1 files changed, 210 insertions, 49 deletions

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index c61e7d332..515eb6042 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -1703,7 +1703,7 @@ void OSDService::queue_recovery_context(
   epoch_t e = get_osdmap_epoch();
 
   uint64_t cost_for_queue = [this, cost] {
-    if (cct->_conf->osd_op_queue == "mclock_scheduler") {
+    if (op_queue_type_t::mClockScheduler == osd->osd_op_queue_type()) {
       return cost;
     } else {
       /* We retain this legacy behavior for WeightedPriorityQueue. It seems to
@@ -1726,14 +1726,32 @@ void OSDService::queue_recovery_context(
       e));
 }
 
-void OSDService::queue_for_snap_trim(PG *pg)
+void OSDService::queue_for_snap_trim(PG *pg, uint64_t cost_per_object)
 {
   dout(10) << "queueing " << *pg << " for snaptrim" << dendl;
+  uint64_t cost_for_queue = [this, cost_per_object] {
+    if (cct->_conf->osd_op_queue == "mclock_scheduler") {
+      /* The cost calculation is valid for most snap trim iterations except
+       * for the following cases:
+       * 1) The penultimate iteration which may return 1 object to trim, in
+       *    which case the cost will be off by a factor equivalent to the
+       *    average object size, and,
+       * 2) The final iteration which returns -ENOENT and performs clean-ups.
+       */
+      return cost_per_object * cct->_conf->osd_pg_max_concurrent_snap_trims;
+    } else {
+      /* We retain this legacy behavior for WeightedPriorityQueue.
+       * This branch should be removed after Squid.
+       */
+      return cct->_conf->osd_snap_trim_cost;
+    }
+  }();
+
   enqueue_back(
     OpSchedulerItem(
       unique_ptr<OpSchedulerItem::OpQueueable>(
 	new PGSnapTrim(pg->get_pgid(), pg->get_osdmap_epoch())),
-      cct->_conf->osd_snap_trim_cost,
+      cost_for_queue,
       cct->_conf->osd_snap_trim_priority,
       ceph_clock_now(),
       0,
@@ -1771,7 +1789,7 @@ int64_t OSDService::get_scrub_cost()
 {
 
   int64_t cost_for_queue = cct->_conf->osd_scrub_cost;
-  if (cct->_conf->osd_op_queue == "mclock_scheduler") {
+  if (op_queue_type_t::mClockScheduler == osd->osd_op_queue_type()) {
     cost_for_queue = cct->_conf->osd_scrub_event_cost *
                      cct->_conf->osd_shallow_scrub_chunk_max;
   }
@@ -2049,7 +2067,7 @@ void OSDService::_queue_for_recovery(
   ceph_assert(ceph_mutex_is_locked_by_me(recovery_lock));
 
   uint64_t cost_for_queue = [this, &reserved_pushes, &p] {
-    if (cct->_conf->osd_op_queue == "mclock_scheduler") {
+    if (op_queue_type_t::mClockScheduler == osd->osd_op_queue_type()) {
       return p.cost_per_object * reserved_pushes;
     } else {
       /* We retain this legacy behavior for WeightedPriorityQueue. It seems to
@@ -2092,6 +2110,22 @@ int heap(CephContext& cct,
 
 } // namespace ceph::osd_cmds
 
+void OSD::write_superblock(CephContext* cct, OSDSuperblock& sb, ObjectStore::Transaction& t)
+{
+  dout(10) << "write_superblock " << sb << dendl;
+
+  //hack: at minimum it's using the baseline feature set
+  if (!sb.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_BASE))
+    sb.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE);
+
+  bufferlist bl;
+  encode(sb, bl);
+  t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl);
+  std::map<std::string, ceph::buffer::list> attrs;
+  attrs.emplace(OSD_SUPERBLOCK_OMAP_KEY, bl);
+  t.omap_setkeys(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, attrs);
+}
+
 int OSD::mkfs(CephContext *cct,
 	      std::unique_ptr<ObjectStore> store,
 	      uuid_d fsid,
@@ -2153,15 +2187,11 @@ int OSD::mkfs(CephContext *cct,
     sb.osd_fsid = store->get_fsid();
     sb.whoami = whoami;
     sb.compat_features = get_osd_initial_compat_set();
-
-    bufferlist bl;
-    encode(sb, bl);
-
     ObjectStore::CollectionHandle ch = store->create_new_collection(
       coll_t::meta());
     ObjectStore::Transaction t;
     t.create_collection(coll_t::meta(), 0);
-    t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl);
+    write_superblock(cct, sb, t);
     ret = store->queue_transaction(ch, std::move(t));
     if (ret) {
       derr << "OSD::mkfs: error while writing OSD_SUPERBLOCK_GOBJECT: "
@@ -2385,13 +2415,55 @@ OSD::OSD(CephContext *cct_,
   trace_endpoint.copy_name(ss.str());
 #endif
 
