summaryrefslogtreecommitdiffstats
path: root/src/test/cls_rgw/test_cls_rgw_stats.cc
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--src/test/cls_rgw/test_cls_rgw_stats.cc646
1 files changed, 646 insertions, 0 deletions
diff --git a/src/test/cls_rgw/test_cls_rgw_stats.cc b/src/test/cls_rgw/test_cls_rgw_stats.cc
new file mode 100644
index 000000000..004ccc6d1
--- /dev/null
+++ b/src/test/cls_rgw/test_cls_rgw_stats.cc
@@ -0,0 +1,646 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <vector>
+#include <boost/circular_buffer.hpp>
+#include <boost/intrusive/set.hpp>
+#include <gtest/gtest.h>
+#include "cls/rgw/cls_rgw_client.h"
+#include "common/debug.h"
+#include "common/dout.h"
+#include "common/errno.h"
+#include "common/random_string.h"
+#include "global/global_context.h"
+#include "test/librados/test_cxx.h"
+
+#define dout_subsys ceph_subsys_rgw
+#define dout_context g_ceph_context
+
+
+// simulation parameters:
+
+// total number of index operations to prepare
+constexpr size_t max_operations = 2048;
+// total number of object names. each operation chooses one at random
+constexpr size_t max_entries = 32;
+// maximum number of pending operations. once the limit is reached, the oldest
+// pending operation is finished before another can start
+constexpr size_t max_pending = 16;
+// object size is randomly distributed between 0 and 4M
+constexpr size_t max_object_size = 4*1024*1024;
+// multipart upload threshold
+constexpr size_t max_part_size = 1024*1024;
+
+
+// create/destroy a pool that's shared by all tests in the process
+struct RadosEnv : public ::testing::Environment {
+ static std::optional<std::string> pool_name;
+ public:
+ static librados::Rados rados;
+ static librados::IoCtx ioctx;
+
+ void SetUp() override {
+ // create pool
+ std::string name = get_temp_pool_name();
+ ASSERT_EQ("", create_one_pool_pp(name, rados));
+ pool_name = name;
+ ASSERT_EQ(rados.ioctx_create(name.c_str(), ioctx), 0);
+ }
+ void TearDown() override {
+ ioctx.close();
+ if (pool_name) {
+ ASSERT_EQ(destroy_one_pool_pp(*pool_name, rados), 0);
+ }
+ }
+};
+std::optional<std::string> RadosEnv::pool_name;
+librados::Rados RadosEnv::rados;
+librados::IoCtx RadosEnv::ioctx;
+
+auto *const rados_env = ::testing::AddGlobalTestEnvironment(new RadosEnv);
+
+
+std::ostream& operator<<(std::ostream& out, const rgw_bucket_category_stats& c) {
+ return out << "{count=" << c.num_entries << " size=" << c.total_size << '}';
+}
+
+
+// librados helpers
+rgw_bucket_entry_ver last_version(librados::IoCtx& ioctx)
+{
+ rgw_bucket_entry_ver ver;
+ ver.pool = ioctx.get_id();
+ ver.epoch = ioctx.get_last_version();
+ return ver;
+}
+
+int index_init(librados::IoCtx& ioctx, const std::string& oid)
+{
+ librados::ObjectWriteOperation op;
+ cls_rgw_bucket_init_index(op);
+ return ioctx.operate(oid, &op);
+}
+
+int index_prepare(librados::IoCtx& ioctx, const std::string& oid,
+ const cls_rgw_obj_key& key, const std::string& tag,
+ RGWModifyOp type)
+{
+ librados::ObjectWriteOperation op;
+ const std::string loc; // empty
+ constexpr bool log_op = false;
+ constexpr int flags = 0;
+ rgw_zone_set zones;
+ cls_rgw_bucket_prepare_op(op, type, tag, key, loc, log_op, flags, zones);
+ return ioctx.operate(oid, &op);
+}
+
+int index_complete(librados::IoCtx& ioctx, const std::string& oid,
+ const cls_rgw_obj_key& key, const std::string& tag,
