaryIndexId); + std::reverse(start_key.begin(), start_key.end()); + + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions ropts; + ropts.snapshot = snapshot; + ropts.total_order_seek = true; + + std::unique_ptr<Iterator> it(db_->NewIterator(ropts)); + for (it->Seek(start_key); it->Valid(); it->Next()) { + Record record; + Status s = record.DecodeSecondaryIndexEntry(it->key(), it->value()); + if (!s.ok()) { + oss << "Cannot decode secondary index entry " << it->key().ToString(true) + << "=>" << it->value().ToString(true); + fprintf(stderr, "%s\n", oss.str().c_str()); + fflush(stderr); + assert(false); + } + // After decoding secondary index entry, we know a and c. Crc is verified + // in decoding phase. + // + // Form a primary key and search in the primary index. + std::string pk = Record::EncodePrimaryKey(record.a_value()); + std::string value; + s = db_->Get(ropts, pk, &value); + if (!s.ok()) { + oss << "Error searching pk " << Slice(pk).ToString(true) << ". " + << s.ToString() << ". sk " << it->key().ToString(true); + fprintf(stderr, "%s\n", oss.str().c_str()); + fflush(stderr); + assert(false); + } + auto result = Record::DecodePrimaryIndexValue(value); + s = std::get<0>(result); + if (!s.ok()) { + oss << "Error decoding primary index value " + << Slice(value).ToString(true) << ". " << s.ToString(); + fprintf(stderr, "%s\n", oss.str().c_str()); + fflush(stderr); + assert(false); + } + uint32_t c_in_primary = std::get<2>(result); + if (c_in_primary != record.c_value()) { + oss << "Pk/sk mismatch. pk: " << Slice(pk).ToString(true) << "=>" + << Slice(value).ToString(true) << " (a=" << record.a_value() + << ", c=" << c_in_primary << "), sk: " << it->key().ToString(true) + << " (c=" << record.c_value() << ")"; + fprintf(stderr, "%s\n", oss.str().c_str()); + fflush(stderr); + assert(false); + } + } +} + +std::pair<uint32_t, uint32_t> MultiOpsTxnsStressTest::ChooseExistingA( + ThreadState* thread) { + uint32_t tid = thread->tid; + auto& key_gen = key_gen_for_a_.at(tid); + return key_gen->ChooseExisting(); +} + +uint32_t MultiOpsTxnsStressTest::GenerateNextA(ThreadState* thread) { + uint32_t tid = thread->tid; + auto& key_gen = key_gen_for_a_.at(tid); + return key_gen->Allocate(); +} + +std::pair<uint32_t, uint32_t> MultiOpsTxnsStressTest::ChooseExistingC( + ThreadState* thread) { + uint32_t tid = thread->tid; + auto& key_gen = key_gen_for_c_.at(tid); + return key_gen->ChooseExisting(); +} + +uint32_t MultiOpsTxnsStressTest::GenerateNextC(ThreadState* thread) { + uint32_t tid = thread->tid; + auto& key_gen = key_gen_for_c_.at(tid); + return key_gen->Allocate(); +} + +#ifndef ROCKSDB_LITE +void MultiOpsTxnsStressTest::ProcessRecoveredPreparedTxnsHelper( + Transaction* txn, SharedState*) { + thread_local Random rand(static_cast<uint32_t>(FLAGS_seed)); + if (rand.OneIn(2)) { + Status s = txn->Commit(); + assert(s.ok()); + } else { + Status s = txn->Rollback(); + assert(s.ok()); + } +} + +Status MultiOpsTxnsStressTest::WriteToCommitTimeWriteBatch(Transaction& txn) { + WriteBatch* ctwb = txn.GetCommitTimeWriteBatch(); + assert(ctwb); + // Do not change the content in key_buf. + static constexpr char key_buf[sizeof(Record::kMetadataPrefix) + 4] = { + '\0', '\0', '\0', '\0', '\0', '\0', '\0', '\xff'}; + + uint64_t counter_val = counter_.Next(); + char val_buf[sizeof(counter_val)]; + EncodeFixed64(val_buf, counter_val); + return ctwb->Put(Slice(key_buf, sizeof(key_buf)), + Slice(val_buf, sizeof(val_buf))); +} + +Status MultiOpsTxnsStressTest::CommitAndCreateTimestampedSnapshotIfNeeded( + ThreadState* thread, Transaction& txn) { + Status s; + if (FLAGS_create_timestamped_snapshot_one_in > 0 && + thread->rand.OneInOpt(FLAGS_create_timestamped_snapshot_one_in)) { + uint64_t ts = db_stress_env->NowNanos(); + std::shared_ptr<const Snapshot> snapshot; + s = txn.CommitAndTryCreateSnapshot(/*notifier=*/nullptr, ts, &snapshot); + } else { + s = txn.Commit(); + } + assert(txn_db_); + if (FLAGS_create_timestamped_snapshot_one_in > 0 && + thread->rand.OneInOpt(50000)) { + uint64_t now = db_stress_env->NowNanos(); + constexpr uint64_t time_diff = static_cast<uint64_t>(1000) * 1000 * 1000; + txn_db_->ReleaseTimestampedSnapshotsOlderThan(now - time_diff); + } + return s; +} + +void MultiOpsTxnsStressTest::SetupSnapshot( + ThreadState* thread, ReadOptions& read_opts, Transaction& txn, + std::shared_ptr<const Snapshot>& snapshot) { + if (thread->rand.OneInOpt(2)) { + snapshot = txn_db_->GetLatestTimestampedSnapshot(); + } + + if (snapshot) { + read_opts.snapshot = snapshot.get(); + } else { + txn.SetSnapshot(); + read_opts.snapshot = txn.GetSnapshot(); + } +} +#endif // !ROCKSDB_LITE + +std::string MultiOpsTxnsStressTest::KeySpaces::EncodeTo() const { + std::string result; + PutFixed32(&result, lb_a); + PutFixed32(&result, ub_a); + PutFixed32(&result, lb_c); + PutFixed32(&result, ub_c); + return result; +} + +bool MultiOpsTxnsStressTest::KeySpaces::DecodeFrom(Slice data) { + if (!GetFixed32(&data, &lb_a) || !GetFixed32(&data, &ub_a) || + !GetFixed32(&data, &lb_c) || !GetFixed32(&data, &ub_c)) { + return false; + } + return true; +} + +void MultiOpsTxnsStressTest::PersistKeySpacesDesc( + const std::string& key_spaces_path, uint32_t lb_a, uint32_t ub_a, + uint32_t lb_c, uint32_t ub_c) { + KeySpaces key_spaces(lb_a, ub_a, lb_c, ub_c); + std::string key_spaces_rep = key_spaces.EncodeTo(); + + std::unique_ptr<WritableFile> wfile; + Status s1 = + Env::Default()->NewWritableFile(key_spaces_path, &wfile, EnvOptions()); + assert(s1.ok()); + assert(wfile); + s1 = wfile->Append(key_spaces_rep); + assert(s1.ok()); +} + +MultiOpsTxnsStressTest::KeySpaces MultiOpsTxnsStressTest::ReadKeySpacesDesc( + const std::string& key_spaces_path) { + KeySpaces key_spaces; + std::unique_ptr<SequentialFile> sfile; + Status s1 = + Env::Default()->NewSequentialFile(key_spaces_path, &sfile, EnvOptions()); + assert(s1.ok()); + assert(sfile); + char buf[16]; + Slice result; + s1 = sfile->Read(sizeof(buf), &result, buf); + assert(s1.ok()); + if (!key_spaces.DecodeFrom(result)) { + assert(false); + } + return key_spaces; +} + +// Create an empty database if necessary and preload it with initial test data. +// Key range [lb_a, ub_a), [lb_c, ub_c). The key ranges will be shared by +// 'threads' threads. +// PreloadDb() also sets up KeyGenerator objects for each sub key range +// operated on by each thread. +// Both [lb_a, ub_a) and [lb_c, ub_c) are partitioned. Each thread operates on +// one sub range, using KeyGenerators to generate keys. +// For example, we choose a from [0, 10000) and c from [0, 100). Number of +// threads is 32, their tids range from 0 to 31. +// Thread k chooses a from [312*k,312*(k+1)) and c from [3*k,3*(k+1)) if k<31. +// Thread 31 chooses a from [9672, 10000) and c from [93, 100). +// Within each subrange: a from [low1, high1), c from [low2, high2). +// high1 - low1 > high2 - low2 +// We reserve {high1 - 1} and {high2 - 1} as unallocated. +// The records are <low1,low2>, <low1+1,low2+1>, ..., +// <low1+k,low2+k%(high2-low2-1), <low1+k+1,low2+(k+1)%(high2-low2-1)>, ... +void MultiOpsTxnsStressTest::PreloadDb(SharedState* shared, int threads, + uint32_t lb_a, uint32_t ub_a, + uint32_t lb_c, uint32_t ub_c) { +#ifdef ROCKSDB_LITE + (void)shared; + (void)threads; + (void)lb_a; + (void)ub_a; + (void)lb_c; + (void)ub_c; +#else + key_gen_for_a_.resize(threads); + key_gen_for_c_.resize(threads); + + assert(ub_a > lb_a && ub_a > lb_a + threads); + assert(ub_c > lb_c && ub_c > lb_c + threads); + + PersistKeySpacesDesc(FLAGS_key_spaces_path, lb_a, ub_a, lb_c, ub_c); + + fprintf(stdout, "a from [%u, %u), c from [%u, %u)\n", + static_cast<unsigned int>(lb_a), static_cast<unsigned int>(ub_a), + static_cast<unsigned int>(lb_c), static_cast<unsigned int>(ub_c)); + + const uint32_t num_c = ub_c - lb_c; + const uint32_t num_c_per_thread = num_c / threads; + const uint32_t num_a = ub_a - lb_a; + const uint32_t num_a_per_thread = num_a / threads; + + WriteOptions wopts; + wopts.disableWAL = FLAGS_disable_wal; + Random rnd(shared->GetSeed()); + assert(txn_db_); + + std::vector<KeySet> existing_a_uniqs(threads); + std::vector<KeySet> non_existing_a_uniqs(threads); + std::vector<KeySet> existing_c_uniqs(threads); + std::vector<KeySet> non_existing_c_uniqs(threads); + + for (uint32_t a = lb_a; a < ub_a; ++a) { + uint32_t tid = (a - lb_a) / num_a_per_thread; + if (tid >= static_cast<uint32_t>(threads)) { + tid = threads - 1; + } + + uint32_t a_base = lb_a + tid * num_a_per_thread; + uint32_t a_hi = (tid < static_cast<uint32_t>(threads - 1)) + ? (a_base + num_a_per_thread) + : ub_a; + uint32_t a_delta = a - a_base; + + if (a == a_hi - 1) { + non_existing_a_uniqs[tid].insert(a); + continue; + } + + uint32_t c_base = lb_c + tid * num_c_per_thread; + uint32_t c_hi = (tid < static_cast<uint32_t>(threads - 1)) + ? (c_base + num_c_per_thread) + : ub_c; + uint32_t c_delta = a_delta % (c_hi - c_base - 1); + uint32_t c = c_base + c_delta; + + uint32_t b = rnd.Next(); + Record record(a, b, c); + WriteBatch wb; + const auto primary_index_entry = record.EncodePrimaryIndexEntry(); + Status s = wb.Put(primary_index_entry.first, primary_index_entry.second); + assert(s.ok()); + + const auto secondary_index_entry = record.EncodeSecondaryIndexEntry(); + s = wb.Put(secondary_index_entry.first, secondary_index_entry.second); + assert(s.ok()); + + s = txn_db_->Write(wopts, &wb); + assert(s.ok()); + + // TODO (yanqin): make the following check optional, especially when data + // size is large. + Record tmp_rec; + tmp_rec.SetB(record.b_value()); + s = tmp_rec.DecodeSecondaryIndexEntry(secondary_index_entry.first, + secondary_index_entry.second); + assert(s.ok()); + assert(tmp_rec == record); + + existing_a_uniqs[tid].insert(a); + existing_c_uniqs[tid].insert(c); + } + + for (int i = 0; i < threads; ++i) { + uint32_t my_seed = i + shared->GetSeed(); + + auto& key_gen_for_a = key_gen_for_a_[i]; + assert(!key_gen_for_a); + uint32_t low = lb_a + i * num_a_per_thread; + uint32_t high = (i < threads - 1) ? (low + num_a_per_thread) : ub_a; + assert(existing_a_uniqs[i].size() == high - low - 1); + assert(non_existing_a_uniqs[i].size() == 1); + key_gen_for_a = std::make_unique<KeyGenerator>( + my_seed, low, high, std::move(existing_a_uniqs[i]), + std::move(non_existing_a_uniqs[i])); + + auto& key_gen_for_c = key_gen_for_c_[i]; + assert(!key_gen_for_c); + low = lb_c + i * num_c_per_thread; + high = (i < threads - 1) ? (low + num_c_per_thread) : ub_c; + non_existing_c_uniqs[i].insert(high - 1); + assert(existing_c_uniqs[i].size() == high - low - 1); + assert(non_existing_c_uniqs[i].size() == 1); + key_gen_for_c = std::make_unique<KeyGenerator>( + my_seed, low, high, std::move(existing_c_uniqs[i]), + std::move(non_existing_c_uniqs[i])); + } +#endif // !ROCKSDB_LITE +} + +// Scan an existing, non-empty database. +// Set up [lb_a, ub_a) and [lb_c, ub_c) as test key ranges. +// Set up KeyGenerator objects for each sub key range operated on by each +// thread. +// Scan the entire database and for each subrange, populate the existing keys +// and non-existing keys. We currently require the non-existing keys be +// non-empty after initialization. +void MultiOpsTxnsStressTest::ScanExistingDb(SharedState* shared, int threads) { + key_gen_for_a_.resize(threads); + key_gen_for_c_.resize(threads); + + KeySpaces key_spaces = ReadKeySpacesDesc(FLAGS_key_spaces_path); + + const uint32_t lb_a = key_spaces.lb_a; + const uint32_t ub_a = key_spaces.ub_a; + const uint32_t lb_c = key_spaces.lb_c; + const uint32_t ub_c = key_spaces.ub_c; + + assert(lb_a < ub_a && lb_c < ub_c); + + fprintf(stdout, "a from [%u, %u), c from [%u, %u)\n", + static_cast<unsigned int>(lb_a), static_cast<unsigned int>(ub_a), + static_cast<unsigned int>(lb_c), static_cast<unsigned int>(ub_c)); + + assert(ub_a > lb_a && ub_a > lb_a + threads); + assert(ub_c > lb_c && ub_c > lb_c + threads); + + const uint32_t num_c = ub_c - lb_c; + const uint32_t num_c_per_thread = num_c / threads; + const uint32_t num_a = ub_a - lb_a; + const uint32_t num_a_per_thread = num_a / threads; + + assert(db_); + ReadOptions ropts; + std::vector<KeySet> existing_a_uniqs(threads); + std::vector<KeySet> non_existing_a_uniqs(threads); + std::vector<KeySet> existing_c_uniqs(threads); + std::vector<KeySet> non_existing_c_uniqs(threads); + { + std::string pk_lb_str = Record::EncodePrimaryKey(0); + std::string pk_ub_str = + Record::EncodePrimaryKey(std::numeric_limits<uint32_t>::max()); + Slice pk_lb = pk_lb_str; + Slice pk_ub = pk_ub_str; + ropts.iterate_lower_bound = &pk_lb; + ropts.iterate_upper_bound = &pk_ub; + ropts.total_order_seek = true; + std::unique_ptr<Iterator> it(db_->NewIterator(ropts)); + + for (it->SeekToFirst(); it->Valid(); it->Next()) { + Record record; + Status s = record.DecodePrimaryIndexEntry(it->key(), it->value()); + if (!s.ok()) { + fprintf(stderr, "Cannot decode primary index entry (%s => %s): %s\n", + it->key().ToString(true).c_str(), + it->value().ToString(true).c_str(), s.ToString().c_str()); + assert(false); + } + uint32_t a = record.a_value(); + assert(a >= lb_a); + assert(a < ub_a); + uint32_t tid = (a - lb_a) / num_a_per_thread; + if (tid >= static_cast<uint32_t>(threads)) { + tid = threads - 1; + } + + existing_a_uniqs[tid].insert(a); + + uint32_t c = record.c_value(); + assert(c >= lb_c); + assert(c < ub_c); + tid = (c - lb_c) / num_c_per_thread; + if (tid >= static_cast<uint32_t>(threads)) { + tid = threads - 1; + } + auto& existing_c_uniq = existing_c_uniqs[tid]; + existing_c_uniq.insert(c); + } + + for (uint32_t a = lb_a; a < ub_a; ++a) { + uint32_t tid = (a - lb_a) / num_a_per_thread; + if (tid >= static_cast<uint32_t>(threads)) { + tid = threads - 1; + } + if (0 == existing_a_uniqs[tid].count(a)) { + non_existing_a_uniqs[tid].insert(a); + } + } + + for (uint32_t c = lb_c; c < ub_c; ++c) { + uint32_t tid = (c - lb_c) / num_c_per_thread; + if (tid >= static_cast<uint32_t>(threads)) { + tid = threads - 1; + } + if (0 == existing_c_uniqs[tid].count(c)) { + non_existing_c_uniqs[tid].insert(c); + } + } + + for (int i = 0; i < threads; ++i) { + uint32_t my_seed = i + shared->GetSeed(); + auto& key_gen_for_a = key_gen_for_a_[i]; + assert(!key_gen_for_a); + uint32_t low = lb_a + i * num_a_per_thread; + uint32_t high = (i < threads - 1) ? (low + num_a_per_thread) : ub_a; + + // The following two assertions assume the test thread count and key + // space remain the same across different runs. Will need to relax. + assert(existing_a_uniqs[i].size() == high - low - 1); + assert(non_existing_a_uniqs[i].size() == 1); + + key_gen_for_a = std::make_unique<KeyGenerator>( + my_seed, low, high, std::move(existing_a_uniqs[i]), + std::move(non_existing_a_uniqs[i])); + + auto& key_gen_for_c = key_gen_for_c_[i]; + assert(!key_gen_for_c); + low = lb_c + i * num_c_per_thread; + high = (i < threads - 1) ? (low + num_c_per_thread) : ub_c; + + // The following two assertions assume the test thread count and key + // space remain the same across different runs. Will need to relax. + assert(existing_c_uniqs[i].size() == high - low - 1); + assert(non_existing_c_uniqs[i].size() == 1); + + key_gen_for_c = std::make_unique<KeyGenerator>( + my_seed, low, high, std::move(existing_c_uniqs[i]), + std::move(non_existing_c_uniqs[i])); + } + } +} + +StressTest* CreateMultiOpsTxnsStressTest() { + return new MultiOpsTxnsStressTest(); +} + +void CheckAndSetOptionsForMultiOpsTxnStressTest() { +#ifndef ROCKSDB_LITE + if (FLAGS_test_batches_snapshots || FLAGS_test_cf_consistency) { + fprintf(stderr, + "-test_multi_ops_txns is not compatible with " + "-test_bathces_snapshots and -test_cf_consistency\n"); + exit(1); + } + if (!FLAGS_use_txn) { + fprintf(stderr, "-use_txn must be true if -test_multi_ops_txns\n"); + exit(1); + } else if (FLAGS_test_secondary > 0) { + fprintf( + stderr, + "secondary instance does not support replaying logs (MANIFEST + WAL) " + "of TransactionDB with write-prepared/write-unprepared policy\n"); + exit(1); + } + if (FLAGS_clear_column_family_one_in > 0) { + fprintf(stderr, + "-test_multi_ops_txns is not compatible with clearing column " + "families\n"); + exit(1); + } + if (FLAGS_column_families > 1) { + // TODO (yanqin) support separating primary index and secondary index in + // different column families. + fprintf(stderr, + "-test_multi_ops_txns currently does not use more than one column " + "family\n"); + exit(1); + } + if (FLAGS_writepercent > 0 || FLAGS_delpercent > 0 || + FLAGS_delrangepercent > 0) { + fprintf(stderr, + "-test_multi_ops_txns requires that -writepercent, -delpercent and " + "-delrangepercent be 0\n"); + exit(1); + } + if (FLAGS_key_spaces_path.empty()) { + fprintf(stderr, + "Must specify a file to store ranges of A and C via " + "-key_spaces_path\n"); + exit(1); + } + if (FLAGS_create_timestamped_snapshot_one_in > 0) { + if (FLAGS_txn_write_policy != + static_cast<uint64_t>(TxnDBWritePolicy::WRITE_COMMITTED)) { + fprintf(stderr, + "Timestamped snapshot is not yet supported by " + "write-prepared/write-unprepared transactions\n"); + exit(1); + } + } + if (FLAGS_sync_fault_injection == 1) { + fprintf(stderr, + "Sync fault injection is currently not supported in " + "-test_multi_ops_txns\n"); + exit(1); + } +#else + fprintf(stderr, "-test_multi_ops_txns not supported in ROCKSDB_LITE mode\n"); + exit(1); +#endif // !ROCKSDB_LITE +} +} // namespace ROCKSDB_NAMESPACE + +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/multi_ops_txns_stress.h b/src/rocksdb/db_stress_tool/multi_ops_txns_stress.h new file mode 100644 index 000000000..7463d05d7 --- /dev/null +++ b/src/rocksdb/db_stress_tool/multi_ops_txns_stress.h @@ -0,0 +1,444 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" + +namespace ROCKSDB_NAMESPACE { + +// This file defines MultiOpsTxnsStress so that we can stress test RocksDB +// transactions on a simple, emulated relational table. +// +// The record format is similar to the example found at +// https://github.com/facebook/mysql-5.6/wiki/MyRocks-record-format. +// +// The table is created by +// ``` +// create table t1 ( +// a int primary key, +// b int, +// c int, +// key(c), +// ) +// ``` +// +// (For simplicity, we use uint32_t for int here.) +// +// For this table, there is a primary index using `a`, as well as a secondary +// index using `c` and `a`. +// +// Primary key format: +// | index id | M(a) | +// Primary index value: +// | b | c | +// M(a) represents the big-endian format of a. +// +// Secondary key