+  // Determine scheduler type for this OSD
+  auto get_op_queue_type = [this, &conf = cct->_conf]() {
+    op_queue_type_t queue_type;
+    if (auto type = conf.get_val<std::string>("osd_op_queue");
+        type != "debug_random") {
+      if (auto qt = get_op_queue_type_by_name(type); qt.has_value()) {
+        queue_type = *qt;
+      } else {
+        // This should never happen
+        dout(0) << "Invalid value passed for 'osd_op_queue': " << type << dendl;
+        ceph_assert(0 == "Unsupported op queue type");
+      }
+    } else {
+      static const std::vector<op_queue_type_t> index_lookup = {
+        op_queue_type_t::mClockScheduler,
+        op_queue_type_t::WeightedPriorityQueue
+      };
+      std::mt19937 random_gen(std::random_device{}());
+      auto which = random_gen() % index_lookup.size();
+      queue_type = index_lookup[which];
+    }
+    return queue_type;
+  };
+  op_queue_type_t op_queue = get_op_queue_type();
+
+  // Determine op queue cutoff
+  auto get_op_queue_cut_off = [&conf = cct->_conf]() {
+    if (conf.get_val<std::string>("osd_op_queue_cut_off") == "debug_random") {
+      std::random_device rd;
+      std::mt19937 random_gen(rd());
+      return (random_gen() % 2 < 1) ? CEPH_MSG_PRIO_HIGH : CEPH_MSG_PRIO_LOW;
+    } else if (conf.get_val<std::string>("osd_op_queue_cut_off") == "high") {
+      return CEPH_MSG_PRIO_HIGH;
+    } else {
+      // default / catch-all is 'low'
+      return CEPH_MSG_PRIO_LOW;
+    }
+  };
+  unsigned op_queue_cut_off = get_op_queue_cut_off();
+
   // initialize shards
   num_shards = get_num_op_shards();
   for (uint32_t i = 0; i < num_shards; i++) {
     OSDShard *one_shard = new OSDShard(
       i,
       cct,
-      this);
+      this,
+      op_queue,
+      op_queue_cut_off);
     shards.push_back(one_shard);
   }
 }
@@ -3121,6 +3193,19 @@ will start to track new ops received afterwards.";
     scrub_purged_snaps();
   }
 
+  else if (prefix == "reset_purged_snaps_last") {
+    lock_guard l(osd_lock);
+    superblock.purged_snaps_last = 0;
+    ObjectStore::Transaction t;
+    dout(10) << __func__ << " updating superblock" << dendl;
+    write_superblock(cct, superblock, t);
+    ret = store->queue_transaction(service.meta_ch, std::move(t), nullptr);
+    if (ret < 0) {
+      ss << "Error writing superblock: " << cpp_strerror(ret);
+      goto out;
+    }
+  }
+
   else if (prefix == "dump_osd_network") {
     lock_guard l(osd_lock);
     int64_t value = 0;
@@ -3762,7 +3847,7 @@ int OSD::init()
     }
 
     ObjectStore::Transaction t;
-    write_superblock(t);
+    write_superblock(cct, superblock, t);
     r = store->queue_transaction(service.meta_ch, std::move(t));
     if (r < 0)
       goto out;
@@ -4300,6 +4385,11 @@ void OSD::final_init()
     "Scrub purged_snaps vs snapmapper index");
   ceph_assert(r == 0);
   r = admin_socket->register_command(
+    "reset_purged_snaps_last",
+    asok_hook,
+    "Reset the superblock's purged_snaps_last");
+  ceph_assert(r == 0);
+  r = admin_socket->register_command(
     "scrubdebug "						\
     "name=pgid,type=CephPgid "	                                \
     "name=cmd,type=CephChoices,strings=block|unblock|set|unset " \
@@ -4572,7 +4662,7 @@ int OSD::shutdown()
   superblock.mounted = service.get_boot_epoch();
   superblock.clean_thru = get_osdmap_epoch();
   ObjectStore::Transaction t;
-  write_superblock(t);
+  write_superblock(cct, superblock, t);
   int r = store->queue_transaction(service.meta_ch, std::move(t));
   if (r) {
     derr << "OSD::shutdown: error writing superblock: "
@@ -4769,31 +4859,81 @@ int OSD::update_crush_device_class()
   }
 }
 