+ RGWModifyOp type, const rgw_bucket_entry_ver& ver,
+ const rgw_bucket_dir_entry_meta& meta,
+ std::list<cls_rgw_obj_key>* remove_objs)
+{
+ librados::ObjectWriteOperation op;
+ constexpr bool log_op = false;
+ constexpr int flags = 0;
+ constexpr rgw_zone_set* zones = nullptr;
+ cls_rgw_bucket_complete_op(op, type, tag, ver, key, meta,
+ remove_objs, log_op, flags, zones);
+ return ioctx.operate(oid, &op);
+}
+
+void read_stats(librados::IoCtx& ioctx, const std::string& oid,
+ rgw_bucket_dir_stats& stats)
+{
+ auto oids = std::map<int, std::string>{{0, oid}};
+ std::map<int, rgw_cls_list_ret> results;
+ ASSERT_EQ(0, CLSRGWIssueGetDirHeader(ioctx, oids, results, 8)());
+ ASSERT_EQ(1, results.size());
+ stats = std::move(results.begin()->second.dir.header.stats);
+}
+
+static void account_entry(rgw_bucket_dir_stats& stats,
+ const rgw_bucket_dir_entry_meta& meta)
+{
+ rgw_bucket_category_stats& c = stats[meta.category];
+ c.num_entries++;
+ c.total_size += meta.accounted_size;
+ c.total_size_rounded += cls_rgw_get_rounded_size(meta.accounted_size);
+ c.actual_size += meta.size;
+}
+
+static void unaccount_entry(rgw_bucket_dir_stats& stats,
+ const rgw_bucket_dir_entry_meta& meta)
+{
+ rgw_bucket_category_stats& c = stats[meta.category];
+ c.num_entries--;
+ c.total_size -= meta.accounted_size;
+ c.total_size_rounded -= cls_rgw_get_rounded_size(meta.accounted_size);
+ c.actual_size -= meta.size;
+}
+
+
+// a map of cached dir entries representing the expected state of cls_rgw
+struct object : rgw_bucket_dir_entry, boost::intrusive::set_base_hook<> {
+ explicit object(const cls_rgw_obj_key& key) {
+ this->key = key;
+ }
+};
+
+struct object_key {
+ using type = cls_rgw_obj_key;
+ const type& operator()(const object& o) const { return o.key; }
+};
+
+using object_map_base = boost::intrusive::set<object,
+ boost::intrusive::key_of_value<object_key>>;
+
+struct object_map : object_map_base {
+ ~object_map() {
+ clear_and_dispose(std::default_delete<object>{});
+ }
+};
+
+
+// models a bucket index operation, starting with cls_rgw_bucket_prepare_op().
+// stores all of the state necessary to complete the operation, either with
+// cls_rgw_bucket_complete_op() or cls_rgw_suggest_changes(). uploads larger
+// than max_part_size are modeled as multipart uploads
+struct operation {
+ RGWModifyOp type;
+ cls_rgw_obj_key key;
+ std::string tag;
+ rgw_bucket_entry_ver ver;
+ std::string upload_id; // empty unless multipart
+ rgw_bucket_dir_entry_meta meta;
+};
+
+
+class simulator {
+ public:
+ simulator(librados::IoCtx& ioctx, std::string oid)
+ : ioctx(ioctx), oid(std::move(oid)), pending(max_pending)
+ {
+ // generate a set of object keys. each operation chooses one at random
+ keys.reserve(max_entries);
+ for (size_t i = 0; i < max_entries; i++) {
+ keys.emplace_back(gen_rand_alphanumeric_upper(g_ceph_context, 12));
+ }
+ }
+
+ void run();
+
+ private:
+ void start();
+ int try_start(const cls_rgw_obj_key& key,
+ const std::string& tag);
+
+ void finish(const operation& op);
+ void complete(const operation& op, RGWModifyOp type);
+ void suggest(const operation& op, char suggestion);
+
+ int init_multipart(const operation& op);
+ void complete_multipart(const operation& op);
+
+ object_map::iterator find_or_create(const cls_rgw_obj_key& key);
+
+ librados::IoCtx& ioctx;
+ std::string oid;
+