format: +// | index id | M(c) | M(a) | +// Secondary index value: +// | crc32 | +// Similarly to M(a), M(c) is the big-endian format of c. +// +// The in-memory representation of a record is defined in class +// MultiOpsTxnsStress:Record that includes a number of helper methods to +// encode/decode primary index keys, primary index values, secondary index keys, +// secondary index values, etc. +// +// Sometimes primary index and secondary index reside on different column +// families, but sometimes they colocate in the same column family. Current +// implementation puts them in the same (default) column family, and this is +// subject to future change if we find it interesting to test the other case. +// +// Class MultiOpsTxnsStressTest has the following transactions for testing. +// +// 1. Primary key update +// UPDATE t1 SET a = 3 WHERE a = 2; +// ``` +// tx->GetForUpdate(primary key a=2) +// tx->GetForUpdate(primary key a=3) +// tx->Delete(primary key a=2) +// tx->Put(primary key a=3, value) +// tx->batch->SingleDelete(secondary key a=2) +// tx->batch->Put(secondary key a=3, value) +// tx->Prepare() +// Tx->Commit() +// ``` +// +// 2. Secondary key update +// UPDATE t1 SET c = 3 WHERE c = 2; +// ``` +// iter->Seek(secondary key) +// // Get corresponding primary key value(s) from iterator +// tx->GetForUpdate(primary key) +// tx->Put(primary key, value c=3) +// tx->batch->SingleDelete(secondary key c=2) +// tx->batch->Put(secondary key c=3) +// tx->Prepare() +// tx->Commit() +// ``` +// +// 3. Primary index value update +// UPDATE t1 SET b = b + 1 WHERE a = 2; +// ``` +// tx->GetForUpdate(primary key a=2) +// tx->Put(primary key a=2, value b=b+1) +// tx->Prepare() +// tx->Commit() +// ``` +// +// 4. Point lookup +// SELECT * FROM t1 WHERE a = 3; +// ``` +// tx->Get(primary key a=3) +// tx->Commit() +// ``` +// +// 5. Range scan +// SELECT * FROM t1 WHERE c = 2; +// ``` +// it = tx->GetIterator() +// it->Seek(secondary key c=2) +// tx->Commit() +// ``` + +class MultiOpsTxnsStressTest : public StressTest { + public: + class Record { + public: + static constexpr uint32_t kMetadataPrefix = 0; + static constexpr uint32_t kPrimaryIndexId = 1; + static constexpr uint32_t kSecondaryIndexId = 2; + + static constexpr size_t kPrimaryIndexEntrySize = 8 + 8; + static constexpr size_t kSecondaryIndexEntrySize = 12 + 4; + + static_assert(kPrimaryIndexId < kSecondaryIndexId, + "kPrimaryIndexId must be smaller than kSecondaryIndexId"); + + static_assert(sizeof(kPrimaryIndexId) == sizeof(uint32_t), + "kPrimaryIndexId must be 4 bytes"); + static_assert(sizeof(kSecondaryIndexId) == sizeof(uint32_t), + "kSecondaryIndexId must be 4 bytes"); + + // Used for generating search key to probe primary index. + static std::string EncodePrimaryKey(uint32_t a); + // Used for generating search prefix to probe secondary index. + static std::string EncodeSecondaryKey(uint32_t c); + // Used for generating search key to probe secondary index. + static std::string EncodeSecondaryKey(uint32_t c, uint32_t a); + + static std::tuple<Status, uint32_t, uint32_t> DecodePrimaryIndexValue( + Slice primary_index_value); + + static std::pair<Status, uint32_t> DecodeSecondaryIndexValue( + Slice secondary_index_value); + + Record() = default; + Record(uint32_t _a, uint32_t _b, uint32_t _c) : a_(_a), b_(_b), c_(_c) {} + + bool operator==(const Record& other) const { + return a_ == other.a_ && b_ == other.b_ && c_ == other.c_; + } + + bool operator!=(const Record& other) const { return !(*this == other); } + + std::pair<std::string, std::string> EncodePrimaryIndexEntry() const; + + std::string EncodePrimaryKey() const; + + std::string EncodePrimaryIndexValue() const; + + std::pair<std::string, std::string> EncodeSecondaryIndexEntry() const; + + std::string EncodeSecondaryKey() const; + + Status DecodePrimaryIndexEntry(Slice primary_index_key, + Slice primary_index_value); + + Status DecodeSecondaryIndexEntry(Slice secondary_index_key, + Slice secondary_index_value); + + uint32_t a_value() const { return a_; } + uint32_t b_value() const { return b_; } + uint32_t c_value() const { return c_; } + + void SetA(uint32_t _a) { a_ = _a; } + void SetB(uint32_t _b) { b_ = _b; } + void SetC(uint32_t _c) { c_ = _c; } + + std::string ToString() const { + std::string ret("("); + ret.append(std::to_string(a_)); + ret.append(","); + ret.append(std::to_string(b_)); + ret.append(","); + ret.append(std::to_string(c_)); + ret.append(")"); + return ret; + } + + private: + friend class InvariantChecker; + + uint32_t a_{0}; + uint32_t b_{0}; + uint32_t c_{0}; + }; + + MultiOpsTxnsStressTest() {} + + ~MultiOpsTxnsStressTest() override {} + + void FinishInitDb(SharedState*) override; + + void ReopenAndPreloadDbIfNeeded(SharedState* shared); + + bool IsStateTracked() const override { return false; } + + Status TestGet(ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + std::vector<Status> TestMultiGet( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + Status TestPrefixScan(ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + // Given a key K, this creates an iterator which scans to K and then + // does a random sequence of Next/Prev operations. + Status TestIterate(ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + Status TestPut(ThreadState* thread, WriteOptions& write_opts, + const ReadOptions& read_opts, const std::vector<int>& cf_ids, + const std::vector<int64_t>& keys, char (&value)[100]) override; + + Status TestDelete(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + Status TestDeleteRange(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + void TestIngestExternalFile(ThreadState* thread, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + void TestCompactRange(ThreadState* thread, int64_t rand_key, + const Slice& start_key, + ColumnFamilyHandle* column_family) override; + + Status TestBackupRestore(ThreadState* thread, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + + Status TestCheckpoint(ThreadState* thread, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; + +#ifndef ROCKSDB_LITE + Status TestApproximateSize(ThreadState* thread, uint64_t iteration, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override; +#endif // !ROCKSDB_LITE + + Status TestCustomOperations( + ThreadState* thread, + const std::vector<int>& rand_column_families) override; + + void RegisterAdditionalListeners() override; + +#ifndef ROCKSDB_LITE + void PrepareTxnDbOptions(SharedState* /*shared*/, + TransactionDBOptions& txn_db_opts) override; +#endif // !ROCKSDB_LITE + + Status PrimaryKeyUpdateTxn(ThreadState* thread, uint32_t old_a, + uint32_t old_a_pos, uint32_t new_a); + + Status SecondaryKeyUpdateTxn(ThreadState* thread, uint32_t old_c, + uint32_t old_c_pos, uint32_t new_c); + + Status UpdatePrimaryIndexValueTxn(ThreadState* thread, uint32_t a, + uint32_t b_delta); + + Status PointLookupTxn(ThreadState* thread, ReadOptions ropts, uint32_t a); + + Status RangeScanTxn(ThreadState* thread, ReadOptions ropts, uint32_t c); + + void VerifyDb(ThreadState* thread) const override; + + void ContinuouslyVerifyDb(ThreadState* thread) const override { + VerifyDb(thread); + } + + void VerifyPkSkFast(int job_id); + + protected: + class Counter { + public: + uint64_t Next() { return value_.fetch_add(1); } + + private: + std::atomic<uint64_t> value_ = Env::Default()->NowNanos(); + }; + + using KeySet = std::set<uint32_t>; + class KeyGenerator { + public: + explicit KeyGenerator(uint32_t s, uint32_t low, uint32_t high, + KeySet&& existing_uniq, KeySet&& non_existing_uniq) + : rand_(s), + low_(low), + high_(high), + existing_uniq_(std::move(existing_uniq)), + non_existing_uniq_(std::move(non_existing_uniq)) {} + ~KeyGenerator() { + assert(!existing_uniq_.empty()); + assert(!non_existing_uniq_.empty()); + } + void FinishInit(); + + std::pair<uint32_t, uint32_t> ChooseExisting(); + void Replace(uint32_t old_val, uint32_t old_pos, uint32_t new_val); + uint32_t Allocate(); + void UndoAllocation(uint32_t new_val); + + std::string ToString() const { + std::ostringstream oss; + oss << "[" << low_ << ", " << high_ << "): " << existing_.size() + << " elements, " << existing_uniq_.size() << " unique values, " + << non_existing_uniq_.size() << " unique non-existing values"; + return oss.str(); + } + + private: + Random rand_; + uint32_t low_ = 0; + uint32_t high_ = 0; + std::vector<uint32_t> existing_{}; + KeySet existing_uniq_{}; + KeySet non_existing_uniq_{}; + bool initialized_ = false; + }; + + // Return <a, pos> + std::pair<uint32_t, uint32_t> ChooseExistingA(ThreadState* thread); + + uint32_t GenerateNextA(ThreadState* thread); + + // Return <c, pos> + std::pair<uint32_t, uint32_t> ChooseExistingC(ThreadState* thread); + + uint32_t GenerateNextC(ThreadState* thread); + +#ifndef ROCKSDB_LITE + // Randomly commit or rollback `txn` + void ProcessRecoveredPreparedTxnsHelper(Transaction* txn, + SharedState*) override; + + // Some applications, e.g. MyRocks writes a KV pair to the database via + // commit-time-write-batch (ctwb) in additional to the transaction's regular + // write batch. The key is usually constant representing some system + // metadata, while the value is monoticailly increasing which represents the + // actual value of