-void OSD::write_superblock(ObjectStore::Transaction& t)
-{
-  dout(10) << "write_superblock " << superblock << dendl;
-
-  //hack: at minimum it's using the baseline feature set
-  if (!superblock.compat_features.incompat.contains(CEPH_OSD_FEATURE_INCOMPAT_BASE))
-    superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_BASE);
-
-  bufferlist bl;
-  encode(superblock, bl);
-  t.write(coll_t::meta(), OSD_SUPERBLOCK_GOBJECT, 0, bl.length(), bl);
-}
 
 int OSD::read_superblock()
 {
+  // Read superblock from both object data and omap metadata
+  // for better robustness.
+  // Use the most recent superblock replica if obtained versions
+  // mismatch.
   bufferlist bl;
-  int r = store->read(service.meta_ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, bl);
-  if (r < 0)
-    return r;
 
-  auto p = bl.cbegin();
-  decode(superblock, p);
+  set<string> keys;
+  keys.insert(OSD_SUPERBLOCK_OMAP_KEY);
+  map<string, bufferlist> vals;
+  OSDSuperblock super_omap;
+  OSDSuperblock super_disk;
+  int r_omap = store->omap_get_values(
+    service.meta_ch, OSD_SUPERBLOCK_GOBJECT, keys, &vals);
+  if (r_omap >= 0 && vals.size() > 0) {
+    try {
+      auto p = vals.begin()->second.cbegin();
+      decode(super_omap, p);
+    } catch(...) {
+      derr << __func__ << " omap replica is corrupted."
+            << dendl;
+      r_omap = -EFAULT;
+    }
+  } else {
+    derr << __func__ << " omap replica is missing."
+         << dendl;
+    r_omap = -ENOENT;
+  }
+  int r_disk = store->read(service.meta_ch, OSD_SUPERBLOCK_GOBJECT, 0, 0, bl);
+  if (r_disk >= 0) {
+    try {
+      auto p = bl.cbegin();
+      decode(super_disk, p);
+    } catch(...) {
+      derr << __func__ << " disk replica is corrupted."
+            << dendl;
+      r_disk = -EFAULT;
+    }
+  } else {
+    derr << __func__ << " disk replica is missing."
+         << dendl;
+    r_disk = -ENOENT;
+  }
 
-  dout(10) << "read_superblock " << superblock << dendl;
+  if (r_omap >= 0 && r_disk < 0) {
+    std::swap(superblock, super_omap);
+    dout(1) << __func__ << " got omap replica but failed to get disk one."
+            << dendl;
+  } else if (r_omap < 0 && r_disk >= 0) {
+    std::swap(superblock, super_disk);
+    dout(1) << __func__ << " got disk replica but failed to get omap one."
+            << dendl;
+  } else if (r_omap < 0 && r_disk < 0) {
+    // error to be logged by the caller
+    return -ENOENT;
+  } else {
+    std::swap(superblock, super_omap); // let omap be the primary source
+    if (superblock.current_epoch != super_disk.current_epoch) {
+      derr << __func__ << " got mismatching superblocks, omap:"
+           << superblock << " vs. disk:" << super_disk
+           << dendl;
+      if (superblock.current_epoch < super_disk.current_epoch) {
+        std::swap(superblock, super_disk);
+        dout(0) << __func__ << " using disk superblock"
+                << dendl;
+      } else {
+        dout(0) << __func__ << " using omap superblock"
+                << dendl;
+      }
+    }
+  }
 
+  dout(10) << "read_superblock " << superblock << dendl;
   return 0;
 }
 
@@ -6695,7 +6835,7 @@ void OSD::handle_get_purged_snaps_reply(MMonGetPurgedSnapsReply *m)
       m->purged_snaps);
   }
   superblock.purged_snaps_last = m->last;
-  write_superblock(t);
+  write_superblock(cct, superblock, t);
   store->queue_transaction(
     service.meta_ch,
     std::move(t));
@@ -7179,7 +7319,7 @@ void OSD::scrub_purged_snaps()
   dout(10) << __func__ << " done queueing pgs, updating superblock" << dendl;
   ObjectStore::Transaction t;
   superblock.last_purged_snaps_scrub = ceph_clock_now();
-  write_superblock(t);
+  write_superblock(cct, superblock, t);
   int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr);
   ceph_assert(tr == 0);
   if (is_active()) {
@@ -7892,7 +8032,7 @@ void OSD::trim_maps(epoch_t oldest, int nreceived, bool skip_maps)
     num++;
     if (num >= cct->_conf->osd_target_transaction_size && num >= nreceived) {
       service.publish_superblock(superblock);
-      write_superblock(t);
+      write_superblock(cct, superblock, t);
       int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr);
       ceph_assert(tr == 0);
       num = 0;
@@ -7908,7 +8048,7 @@ void OSD::trim_maps(epoch_t oldest, int nreceived, bool skip_maps)
   }
   if (num > 0) {
     service.publish_superblock(superblock);
-    write_superblock(t);
+    write_superblock(cct, superblock, t);
     int tr = store->queue_transaction(service.meta_ch, std::move(t), nullptr);
     ceph_assert(tr == 0);
   }
@@ -8220,7 +8360,19 @@ void OSD::handle_osd_map(MOSDMap *m)
   {
     bufferlist bl;
     ::encode(pg_num_history, bl);
-    t.write(coll_t::meta(), make_pg_num_history_oid(), 0, bl.length(), bl);
+    auto oid = make_pg_num_history_oid();
+    t.truncate(coll_t::meta(), oid, 0); // we don't need bytes left if new data
+                                        // block is shorter than the previous
+                                        // one. And better to trim them, e.g.
+                                        // this allows to avoid csum eroors
+                                        // when issuing overwrite
+                                        // (which happens to be partial)
+                                        // and original data is corrupted.
+                                        // Another side effect is that the
+                                        // superblock is not permanently
+                                        // anchored to a fixed disk location
+                                        // any more.
+    t.write(coll_t::meta(), oid, 0, bl.length(), bl);
     dout(20) << __func__ << " pg_num_history " << pg_num_history << dendl;
   }
 