+ std::vector<cls_rgw_obj_key> keys;
+ object_map objects;
+ boost::circular_buffer<operation> pending;
+ rgw_bucket_dir_stats stats;
+};
+
+
+void simulator::run()
+{
+ // init the bucket index object
+ ASSERT_EQ(0, index_init(ioctx, oid));
+ // run the simulation for N steps
+ for (size_t i = 0; i < max_operations; i++) {
+ if (pending.full()) {
+ // if we're at max_pending, finish the oldest operation
+ auto& op = pending.front();
+ finish(op);
+ pending.pop_front();
+
+ // verify bucket stats
+ rgw_bucket_dir_stats stored_stats;
+ read_stats(ioctx, oid, stored_stats);
+
+ const rgw_bucket_dir_stats& expected_stats = stats;
+ ASSERT_EQ(expected_stats, stored_stats);
+ }
+
+ // initiate the next operation
+ start();
+
+ // verify bucket stats
+ rgw_bucket_dir_stats stored_stats;
+ read_stats(ioctx, oid, stored_stats);
+
+ const rgw_bucket_dir_stats& expected_stats = stats;
+ ASSERT_EQ(expected_stats, stored_stats);
+ }
+}
+
+object_map::iterator simulator::find_or_create(const cls_rgw_obj_key& key)
+{
+ object_map::insert_commit_data commit;
+ auto result = objects.insert_check(key, std::less<cls_rgw_obj_key>{}, commit);
+ if (result.second) { // inserting new entry
+ auto p = new object(key);
+ result.first = objects.insert_commit(*p, commit);
+ }
+ return result.first;
+}
+
+int simulator::try_start(const cls_rgw_obj_key& key, const std::string& tag)
+{
+ // choose randomly betwen create and delete
+ const auto type = static_cast<RGWModifyOp>(
+ ceph::util::generate_random_number<size_t, size_t>(CLS_RGW_OP_ADD,
+ CLS_RGW_OP_DEL));
+ auto op = operation{type, key, tag};
+
+ op.meta.category = RGWObjCategory::Main;
+ op.meta.size = op.meta.accounted_size =
+ ceph::util::generate_random_number(1, max_object_size);
+
+ if (type == CLS_RGW_OP_ADD && op.meta.size > max_part_size) {
+ // simulate multipart for uploads over the max_part_size threshold
+ op.upload_id = gen_rand_alphanumeric_upper(g_ceph_context, 12);
+
+ int r = init_multipart(op);
+ if (r != 0) {
+ derr << "> failed to prepare multipart upload key=" << key
+ << " upload=" << op.upload_id << " tag=" << tag
+ << " type=" << type << ": " << cpp_strerror(r) << dendl;
+ return r;
+ }
+
+ dout(1) << "> prepared multipart upload key=" << key
+ << " upload=" << op.upload_id << " tag=" << tag
+ << " type=" << type << " size=" << op.meta.size << dendl;
+ } else {
+ // prepare operation
+ int r = index_prepare(ioctx, oid, op.key, op.tag, op.type);
+ if (r != 0) {
+ derr << "> failed to prepare operation key=" << key
+ << " tag=" << tag << " type=" << type
+ << ": " << cpp_strerror(r) << dendl;
+ return r;
+ }
+
+ dout(1) << "> prepared operation key=" << key
+ << " tag=" << tag << " type=" << type
+ << " size=" << op.meta.size << dendl;
+ }
+ op.ver = last_version(ioctx);
+
+ ceph_assert(!pending.full());
+ pending.push_back(std::move(op));
+ return 0;
+}
+
+void simulator::start()
+{
+ // choose a random object key
+ const size_t index = ceph::util::generate_random_number(0, keys.size() - 1);
+ const auto& key = keys[index];
+ // generate a random tag
+ const auto tag = gen_rand_alphanumeric_upper(g_ceph_context, 12);
+
+ // retry until success. failures don't count towards max_operations
+ while (try_start(key, tag) != 0)
+ ;
+}
+
+void simulator::finish(const operation& op)
+{
+ if (op.type == CLS_RGW_OP_ADD && !op.upload_id.empty()) {