the metadata. Method WriteToCommitTimeWriteBatch() + // emulates this scenario. + Status WriteToCommitTimeWriteBatch(Transaction& txn); + + Status CommitAndCreateTimestampedSnapshotIfNeeded(ThreadState* thread, + Transaction& txn); + + void SetupSnapshot(ThreadState* thread, ReadOptions& read_opts, + Transaction& txn, + std::shared_ptr<const Snapshot>& snapshot); +#endif //! ROCKSDB_LITE + + std::vector<std::unique_ptr<KeyGenerator>> key_gen_for_a_; + std::vector<std::unique_ptr<KeyGenerator>> key_gen_for_c_; + + Counter counter_{}; + + private: + struct KeySpaces { + uint32_t lb_a = 0; + uint32_t ub_a = 0; + uint32_t lb_c = 0; + uint32_t ub_c = 0; + + explicit KeySpaces() = default; + explicit KeySpaces(uint32_t _lb_a, uint32_t _ub_a, uint32_t _lb_c, + uint32_t _ub_c) + : lb_a(_lb_a), ub_a(_ub_a), lb_c(_lb_c), ub_c(_ub_c) {} + + std::string EncodeTo() const; + bool DecodeFrom(Slice data); + }; + + void PersistKeySpacesDesc(const std::string& key_spaces_path, uint32_t lb_a, + uint32_t ub_a, uint32_t lb_c, uint32_t ub_c); + + KeySpaces ReadKeySpacesDesc(const std::string& key_spaces_path); + + void PreloadDb(SharedState* shared, int threads, uint32_t lb_a, uint32_t ub_a, + uint32_t lb_c, uint32_t ub_c); + + void ScanExistingDb(SharedState* shared, int threads); +}; + +class InvariantChecker { + public: + static_assert(sizeof(MultiOpsTxnsStressTest::Record().a_) == sizeof(uint32_t), + "MultiOpsTxnsStressTest::Record::a_ must be 4 bytes"); + static_assert(sizeof(MultiOpsTxnsStressTest::Record().b_) == sizeof(uint32_t), + "MultiOpsTxnsStressTest::Record::b_ must be 4 bytes"); + static_assert(sizeof(MultiOpsTxnsStressTest::Record().c_) == sizeof(uint32_t), + "MultiOpsTxnsStressTest::Record::c_ must be 4 bytes"); +}; + +class MultiOpsTxnsStressListener : public EventListener { + public: + explicit MultiOpsTxnsStressListener(MultiOpsTxnsStressTest* stress_test) + : stress_test_(stress_test) { + assert(stress_test_); + } + +#ifndef ROCKSDB_LITE + ~MultiOpsTxnsStressListener() override {} + + void OnFlushCompleted(DB* db, const FlushJobInfo& info) override { + assert(db); +#ifdef NDEBUG + (void)db; +#endif + assert(info.cf_id == 0); + stress_test_->VerifyPkSkFast(info.job_id); + } + + void OnCompactionCompleted(DB* db, const CompactionJobInfo& info) override { + assert(db); +#ifdef NDEBUG + (void)db; +#endif + assert(info.cf_id == 0); + stress_test_->VerifyPkSkFast(info.job_id); + } +#endif //! ROCKSDB_LITE + + private: + MultiOpsTxnsStressTest* const stress_test_ = nullptr; +}; + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS diff --git a/src/rocksdb/db_stress_tool/no_batched_ops_stress.cc b/src/rocksdb/db_stress_tool/no_batched_ops_stress.cc new file mode 100644 index 000000000..bf01b788f --- /dev/null +++ b/src/rocksdb/db_stress_tool/no_batched_ops_stress.cc @@ -0,0 +1,1505 @@ +// Copyright (c) 2011-present, Facebook, Inc. All rights reserved. +// This source code is licensed under both the GPLv2 (found in the +// COPYING file in the root directory) and Apache 2.0 License +// (found in the LICENSE.Apache file in the root directory). +// +// Copyright (c) 2011 The LevelDB Authors. All rights reserved. +// Use of this source code is governed by a BSD-style license that can be +// found in the LICENSE file. See the AUTHORS file for names of contributors. + +#ifdef GFLAGS +#include "db_stress_tool/db_stress_common.h" +#include "rocksdb/utilities/transaction_db.h" +#include "utilities/fault_injection_fs.h" + +namespace ROCKSDB_NAMESPACE { +class NonBatchedOpsStressTest : public StressTest { + public: + NonBatchedOpsStressTest() {} + + virtual ~NonBatchedOpsStressTest() {} + + void VerifyDb(ThreadState* thread) const override { + // This `ReadOptions` is for validation purposes. Ignore + // `FLAGS_rate_limit_user_ops` to avoid slowing any validation. + ReadOptions options(FLAGS_verify_checksum, true); + std::string ts_str; + Slice ts; + if (FLAGS_user_timestamp_size > 0) { + ts_str = GetNowNanos(); + ts = ts_str; + options.timestamp = &ts; + } + + auto shared = thread->shared; + const int64_t max_key = shared->GetMaxKey(); + const int64_t keys_per_thread = max_key / shared->GetNumThreads(); + int64_t start = keys_per_thread * thread->tid; + int64_t end = start + keys_per_thread; + uint64_t prefix_to_use = + (FLAGS_prefix_size < 0) ? 1 : static_cast<size_t>(FLAGS_prefix_size); + + if (thread->tid == shared->GetNumThreads() - 1) { + end = max_key; + } + + for (size_t cf = 0; cf < column_families_.size(); ++cf) { + if (thread->shared->HasVerificationFailedYet()) { + break; + } + + enum class VerificationMethod { + kIterator, + kGet, + kMultiGet, + kGetMergeOperands, + // Add any new items above kNumberOfMethods + kNumberOfMethods + }; + + constexpr int num_methods = + static_cast<int>(VerificationMethod::kNumberOfMethods); + + const VerificationMethod method = + static_cast<VerificationMethod>(thread->rand.Uniform( + (FLAGS_user_timestamp_size > 0) ? num_methods - 1 : num_methods)); + + if (method == VerificationMethod::kIterator) { + std::unique_ptr<Iterator> iter( + db_->NewIterator(options, column_families_[cf])); + + std::string seek_key = Key(start); + iter->Seek(seek_key); + + Slice prefix(seek_key.data(), prefix_to_use); + + for (int64_t i = start; i < end; ++i) { + if (thread->shared->HasVerificationFailedYet()) { + break; + } + + const std::string key = Key(i); + const Slice k(key); + const Slice pfx(key.data(), prefix_to_use); + + // Reseek when the prefix changes + if (prefix_to_use > 0 && prefix.compare(pfx) != 0) { + iter->Seek(k); + seek_key = key; + prefix = Slice(seek_key.data(), prefix_to_use); + } + + Status s = iter->status(); + + std::string from_db; + + if (iter->Valid()) { + const int diff = iter->key().compare(k); + + if (diff > 0) { + s = Status::NotFound(); + } else if (diff == 0) { + const WideColumns expected_columns = GenerateExpectedWideColumns( + GetValueBase(iter->value()), iter->value()); + if (iter->columns() != expected_columns) { + VerificationAbort(shared, static_cast<int>(cf), i, + iter->value(), iter->columns(), + expected_columns); + break; + } + + from_db = iter->value().ToString(); + iter->Next(); + } else { + assert(diff < 0); + + VerificationAbort(shared, "An out of range key was found", + static_cast<int>(cf), i); + } + } else { + // The iterator found no value for the key in question, so do not + // move to the next item in the iterator + s = Status::NotFound(); + } + + VerifyOrSyncValue(static_cast<int>(cf), i, options, shared, from_db, + s, /* strict */ true); + + if (!from_db.empty()) { + PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i), + from_db.data(), from_db.size()); + } + } + } else if (method == VerificationMethod::kGet) { + for (int64_t i = start; i < end; ++i) { + if (thread->shared->HasVerificationFailedYet()) { + break; + } + + const std::string key = Key(i); + std::string from_db; + + Status s = db_->Get(options, column_families_[cf], key, &from_db); + + VerifyOrSyncValue(static_cast<int>(cf), i, options, shared, from_db, + s, /* strict */ true); + + if (!from_db.empty()) { + PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i), + from_db.data(), from_db.size()); + } + } + } else if (method == VerificationMethod::kMultiGet) { + for (int64_t i = start; i < end;) { + if (thread->shared->HasVerificationFailedYet()) { + break; + } + + // Keep the batch size to some reasonable value + size_t batch_size = thread->rand.Uniform(128) + 1; + batch_size = std::min<size_t>(batch_size, end - i); + + std::vector<std::string> keystrs(batch_size); + std::vector<Slice> keys(batch_size); + std::vector<PinnableSlice> values(batch_size); + std::vector<Status> statuses(batch_size); + + for (size_t j = 0; j < batch_size; ++j) { + keystrs[j] = Key(i + j); + keys[j] = Slice(keystrs[j].data(), keystrs[j].size()); + } + + db_->MultiGet(options, column_families_[cf], batch_size, keys.data(), + values.data(), statuses.data()); + + for (size_t j = 0; j < batch_size; ++j) { + const std::string from_db = values[j].ToString(); + + VerifyOrSyncValue(static_cast<int>(cf), i + j, options, shared, + from_db, statuses[j], /* strict */ true); + + if (!from_db.empty()) { + PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i + j), + from_db.data(), from_db.size()); + } + } + + i += batch_size; + } + } else { + assert(method == VerificationMethod::kGetMergeOperands); + + // Start off with small size that will be increased later if necessary + std::vector<PinnableSlice> values(4); + + GetMergeOperandsOptions merge_operands_info; + merge_operands_info.expected_max_number_of_operands = + static_cast<int>(values.size()); + + for (int64_t i = start; i < end; ++i) { + if (thread->shared->HasVerificationFailedYet()) { + break; + } + + const std::string key = Key(i); + const Slice k(key); + std::string from_db; + int number_of_operands = 0; + + Status s = db_->GetMergeOperands(options, column_families_[cf], k, + values.data(), &merge_operands_info, + &number_of_operands); + + if (s.IsIncomplete()) { + // Need to resize values as there are more than values.size() merge + // operands on this key. Should only happen a few times when we + // encounter a key that had more merge operands than any key seen so + // far + values.resize(number_of_operands); + merge_operands_info.expected_max_number_of_operands = + static_cast<int>(number_of_operands); + s = db_->GetMergeOperands(options, column_families_[cf], k, + values.data(), &merge_operands_info, + &number_of_operands); + } + // Assumed here that GetMergeOperands always sets number_of_operand + if (number_of_operands) { + from_db = values[number_of_operands - 1].ToString(); + } + + VerifyOrSyncValue(static_cast<int>(cf), i, options, shared, from_db, + s, /* strict */ true); + + if (!from_db.empty()) { + PrintKeyValue(static_cast<int>(cf), static_cast<uint32_t>(i), + from_db.data(), from_db.size()); + } + } + } + } + } + +#ifndef ROCKSDB_LITE + void ContinuouslyVerifyDb(ThreadState* thread) const override { + if (!cmp_db_) { + return; + } + assert(cmp_db_); + assert(!cmp_cfhs_.empty()); + Status s = cmp_db_->TryCatchUpWithPrimary(); + if (!s.ok()) { + assert(false); + exit(1); + } + + const auto checksum_column_family = [](Iterator* iter, + uint32_t* checksum) -> Status { + assert(nullptr != checksum); + uint32_t ret = 0; + for (iter->SeekToFirst(); iter->Valid(); iter->Next()) { + ret = crc32c::Extend(ret, iter->key().data(), iter->key().size()); + ret = crc32c::Extend(ret, iter->value().data(), iter->value().size()); + } + *checksum = ret; + return iter->status(); + }; + + auto* shared = thread->shared; + assert(shared); + const int64_t max_key = shared->GetMaxKey(); + ReadOptions read_opts(FLAGS_verify_checksum, true); + std::string ts_str; + Slice ts; + if (FLAGS_user_timestamp_size > 0) { + ts_str = GetNowNanos(); + ts = ts_str; + read_opts.timestamp = &ts; + } + + static Random64 rand64(shared->GetSeed()); + + { + uint32_t crc = 0; + std::unique_ptr<Iterator> it(cmp_db_->NewIterator(read_opts)); + s = checksum_column_family(it.get(), &crc); + if (!s.ok()) { + fprintf(stderr, "Computing checksum of default cf: %s\n", + s.ToString().c_str()); + assert(false); + } + } + + for (auto* handle : cmp_cfhs_) { + if (thread->rand.OneInOpt(3)) { + // Use Get() + uint64_t key = rand64.Uniform(static_cast<uint64_t>(max_key)); + std::string key_str = Key(key); + std::string value; + std::string key_ts; + s = cmp_db_->Get(read_opts, handle, key_str, &value, + FLAGS_user_timestamp_size > 0 ? &key_ts : nullptr); + s.PermitUncheckedError(); + } else { + // Use range scan + std::unique_ptr<Iterator> iter(cmp_db_->NewIterator(read_opts, handle)); + uint32_t rnd = (thread->rand.Next()) % 4; + if (0 == rnd) { + // SeekToFirst() + Next()*5 + read_opts.total_order_seek = true; + iter->SeekToFirst(); + for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Next()) { + } + } else if (1 == rnd) { + // SeekToLast() + Prev()*5 + read_opts.total_order_seek = true; + iter->SeekToLast(); + for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Prev()) { + } + } else if (2 == rnd) { + // Seek() +Next()*5 + uint64_t key = rand64.Uniform(static_cast<uint64_t>(max_key)); + std::string key_str = Key(key); + iter->Seek(key_str); + for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Next()) { + } + } else { + // SeekForPrev() + Prev()*5 + uint64_t key = rand64.Uniform(static_cast<uint64_t>(max_key)); + std::string key_str = Key(key); + iter->SeekForPrev(key_str); + for (int i = 0; i < 5 && iter->Valid(); ++i, iter->Prev()) { + } + } + } + } + } +#else + void ContinuouslyVerifyDb(ThreadState* /*thread*/) const override {} +#endif // ROCKSDB_LITE + + void MaybeClearOneColumnFamily(ThreadState* thread) override { + if (FLAGS_column_families > 1) { + if (thread->rand.OneInOpt(FLAGS_clear_column_family_one_in)) { + // drop column family and then create it again (can't drop default) + int cf = thread->rand.Next() % (FLAGS_column_families - 1) + 1; + std::string new_name = + std::to_string(new_column_family_name_.fetch_add(1)); + { + MutexLock l(thread->shared->GetMutex()); + fprintf( + stdout, + "[CF %d] Dropping and recreating column family. new name: %s\n", + cf, new_name.c_str()); + } + thread->shared->LockColumnFamily(cf); + Status s = db_->DropColumnFamily(column_families_[cf]); + delete column_families_[cf]; + if (!s.ok()) { + fprintf(stderr, "dropping column family error: %s\n", + s.ToString().c_str()); + std::terminate(); + } + s = db_->CreateColumnFamily(ColumnFamilyOptions(options_), new_name, + &column_families_[cf]); + column_family_names_[cf] = new_name; + thread->shared->ClearColumnFamily(cf); + if (!s.ok()) { + fprintf(stderr, "creating column family error: %s\n", + s.ToString().c_str()); + std::terminate(); + } + thread->shared->UnlockColumnFamily(cf); + } + } + } + + bool ShouldAcquireMutexOnKey() const override { return true; } + + bool IsStateTracked() const override { return true; } + + Status TestGet(ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + auto cfh = column_families_[rand_column_families[0]]; + std::string key_str = Key(rand_keys[0]); + Slice key = key_str; + std::string from_db; + int error_count = 0; + + if (fault_fs_guard) { + fault_fs_guard->EnableErrorInjection(); + SharedState::ignore_read_error = false; + } + + std::unique_ptr<MutexLock> lock(new MutexLock( + thread->shared->GetMutexForKey(rand_column_families[0], rand_keys[0]))); + + ReadOptions read_opts_copy = read_opts; + std::string read_ts_str; + Slice read_ts_slice; + bool read_older_ts = MaybeUseOlderTimestampForPointLookup( + thread, read_ts_str, read_ts_slice, read_opts_copy); + + Status s = db_->Get(read_opts_copy, cfh, key, &from_db); + if (fault_fs_guard) { + error_count = fault_fs_guard->GetAndResetErrorCount(); + } + if (s.ok()) { + if (fault_fs_guard) { + if (error_count && !SharedState::ignore_read_error) { + // Grab mutex so multiple thread don't try to print the + // stack trace at the same time + MutexLock l(thread->shared->GetMutex()); + fprintf(stderr, "Didn't get expected error from Get\n"); + fprintf(stderr, "Callstack that injected the fault\n"); + fault_fs_guard->PrintFaultBacktrace(); + std::terminate(); + } + } + // found case + thread->stats.AddGets(1, 1); + // we only have the latest expected state + if (!FLAGS_skip_verifydb && !read_opts_copy.timestamp && + thread->shared->Get(rand_column_families[0], rand_keys[0]) == + SharedState::DELETION_SENTINEL) { + thread->shared->SetVerificationFailure(); + fprintf(stderr, + "error : inconsistent values for key %s: Get returns %s, " + "expected state does not have the key.\n", + key.ToString(true).c_str(), StringToHex(from_db).c_str()); + } + } else if (s.IsNotFound()) { + // not found case + thread->stats.AddGets(1, 0); + if (!FLAGS_skip_verifydb && !read_older_ts) { + auto expected = + thread->shared->Get(rand_column_families[0], rand_keys[0]); + if (expected != SharedState::DELETION_SENTINEL && + expected != SharedState::UNKNOWN_SENTINEL) { + thread->shared->SetVerificationFailure(); + fprintf(stderr, + "error : inconsistent values for key %s: expected state has " + "the key, Get() returns NotFound.\n", + key.ToString(true).c_str()); + } + } + } else { + if (error_count == 0) { + // errors case + thread->stats.AddErrors(1); + } else { + thread->stats.AddVerifiedErrors(1); + } + } + if (fault_fs_guard) { + fault_fs_guard->DisableErrorInjection(); + } + return s; + } + + std::vector<Status> TestMultiGet( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + size_t num_keys = rand_keys.size(); + std::vector<std::string> key_str; + std::vector<Slice> keys; + key_str.reserve(num_keys); + keys.reserve(num_keys); + std::vector<PinnableSlice> values(num_keys); + std::vector<Status> statuses(num_keys); + ColumnFamilyHandle* cfh = column_families_[rand_column_families[0]]; + int error_count = 0; + // Do a consistency check between Get and MultiGet. Don't do it too + // often as it will slow db_stress down + bool do_consistency_check = thread->rand.OneIn(4); + + ReadOptions readoptionscopy = read_opts; + if (do_consistency_check) { + readoptionscopy.snapshot = db_->GetSnapshot(); + } + + std::string read_ts_str; + Slice read_ts_slice; + MaybeUseOlderTimestampForPointLookup(thread, read_ts_str, read_ts_slice, + readoptionscopy); + + readoptionscopy.rate_limiter_priority = + FLAGS_rate_limit_user_ops ? Env::IO_USER : Env::IO_TOTAL; + + // To appease clang analyzer + const bool use_txn = FLAGS_use_txn; + + // Create a transaction in order to write some data. The purpose is to + // exercise WriteBatchWithIndex::MultiGetFromBatchAndDB. The transaction + // will be rolled back once MultiGet returns. +#ifndef ROCKSDB_LITE + Transaction* txn = nullptr; + if (use_txn) { + WriteOptions wo; + if (FLAGS_rate_limit_auto_wal_flush) { + wo.rate_limiter_priority = Env::IO_USER; + } + Status s = NewTxn(wo, &txn); + if (!s.ok()) { + fprintf(stderr, "NewTxn: %s\n", s.ToString().c_str()); + std::terminate(); + } + } +#endif + for (size_t i = 0; i < num_keys; ++i) { + key_str.emplace_back(Key(rand_keys[i])); + keys.emplace_back(key_str.back()); +#ifndef ROCKSDB_LITE + if (use_txn) { + // With a 1 in 10 probability, insert the just added key in the batch + // into the transaction. This will create an overlap with the MultiGet + // keys and exercise some corner cases in the code + if (thread->rand.OneIn(10)) { + int op = thread->rand.Uniform(2); + Status s; + switch (op) { + case 0: + case 1: { + uint32_t value_base = + thread->rand.Next() % thread->shared->UNKNOWN_SENTINEL; + char value[100]; + size_t sz = GenerateValue(value_base, value, sizeof(value)); + Slice v(value, sz); + if (op == 0) { + s = txn->Put(cfh, keys.back(), v); + } else { + s = txn->Merge(cfh, keys.back(), v); + } + break; + } + case 2: + s = txn->Delete(cfh, keys.back()); + break; + default: + assert(false); + } + if (!s.ok()) { + fprintf(stderr, "Transaction put: %s\n", s.ToString().c_str()); + std::terminate(); + } + } + } +#endif + } + + if (!use_txn) { + if (fault_fs_guard) { + fault_fs_guard->EnableErrorInjection(); + SharedState::ignore_read_error = false; + } + db_->MultiGet(readoptionscopy, cfh, num_keys, keys.data(), values.data(), + statuses.data()); + if (fault_fs_guard) { + error_count = fault_fs_guard->GetAndResetErrorCount(); + } + } else { +#ifndef ROCKSDB_LITE + txn->MultiGet(readoptionscopy, cfh, num_keys, keys.data(), values.data(), + statuses.data()); +#endif + } + + if (fault_fs_guard && error_count && !SharedState::ignore_read_error) { + int stat_nok = 0; + for (const auto& s : statuses) { + if (!s.ok() && !s.IsNotFound()) { + stat_nok++; + } + } + + if (stat_nok < error_count) { + // Grab mutex so multiple thread don't try to print the + // stack trace at the same time + MutexLock l(thread->shared->GetMutex()); + fprintf(stderr, "Didn't get expected error from MultiGet. \n"); + fprintf(stderr, "num_keys %zu Expected %d errors, seen %d\n", num_keys, + error_count, stat_nok); + fprintf(stderr, "Callstack that injected the fault\n"); + fault_fs_guard->PrintFaultBacktrace(); + std::terminate(); + } + } + if (fault_fs_guard) { + fault_fs_guard->DisableErrorInjection(); + } + + for (size_t i = 0; i < statuses.size(); ++i) { + Status s = statuses[i]; + bool is_consistent = true; + // Only do the consistency check if no error was injected and MultiGet + // didn't return an unexpected error + if (do_consistency_check && !error_count && (s.ok() || s.IsNotFound())) { + Status tmp_s; + std::string value; + + if (use_txn) { +#ifndef ROCKSDB_LITE + tmp_s = txn->Get(readoptionscopy, cfh, keys[i], &value); +#endif // ROCKSDB_LITE + } else { + tmp_s = db_->Get(readoptionscopy, cfh, keys[i], &value); + } + if (!tmp_s.ok() && !tmp_s.IsNotFound()) { + fprintf(stderr, "Get error: %s\n", s.ToString().c_str()); + is_consistent = false; + } else if (!s.ok() && tmp_s.ok()) { + fprintf(stderr, "MultiGet returned different results with key %s\n", + keys[i].ToString(true).c_str()); + fprintf(stderr, "Get returned ok, MultiGet returned not found\n"); + is_consistent = false; + } else if (s.ok() && tmp_s.IsNotFound()) { + fprintf(stderr, "MultiGet returned different results with key %s\n", + keys[i].ToString(true).c_str()); + fprintf(stderr, "MultiGet returned ok, Get returned not found\n"); + is_consistent = false; + } else if (s.ok() && value != values[i].ToString()) { + fprintf(stderr, "MultiGet returned different results with key %s\n", + keys[i].ToString(true).c_str()); + fprintf(stderr, "MultiGet returned value %s\n", + values[i].ToString(true).c_str()); + fprintf(stderr, "Get returned value %s\n", + Slice(value).ToString(true /* hex */).c_str()); + is_consistent = false; + } + } + + if (!is_consistent) { + fprintf(stderr, "TestMultiGet error: is_consistent is false\n"); + thread->stats.AddErrors(1); + // Fail fast to preserve the DB state + thread->shared->SetVerificationFailure(); + break; + } else if (s.ok()) { + // found case + thread->stats.AddGets(1, 1); + } else if (s.IsNotFound()) { + // not found case + thread->stats.AddGets(1, 0); + } else if (s.IsMergeInProgress() && use_txn) { + // With txn this is sometimes expected. + thread->stats.AddGets(1, 1); + } else { + if (error_count == 0) { + // errors case + fprintf(stderr, "MultiGet error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + } else { + thread->stats.AddVerifiedErrors(1); + } + } + } + + if (readoptionscopy.snapshot) { + db_->ReleaseSnapshot(readoptionscopy.snapshot); + } + if (use_txn) { +#ifndef ROCKSDB_LITE + RollbackTxn(txn); +#endif + } + return statuses; + } + + Status TestPrefixScan(ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + ColumnFamilyHandle* const cfh = column_families_[rand_column_families[0]]; + assert(cfh); + + const std::string key = Key(rand_keys[0]); + const Slice prefix(key.data(), FLAGS_prefix_size); + + std::string upper_bound; + Slice ub_slice; + ReadOptions ro_copy = read_opts; + + // Get the next prefix first and then see if we want to set upper bound. + // We'll use the next prefix in an assertion later on + if (GetNextPrefix(prefix, &upper_bound) && thread->rand.OneIn(2)) { + // For half of the time, set the upper bound to the next prefix + ub_slice = Slice(upper_bound); + ro_copy.iterate_upper_bound = &ub_slice; + } + + std::string read_ts_str; + Slice read_ts_slice; + MaybeUseOlderTimestampForRangeScan(thread, read_ts_str, read_ts_slice, + ro_copy); + + std::unique_ptr<Iterator> iter(db_->NewIterator(ro_copy, cfh)); + + uint64_t count = 0; + Status s; + + for (iter->Seek(prefix); iter->Valid() && iter->key().starts_with(prefix); + iter->Next()) { + ++count; + + // When iter_start_ts is set, iterator exposes internal keys, including + // tombstones; however, we want to perform column validation only for + // value-like types. + if (ro_copy.iter_start_ts) { + const ValueType value_type = ExtractValueType(iter->key()); + if (value_type != kTypeValue && value_type != kTypeBlobIndex && + value_type != kTypeWideColumnEntity) { + continue; + } + } + + const WideColumns expected_columns = GenerateExpectedWideColumns( + GetValueBase(iter->value()), iter->value()); + if (iter->columns() != expected_columns) { + s = Status::Corruption( + "Value and columns inconsistent", + DebugString(iter->value(), iter->columns(), expected_columns)); + break; + } + } + + if (ro_copy.iter_start_ts == nullptr) { + assert(count <= GetPrefixKeyCount(prefix.ToString(), upper_bound)); + } + + if (s.ok()) { + s = iter->status(); + } + + if (!s.ok()) { + fprintf(stderr, "TestPrefixScan error: %s\n", s.ToString().c_str()); + thread->stats.AddErrors(1); + + return s; + } + + thread->stats.AddPrefixes(1, count); + + return Status::OK(); + } + + Status TestPut(ThreadState* thread, WriteOptions& write_opts, + const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys, + char (&value)[100]) override { + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + auto shared = thread->shared; + assert(shared); + + const int64_t max_key = shared->GetMaxKey(); + + int64_t rand_key = rand_keys[0]; + int rand_column_family = rand_column_families[0]; + std::string write_ts; + + std::unique_ptr<MutexLock> lock( + new MutexLock(shared->GetMutexForKey(rand_column_family, rand_key))); + while (!shared->AllowsOverwrite(rand_key) && + (FLAGS_use_merge || shared->Exists(rand_column_family, rand_key))) { + lock.reset(); + + rand_key = thread->rand.Next() % max_key; + rand_column_family = thread->rand.Next() % FLAGS_column_families; + + lock.reset( + new MutexLock(shared->GetMutexForKey(rand_column_family, rand_key))); + if (FLAGS_user_timestamp_size > 0) { + write_ts = GetNowNanos(); + } + } + + if (write_ts.empty() && FLAGS_user_timestamp_size) { + write_ts = GetNowNanos(); + } + + const std::string k = Key(rand_key); + + ColumnFamilyHandle* const cfh = column_families_[rand_column_family]; + assert(cfh); + + if (FLAGS_verify_before_write) { + std::string from_db; + Status s = db_->Get(read_opts, cfh, k, &from_db); + if (!VerifyOrSyncValue(rand_column_family, rand_key, read_opts, shared, + from_db, s, /* strict */ true)) { + return s; + } + } + + const uint32_t value_base = thread->rand.Next() % shared->UNKNOWN_SENTINEL; + const size_t sz = GenerateValue(value_base, value, sizeof(value)); + const Slice v(value, sz); + + shared->Put(rand_column_family, rand_key, value_base, true /* pending */); + + Status s; + + if (FLAGS_use_merge) { + if (!FLAGS_use_txn) { + if (FLAGS_user_timestamp_size == 0) { + s = db_->Merge(write_opts, cfh, k, v); + } else { + s = db_->Merge(write_opts, cfh, k, write_ts, v); + } + } else { +#ifndef ROCKSDB_LITE + Transaction* txn; + s = NewTxn(write_opts, &txn); + if (s.ok()) { + s = txn->Merge(cfh, k, v); + if (s.ok()) { + s = CommitTxn(txn, thread); + } + } +#endif + } + } else if (FLAGS_use_put_entity_one_in > 0 && + (value_base % FLAGS_use_put_entity_one_in) == 0) { + s = db_->PutEntity(write_opts, cfh, k, + GenerateWideColumns(value_base, v)); + } else { + if (!FLAGS_use_txn) { + if (FLAGS_user_timestamp_size == 0) { + s = db_->Put(write_opts, cfh, k, v); + } else { + s = db_->Put(write_opts, cfh, k, write_ts, v); + } + } else { +#ifndef ROCKSDB_LITE + Transaction* txn; + s = NewTxn(write_opts, &txn); + if (s.ok()) { + s = txn->Put(cfh, k, v); + if (s.ok()) { + s = CommitTxn(txn, thread); + } + } +#endif + } + } + + shared->Put(rand_column_family, rand_key, value_base, false /* pending */); + + if (!s.ok()) { + if (FLAGS_injest_error_severity >= 2) { + if (!is_db_stopped_ && s.severity() >= Status::Severity::kFatalError) { + is_db_stopped_ = true; + } else if (!is_db_stopped_ || + s.severity() < Status::Severity::kFatalError) { + fprintf(stderr, "put or merge error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } else { + fprintf(stderr, "put or merge error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } + + thread->stats.AddBytesForWrites(1, sz); + PrintKeyValue(rand_column_family, static_cast<uint32_t>(rand_key), value, + sz); + return s; + } + + Status TestDelete(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + int64_t rand_key = rand_keys[0]; + int rand_column_family = rand_column_families[0]; + auto shared = thread->shared; + + std::unique_ptr<MutexLock> lock( + new MutexLock(shared->GetMutexForKey(rand_column_family, rand_key))); + + // OPERATION delete + std::string write_ts_str = GetNowNanos(); + Slice write_ts = write_ts_str; + + std::string key_str = Key(rand_key); + Slice key = key_str; + auto cfh = column_families_[rand_column_family]; + + // Use delete if the key may be overwritten and a single deletion + // otherwise. + Status s; + if (shared->AllowsOverwrite(rand_key)) { + shared->Delete(rand_column_family, rand_key, true /* pending */); + if (!FLAGS_use_txn) { + if (FLAGS_user_timestamp_size == 0) { + s = db_->Delete(write_opts, cfh, key); + } else { + s = db_->Delete(write_opts, cfh, key, write_ts); + } + } else { +#ifndef ROCKSDB_LITE + Transaction* txn; + s = NewTxn(write_opts, &txn); + if (s.ok()) { + s = txn->Delete(cfh, key); + if (s.ok()) { + s = CommitTxn(txn, thread); + } + } +#endif + } + shared->Delete(rand_column_family, rand_key, false /* pending */); + thread->stats.AddDeletes(1); + if (!s.ok()) { + if (FLAGS_injest_error_severity >= 2) { + if (!is_db_stopped_ && + s.severity() >= Status::Severity::kFatalError) { + is_db_stopped_ = true; + } else if (!is_db_stopped_ || + s.severity() < Status::Severity::kFatalError) { + fprintf(stderr, "delete error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } else { + fprintf(stderr, "delete error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } + } else { + shared->SingleDelete(rand_column_family, rand_key, true /* pending */); + if (!FLAGS_use_txn) { + if (FLAGS_user_timestamp_size == 0) { + s = db_->SingleDelete(write_opts, cfh, key); + } else { + s = db_->SingleDelete(write_opts, cfh, key, write_ts); + } + } else { +#ifndef ROCKSDB_LITE + Transaction* txn; + s = NewTxn(write_opts, &txn); + if (s.ok()) { + s = txn->SingleDelete(cfh, key); + if (s.ok()) { + s = CommitTxn(txn, thread); + } + } +#endif + } + shared->SingleDelete(rand_column_family, rand_key, false /* pending */); + thread->stats.AddSingleDeletes(1); + if (!s.ok()) { + if (FLAGS_injest_error_severity >= 2) { + if (!is_db_stopped_ && + s.severity() >= Status::Severity::kFatalError) { + is_db_stopped_ = true; + } else if (!is_db_stopped_ || + s.severity() < Status::Severity::kFatalError) { + fprintf(stderr, "single delete error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } else { + fprintf(stderr, "single delete error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } + } + return s; + } + + Status TestDeleteRange(ThreadState* thread, WriteOptions& write_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + // OPERATION delete range + std::vector<std::unique_ptr<MutexLock>> range_locks; + // delete range does not respect disallowed overwrites. the keys for + // which overwrites are disallowed are randomly distributed so it + // could be expensive to find a range where each key allows + // overwrites. + int64_t rand_key = rand_keys[0]; + int rand_column_family = rand_column_families[0]; + auto shared = thread->shared; + int64_t max_key = shared->GetMaxKey(); + if (rand_key > max_key - FLAGS_range_deletion_width) { + rand_key = + thread->rand.Next() % (max_key - FLAGS_range_deletion_width + 1); + } + for (int j = 0; j < FLAGS_range_deletion_width; ++j) { + if (j == 0 || + ((rand_key + j) & ((1 << FLAGS_log2_keys_per_lock) - 1)) == 0) { + range_locks.emplace_back(new MutexLock( + shared->GetMutexForKey(rand_column_family, rand_key + j))); + } + } + shared->DeleteRange(rand_column_family, rand_key, + rand_key + FLAGS_range_deletion_width, + true /* pending */); + + std::string keystr = Key(rand_key); + Slice key = keystr; + auto cfh = column_families_[rand_column_family]; + std::string end_keystr = Key(rand_key + FLAGS_range_deletion_width); + Slice end_key = end_keystr; + std::string write_ts_str; + Slice write_ts; + Status s; + if (FLAGS_user_timestamp_size) { + write_ts_str = GetNowNanos(); + write_ts = write_ts_str; + s = db_->DeleteRange(write_opts, cfh, key, end_key, write_ts); + } else { + s = db_->DeleteRange(write_opts, cfh, key, end_key); + } + if (!s.ok()) { + if (FLAGS_injest_error_severity >= 2) { + if (!is_db_stopped_ && s.severity() >= Status::Severity::kFatalError) { + is_db_stopped_ = true; + } else if (!is_db_stopped_ || + s.severity() < Status::Severity::kFatalError) { + fprintf(stderr, "delete range error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } else { + fprintf(stderr, "delete range error: %s\n", s.ToString().c_str()); + std::terminate(); + } + } + int covered = shared->DeleteRange(rand_column_family, rand_key, + rand_key + FLAGS_range_deletion_width, + false /* pending */); + thread->stats.AddRangeDeletions(1); + thread->stats.AddCoveredByRangeDeletions(covered); + return s; + } + +#ifdef ROCKSDB_LITE + void TestIngestExternalFile( + ThreadState* /* thread */, + const std::vector<int>& /* rand_column_families */, + const std::vector<int64_t>& /* rand_keys */) override { + assert(false); + fprintf(stderr, + "RocksDB lite does not support " + "TestIngestExternalFile\n"); + std::terminate(); + } +#else + void TestIngestExternalFile(ThreadState* thread, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + const std::string sst_filename = + FLAGS_db + "/." + std::to_string(thread->tid) + ".sst"; + Status s; + if (db_stress_env->FileExists(sst_filename).ok()) { + // Maybe we terminated abnormally before, so cleanup to give this file + // ingestion a clean slate + s = db_stress_env->DeleteFile(sst_filename); + } + + SstFileWriter sst_file_writer(EnvOptions(options_), options_); + if (s.ok()) { + s = sst_file_writer.Open(sst_filename); + } + int64_t key_base = rand_keys[0]; + int column_family = rand_column_families[0]; + std::vector<std::unique_ptr<MutexLock>> range_locks; + range_locks.reserve(FLAGS_ingest_external_file_width); + std::vector<int64_t> keys; + keys.reserve(FLAGS_ingest_external_file_width); + std::vector<uint32_t> values; + values.reserve(FLAGS_ingest_external_file_width); + SharedState* shared = thread->shared; + + assert(FLAGS_nooverwritepercent < 100); + // Grab locks, set pending state on expected values, and add keys + for (int64_t key = key_base; + s.ok() && key < shared->GetMaxKey() && + static_cast<int32_t>(keys.size()) < FLAGS_ingest_external_file_width; + ++key) { + if (key == key_base || + (key & ((1 << FLAGS_log2_keys_per_lock) - 1)) == 0) { + range_locks.emplace_back( + new MutexLock(shared->GetMutexForKey(column_family, key))); + } + if (!shared->AllowsOverwrite(key)) { + // We could alternatively include `key` on the condition its current + // value is `DELETION_SENTINEL`. + continue; + } + keys.push_back(key); + + uint32_t value_base = thread->rand.Next() % shared->UNKNOWN_SENTINEL; + values.push_back(value_base); + shared->Put(column_family, key, value_base, true /* pending */); + + char value[100]; + size_t value_len = GenerateValue(value_base, value, sizeof(value)); + auto key_str = Key(key); + s = sst_file_writer.Put(Slice(key_str), Slice(value, value_len)); + } + + if (s.ok() && keys.empty()) { + return; + } + + if (s.ok()) { + s = sst_file_writer.Finish(); + } + if (s.ok()) { + s = db_->IngestExternalFile(column_families_[column_family], + {sst_filename}, IngestExternalFileOptions()); + } + if (!s.ok()) { + fprintf(stderr, "file ingestion error: %s\n", s.ToString().c_str()); + std::terminate(); + } + for (size_t i = 0; i < keys.size(); ++i) { + shared->Put(column_family, keys[i], values[i], false /* pending */); + } + } +#endif // ROCKSDB_LITE + + // Given a key K, this creates an iterator which scans the range + // [K, K + FLAGS_num_iterations) forward and backward. + // Then does a random sequence of Next/Prev operations. + Status TestIterateAgainstExpected( + ThreadState* thread, const ReadOptions& read_opts, + const std::vector<int>& rand_column_families, + const std::vector<int64_t>& rand_keys) override { + assert(thread); + assert(!rand_column_families.empty()); + assert(!rand_keys.empty()); + + auto shared = thread->shared; + assert(shared); + + int64_t max_key = shared->GetMaxKey(); + + const int64_t num_iter = static_cast<int64_t>(FLAGS_num_iterations); + + int64_t lb = rand_keys[0]; + if (lb > max_key - num_iter) { + lb = thread->rand.Next() % (max_key - num_iter + 1); + } + + const int64_t ub = lb + num_iter; + + // Lock the whole range over which we might iterate to ensure it doesn't + // change under us. + const int rand_column_family = rand_column_families[0]; + std::vector<std::unique_ptr<MutexLock>> range_locks = + shared->GetLocksForKeyRange(rand_column_family, lb, ub); + + ReadOptions ro(read_opts); + ro.total_order_seek = true; + + std::string read_ts_str; + Slice read_ts; + if (FLAGS_user_timestamp_size > 0) { + read_ts_str = GetNowNanos(); + read_ts = read_ts_str; + ro.timestamp = &read_ts; + } + + std::string max_key_str; + Slice max_key_slice; + if (!FLAGS_destroy_db_initially) { + max_key_str = Key(max_key); + max_key_slice = max_key_str; + // to restrict iterator from reading keys written in batched_op_stress + // that do not have expected state updated and may not be parseable by + // GetIntVal(). + ro.iterate_upper_bound = &max_key_slice; + } + + ColumnFamilyHandle* const cfh = column_families_[rand_column_family]; + assert(cfh); + + std::unique_ptr<Iterator> iter(db_->NewIterator(ro, cfh)); + + std::string op_logs; + + auto check_columns = [&]() { + assert(iter); + assert(iter->Valid()); + + const WideColumns expected_columns = GenerateExpectedWideColumns( + GetValueBase(iter->value()), iter->value()); + if (iter->columns() != expected_columns) { + shared->SetVerificationFailure(); + + fprintf(stderr, + "Verification failed for key %s: " + "Value and columns inconsistent: %s\n", + Slice(iter->key()).ToString(/* hex */ true).c_str(), + DebugString(iter->value(), iter->columns(), expected_columns) + .c_str()); + fprintf(stderr, "Column family: %s, op_logs: %s\n", + cfh->GetName().c_str(), op_logs.c_str()); + + thread->stats.AddErrors(1); + + return false; + } + + return true; + }; + + auto check_no_key_in_range = [&](int64_t start, int64_t end) { + for (auto j = std::max(start, lb); j < std::min(end, ub); ++j) { + auto expected_value = + shared->Get(rand_column_family, static_cast<int64_t>(j)); + if (expected_value != shared->DELETION_SENTINEL && + expected_value != shared->UNKNOWN_SENTINEL) { + // Fail fast to preserve the DB state. + thread->shared->SetVerificationFailure(); + if (iter->Valid()) { + fprintf(stderr, + "Expected state has key %s, iterator is at key %s\n", + Slice(Key(j)).ToString(true).c_str(), + iter->key().ToString(true).c_str()); + } else { + fprintf(stderr, "Expected state has key %s, iterator is invalid\n", + Slice(Key(j)).ToString(true).c_str()); + } + fprintf(stderr, "Column family: %s, op_logs: %s\n", + cfh->GetName().c_str(), op_logs.c_str()); + thread->stats.AddErrors(1); + return false; + } + } + return true; + }; + + // Forward and backward scan to ensure we cover the entire range [lb, ub). + // The random sequence Next and Prev test below tends to be very short + // ranged. + int64_t last_key = lb - 1; + + std::string key_str = Key(lb); + iter->Seek(key_str); + + op_logs += "S " + Slice(key_str).ToString(true) + " "; + + uint64_t curr = 0; + while (true) { + if (!iter->Valid()) { + if (!iter->status().ok()) { + thread->shared->SetVerificationFailure(); + fprintf(stderr, "TestIterate against expected state error: %s\n", + iter->status().ToString().c_str()); + fprintf(stderr, "Column family: %s, op_logs: %s\n", + cfh->GetName().c_str(), op_logs.c_str()); + thread->stats.AddErrors(1); + return iter->status(); + } + if (!check_no_key_in_range(last_key + 1, ub)) { + return Status::OK(); + } + break; + } + + if (!check_columns()) { + return Status::OK(); + } + + // iter is valid, the range (last_key, current key) was skipped + GetIntVal(iter->key().ToString(), &curr); + if (!check_no_key_in_range(last_key + 1, static_cast<int64_t>(curr))) { + return Status::OK(); + } + + last_key = static_cast<int64_t>(curr); + if (last_key >= ub - 1) { + break; + } + + iter->Next(); + + op_logs += "N"; + } + + // backward scan + key_str = Key(ub - 1); + iter->SeekForPrev(key_str); + + op_logs += " SFP " + Slice(key_str).ToString(true) + " "; + + last_key = ub; + while (true) { + if (!iter->Valid()) { + if (!iter->status().ok()) { + thread->shared->SetVerificationFailure(); + fprintf(stderr, "TestIterate against expected state error: %s\n", + iter->status().ToString().c_str()); + fprintf(stderr, "Column family: %s, op_logs: %s\n", + cfh->GetName().c_str(), op_logs.c_str()); + thread->stats.AddErrors(1); + return iter->status(); + } + if (!check_no_key_in_range(lb, last_key)) { + return Status::OK(); + } + break; + } + + if (!check_columns()) { + return Status::OK(); + } + + // the range (current key, last key) was skipped + GetIntVal(iter->key().ToString(), &curr); + if (!check_no_key_in_range(static_cast<int64_t>(curr + 1), last_key)) { + return Status::OK(); + } + + last_key = static_cast<int64_t>(curr); + if (last_key <= lb) { + break; + } + + iter->Prev(); + + op_logs += "P"; + } + + if (thread->rand.OneIn(2)) { + // Refresh after forward/backward scan to allow higher chance of SV + // change. It is safe to refresh since the testing key range is locked. + iter->Refresh(); + } + + // start from middle of [lb, ub) otherwise it is easy to iterate out of + // locked range + const int64_t mid = lb + num_iter / 2; + + key_str = Key(mid); + const Slice key(key_str); + + if (thread->rand.OneIn(2)) { + iter->Seek(key); + op_logs += " S " + key.ToString(true) + " "; + if (!iter->Valid() && iter->status().ok()) { + if (!check_no_key_in_range(mid, ub)) { + return Status::OK(); + } + } + } else { + iter->SeekForPrev(key); + op_logs += " SFP " + key.ToString(true) + " "; + if (!iter->Valid() && iter->status().ok()) { + // iterator says nothing <= mid + if (!check_no_key_in_range(lb, mid + 1)) { + return Status::OK(); + } + } + } + + for (int64_t i = 0; i < num_iter && iter->Valid(); ++i) { + if (!check_columns()) { + return Status::OK(); + } + + GetIntVal(iter->key().ToString(), &curr); + if (static_cast<int64_t>(curr) < lb) { + iter->Next(); + op_logs += "N"; + } else if (static_cast<int64_t>(curr) >= ub) { + iter->Prev(); + op_logs += "P"; + } else { + const uint32_t expected_value = + shared->Get(rand_column_family, static_cast<int64_t>(curr)); + if (expected_value == shared->DELETION_SENTINEL) { + // Fail fast to preserve the DB state. + thread->shared->SetVerificationFailure(); + fprintf(stderr, "Iterator has key %s, but expected state does not.\n", + iter->key().ToString(true).c_str()); + fprintf(stderr, "Column family: %s, op_logs: %s\n", + cfh->GetName().c_str(), op_logs.c_str()); + thread->stats.AddErrors(1); + break; + } + + if (thread->rand.OneIn(2)) { + iter->Next(); + op_logs += "N"; + if (!iter->Valid()) { + break; + } + uint64_t next = 0; + GetIntVal(iter->key().ToString(), &next); + if (!check_no_key_in_range(static_cast<int64_t>(curr + 1), + static_cast<int64_t>(next))) { + return Status::OK(); + } + } else { + iter->Prev(); + op_logs += "P"; + if (!iter->Valid()) { + break; + } + uint64_t prev = 0; + GetIntVal(iter->key().ToString(), &prev); + if (!check_no_key_in_range(static_cast<int64_t>(prev + 1), + static_cast<int64_t>(curr))) { + return Status::OK(); + } + } + } + } + + if (!iter->status().ok()) { + thread->shared->SetVerificationFailure(); + fprintf(stderr, "TestIterate against expected state error: %s\n", + iter->status().ToString().c_str()); + fprintf(stderr, "Column family: %s, op_logs: %s\n", + cfh->GetName().c_str(), op_logs.c_str()); + thread->stats.AddErrors(1); + return iter->status(); + } + + thread->stats.AddIterations(1); + + return Status::OK(); + } + + bool VerifyOrSyncValue(int cf, int64_t key, const ReadOptions& /*opts*/, + SharedState* shared, const std::string& value_from_db, + const Status& s, bool strict = false) const { + if (shared->HasVerificationFailedYet()) { + return false; + } + // compare value_from_db with the value in the shared state + uint32_t value_base = shared->Get(cf, key); + if (value_base == SharedState::UNKNOWN_SENTINEL) { + if (s.ok()) { + // Value exists in db, update state to reflect that + Slice slice(value_from_db); + value_base = GetValueBase(slice); + shared->Put(cf, key, value_base, false); + } else if (s.IsNotFound()) { + // Value doesn't exist in db, update state to reflect that + shared->SingleDelete(cf, key, false); + } + return true; + } + if (value_base == SharedState::DELETION_SENTINEL && !strict) { + return true; + } + + if (s.ok()) { + char value[kValueMaxLen]; + if (value_base == SharedState::DELETION_SENTINEL) { + VerificationAbort(shared, "Unexpected value found", cf, key, + value_from_db, ""); + return false; + } + size_t sz = GenerateValue(value_base, value, sizeof(value)); + if (value_from_db.length() != sz) { + VerificationAbort(shared, "Length of value read is not equal", cf, key, + value_from_db, Slice(value, sz)); + return false; + } + if (memcmp(value_from_db.data(), value, sz) != 0) { + VerificationAbort(shared, "Contents of value read don't match", cf, key, + value_from_db, Slice(value, sz)); + return false; + } + } else { + if (value_base != SharedState::DELETION_SENTINEL) { + char value[kValueMaxLen]; + size_t sz = GenerateValue(value_base, value, sizeof(value)); + VerificationAbort(shared, "Value not found: " + s.ToString(), cf, key, + "", Slice(value, sz)); + return false; + } + } + return true; + } + +#ifndef ROCKSDB_LITE + void PrepareTxnDbOptions(SharedState* shared, + TransactionDBOptions& txn_db_opts) override { + txn_db_opts.rollback_deletion_type_callback = + [shared](TransactionDB*, ColumnFamilyHandle*, const Slice& key) { + assert(shared); + uint64_t key_num = 0; + bool ok = GetIntVal(key.ToString(), &key_num); + assert(ok); + (void)ok; + return !shared->AllowsOverwrite(key_num); + }; + } +#endif // ROCKSDB_LITE +}; + +StressTest* CreateNonBatchedOpsStressTest() { + return new NonBatchedOpsStressTest(); +} + +} // namespace ROCKSDB_NAMESPACE +#endif // GFLAGS |