+ // multipart uploads either complete or abort based on part uploads
+ complete_multipart(op);
+ return;
+ }
+
+ // complete most operations, but finish some with cancel or dir suggest
+ constexpr int cancel_percent = 10;
+ constexpr int suggest_update_percent = 10;
+ constexpr int suggest_remove_percent = 10;
+
+ int result = ceph::util::generate_random_number(0, 99);
+ if (result < cancel_percent) {
+ complete(op, CLS_RGW_OP_CANCEL);
+ return;
+ }
+ result -= cancel_percent;
+ if (result < suggest_update_percent) {
+ suggest(op, CEPH_RGW_UPDATE);
+ return;
+ }
+ result -= suggest_update_percent;
+ if (result < suggest_remove_percent) {
+ suggest(op, CEPH_RGW_REMOVE);
+ return;
+ }
+ complete(op, op.type);
+}
+
+void simulator::complete(const operation& op, RGWModifyOp type)
+{
+ int r = index_complete(ioctx, oid, op.key, op.tag, type,
+ op.ver, op.meta, nullptr);
+ if (r != 0) {
+ derr << "< failed to complete operation key=" << op.key
+ << " tag=" << op.tag << " type=" << op.type
+ << " size=" << op.meta.size
+ << ": " << cpp_strerror(r) << dendl;
+ return;
+ }
+
+ if (type == CLS_RGW_OP_CANCEL) {
+ dout(1) << "< canceled operation key=" << op.key
+ << " tag=" << op.tag << " type=" << op.type
+ << " size=" << op.meta.size << dendl;
+ } else if (type == CLS_RGW_OP_ADD) {
+ auto obj = find_or_create(op.key);
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ obj->exists = true;
+ obj->meta = op.meta;
+ account_entry(stats, obj->meta);
+ dout(1) << "< completed write operation key=" << op.key
+ << " tag=" << op.tag << " type=" << type
+ << " size=" << op.meta.size << dendl;
+ } else {
+ ceph_assert(type == CLS_RGW_OP_DEL);
+ auto obj = objects.find(op.key, std::less<cls_rgw_obj_key>{});
+ if (obj != objects.end()) {
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ objects.erase_and_dispose(obj, std::default_delete<object>{});
+ }
+ dout(1) << "< completed delete operation key=" << op.key
+ << " tag=" << op.tag << " type=" << type << dendl;
+ }
+}
+
+void simulator::suggest(const operation& op, char suggestion)
+{
+ // read and decode the current dir entry
+ rgw_cls_bi_entry bi_entry;
+ int r = cls_rgw_bi_get(ioctx, oid, BIIndexType::Plain, op.key, &bi_entry);
+ if (r != 0) {
+ derr << "< no bi entry to suggest for operation key=" << op.key
+ << " tag=" << op.tag << " type=" << op.type
+ << " size=" << op.meta.size
+ << ": " << cpp_strerror(r) << dendl;
+ return;
+ }
+ ASSERT_EQ(bi_entry.type, BIIndexType::Plain);
+
+ rgw_bucket_dir_entry entry;
+ auto p = bi_entry.data.cbegin();
+ ASSERT_NO_THROW(decode(entry, p));
+
+ ASSERT_EQ(entry.key, op.key);
+
+ // clear pending info and write it back; this cancels those pending
+ // operations (we'll see EINVAL when we try to complete them), but dir
+ // suggest is ignored unless the pending_map is empty
+ entry.pending_map.clear();
+
+ bi_entry.data.clear();
+ encode(entry, bi_entry.data);
+
+ r = cls_rgw_bi_put(ioctx, oid, bi_entry);
+ ASSERT_EQ(0, r);
+
+ // now suggest changes for this entry
+ entry.ver = last_version(ioctx);
+ entry.exists = (suggestion == CEPH_RGW_UPDATE);
+ entry.meta = op.meta;
+
+ bufferlist update;
+ cls_rgw_encode_suggestion(suggestion, entry, update);
+
+ librados::ObjectWriteOperation write_op;
+ cls_rgw_suggest_changes(write_op, update);
+ r = ioctx.operate(oid, &write_op);
+ if (r != 0) {
+ derr << "< failed to suggest operation key=" << op.key
+ << " tag=" << op.tag << " type=" << op.type
+ << " size=" << op.meta.size
+ << ": " << cpp_strerror(r) << dendl;
+ return;
+ }
+
+ // update our cache accordingly
+ if (suggestion == CEPH_RGW_UPDATE) {
+ auto obj = find_or_create(op.key);
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ obj->exists = true;
+ obj->meta = op.meta;
+ account_entry(stats, obj->meta);
+ dout(1) << "< suggested update operation key=" << op.key
+ << " tag=" << op.tag << " type=" << op.type
+ << " size=" << op.meta.size << dendl;
+ } else {
+ ceph_assert(suggestion == CEPH_RGW_REMOVE);
+ auto obj = objects.find(op.key, std::less<cls_rgw_obj_key>{});
+ if (obj != objects.end()) {
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ objects.erase_and_dispose(obj, std::default_delete<object>{});
+ }
+ dout(1) << "< suggested remove operation key=" << op.key
+ << " tag=" << op.tag << " type=" << op.type << dendl;
+ }
+}
+
+int simulator::init_multipart(const operation& op)
+{
+ // create (not just prepare) the meta object
+ const auto meta_key = cls_rgw_obj_key{
+ fmt::format("_multipart_{}.2~{}.meta", op.key.name, op.upload_id)};
+ const std::string empty_tag; // empty tag enables complete without prepare
+ const rgw_bucket_entry_ver empty_ver;
+ rgw_bucket_dir_entry_meta meta_meta;
+ meta_meta.category = RGWObjCategory::MultiMeta;
+ int r = index_complete(ioctx, oid, meta_key, empty_tag, CLS_RGW_OP_ADD,
+ empty_ver, meta_meta, nullptr);
+ if (r != 0) {
+ derr << " < failed to create multipart meta key=" << meta_key
+ << ": " << cpp_strerror(r) << dendl;
+ return r;
+ } else {
+ // account for meta object
+ auto obj = find_or_create(meta_key);
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ obj->exists = true;
+ obj->meta = meta_meta;
+ account_entry(stats, obj->meta);
+ }
+
+ // prepare part uploads
+ std::list<cls_rgw_obj_key> remove_objs;
+ size_t part_id = 0;
+
+ size_t remaining = op.meta.size;
+ while (remaining > max_part_size) {
+ remaining -= max_part_size;
+ const auto part_size = std::min(remaining, max_part_size);
+ const auto part_key = cls_rgw_obj_key{
+ fmt::format("_multipart_{}.2~{}.{}", op.key.name, op.upload_id, part_id)};
+ part_id++;
+
+ r = index_prepare(ioctx, oid, part_key, op.tag, op.type);
+ if (r != 0) {
+ // if part prepare fails, remove the meta object and remove_objs
+ [[maybe_unused]] int ignored =
+ index_complete(ioctx, oid, meta_key, empty_tag, CLS_RGW_OP_DEL,
+ empty_ver, meta_meta, &remove_objs);
+ derr << " > failed to prepare part key=" << part_key
+ << " size=" << part_size << dendl;
+ return r; // return the error from prepare
+ }
+ dout(1) << " > prepared part key=" << part_key
+ << " size=" << part_size << dendl;
+ remove_objs.push_back(part_key);
+ }
+ return 0;
+}
+
+void simulator::complete_multipart(const operation& op)
+{
+ const std::string empty_tag; // empty tag enables complete without prepare
+ const rgw_bucket_entry_ver empty_ver;
+
+ // try to finish part uploads
+ size_t part_id = 0;
+ std::list<cls_rgw_obj_key> remove_objs;
+
+ RGWModifyOp type = op.type; // OP_ADD, or OP_CANCEL for abort
+
+ size_t remaining = op.meta.size;
+ while (remaining > max_part_size) {
+ remaining -= max_part_size;
+ const auto part_size = std::min(remaining, max_part_size);
+ const auto part_key = cls_rgw_obj_key{
+ fmt::format("_multipart_{}.2~{}.{}", op.key.name, op.upload_id, part_id)};
+ part_id++;
+
+ // cancel 10% of part uploads (and abort the multipart upload)
+ constexpr int cancel_percent = 10;
+ const int result = ceph::util::generate_random_number(0, 99);
+ if (result < cancel_percent) {
+ type = CLS_RGW_OP_CANCEL; // abort multipart
+ dout(1) << " < canceled part key=" << part_key
+ << " size=" << part_size << dendl;
+ } else {
+ rgw_bucket_dir_entry_meta meta;
+ meta.category = op.meta.category;
+ meta.size = meta.accounted_size = part_size;
+
+ int r = index_complete(ioctx, oid, part_key, op.tag, op.type,
+ empty_ver, meta, nullptr);
+ if (r != 0) {
+ derr << " < failed to complete part key=" << part_key
+ << " size=" << meta.size << ": " << cpp_strerror(r) << dendl;
+ type = CLS_RGW_OP_CANCEL; // abort multipart
+ } else {
+ dout(1) << " < completed part key=" << part_key
+ << " size=" << meta.size << dendl;
+ // account for successful part upload
+ auto obj = find_or_create(part_key);
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ obj->exists = true;
+ obj->meta = meta;
+ account_entry(stats, obj->meta);
+ }
+ }
+ remove_objs.push_back(part_key);
+ }
+
+ // delete the multipart meta object
+ const auto meta_key = cls_rgw_obj_key{
+ fmt::format("_multipart_{}.2~{}.meta", op.key.name, op.upload_id)};
+ rgw_bucket_dir_entry_meta meta_meta;
+ meta_meta.category = RGWObjCategory::MultiMeta;
+
+ int r = index_complete(ioctx, oid, meta_key, empty_tag, CLS_RGW_OP_DEL,
+ empty_ver, meta_meta, nullptr);
+ if (r != 0) {
+ derr << " < failed to remove multipart meta key=" << meta_key
+ << ": " << cpp_strerror(r) << dendl;
+ } else {
+ // unaccount for meta object
+ auto obj = objects.find(meta_key, std::less<cls_rgw_obj_key>{});
+ if (obj != objects.end()) {
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ objects.erase_and_dispose(obj, std::default_delete<object>{});
+ }
+ }
+
+ // create or cancel the head object
+ r = index_complete(ioctx, oid, op.key, empty_tag, type,
+ empty_ver, op.meta, &remove_objs);
+ if (r != 0) {
+ derr << "< failed to complete multipart upload key=" << op.key
+ << " upload=" << op.upload_id << " tag=" << op.tag
+ << " type=" << type << " size=" << op.meta.size
+ << ": " << cpp_strerror(r) << dendl;
+ return;
+ }
+
+ if (type == CLS_RGW_OP_ADD) {
+ dout(1) << "< completed multipart upload key=" << op.key
+ << " upload=" << op.upload_id << " tag=" << op.tag
+ << " type=" << op.type << " size=" << op.meta.size << dendl;
+
+ // account for head stats
+ auto obj = find_or_create(op.key);
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ obj->exists = true;
+ obj->meta = op.meta;
+ account_entry(stats, obj->meta);
+ } else {
+ dout(1) << "< canceled multipart upload key=" << op.key
+ << " upload=" << op.upload_id << " tag=" << op.tag
+ << " type=" << op.type << " size=" << op.meta.size << dendl;
+ }
+
+ // unaccount for remove_objs
+ for (const auto& part_key : remove_objs) {
+ auto obj = objects.find(part_key, std::less<cls_rgw_obj_key>{});
+ if (obj != objects.end()) {
+ if (obj->exists) {
+ unaccount_entry(stats, obj->meta);
+ }
+ objects.erase_and_dispose(obj, std::default_delete<object>{});
+ }
+ }
+}
+
+TEST(cls_rgw_stats, simulate)
+{
+ const char* bucket_oid = __func__;
+ auto sim = simulator{RadosEnv::ioctx, bucket_oid};
+ sim.run();
+}
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-06 15:59:44 +0000