@@ -8240,7 +8392,7 @@ void OSD::handle_osd_map(MOSDMap *m)
   }
 
   // superblock and commit
-  write_superblock(t);
+  write_superblock(cct, superblock, t);
   t.register_on_commit(new C_OnMapCommit(this, start, last, m));
   store->queue_transaction(
     service.meta_ch,
@@ -8558,7 +8710,7 @@ void OSD::check_osdmap_features()
       dout(0) << __func__ << " enabling on-disk ERASURE CODES compat feature" << dendl;
       superblock.compat_features.incompat.insert(CEPH_OSD_FEATURE_INCOMPAT_SHARDS);
       ObjectStore::Transaction t;
-      write_superblock(t);
+      write_superblock(cct, superblock, t);
       int err = store->queue_transaction(service.meta_ch, std::move(t), NULL);
       ceph_assert(err == 0);
     }
@@ -9888,7 +10040,7 @@ void OSD::maybe_override_max_osd_capacity_for_qos()
   // If the scheduler enabled is mclock, override the default
   // osd capacity with the value obtained from running the
   // osd bench test. This is later used to setup mclock.
-  if ((cct->_conf.get_val<std::string>("osd_op_queue") == "mclock_scheduler") &&
+  if ((op_queue_type_t::mClockScheduler == osd_op_queue_type()) &&
       (cct->_conf.get_val<bool>("osd_mclock_skip_benchmark") == false) &&
       (!unsupported_objstore_for_qos())) {
     std::string max_capacity_iops_config;
@@ -9988,7 +10140,7 @@ bool OSD::maybe_override_options_for_qos(const std::set<std::string> *changed)
 {
   // Override options only if the scheduler enabled is mclock and the
   // underlying objectstore is supported by mclock
-  if (cct->_conf.get_val<std::string>("osd_op_queue") == "mclock_scheduler" &&
+  if (op_queue_type_t::mClockScheduler == osd_op_queue_type() &&
       !unsupported_objstore_for_qos()) {
     static const std::map<std::string, uint64_t> recovery_qos_defaults {
       {"osd_recovery_max_active", 0},
@@ -10090,9 +10242,8 @@ void OSD::maybe_override_sleep_options_for_qos()
 {
   // Override options only if the scheduler enabled is mclock and the
   // underlying objectstore is supported by mclock
-  if (cct->_conf.get_val<std::string>("osd_op_queue") == "mclock_scheduler" &&
+  if (op_queue_type_t::mClockScheduler == osd_op_queue_type() &&
       !unsupported_objstore_for_qos()) {
-
     // Override the various sleep settings
     // Disable recovery sleep
     cct->_conf.set_val("osd_recovery_sleep", std::to_string(0));
@@ -10121,7 +10272,7 @@ void OSD::maybe_override_cost_for_qos()
 {
   // If the scheduler enabled is mclock, override the default PG deletion cost
   // so that mclock can meet the QoS goals.
-  if (cct->_conf.get_val<std::string>("osd_op_queue") == "mclock_scheduler" &&
+  if (op_queue_type_t::mClockScheduler == osd_op_queue_type() &&
       !unsupported_objstore_for_qos()) {
     uint64_t pg_delete_cost = 15728640;
     cct->_conf.set_val("osd_pg_delete_cost", std::to_string(pg_delete_cost));
@@ -10195,6 +10346,16 @@ bool OSD::unsupported_objstore_for_qos()
                    store->get_type()) != unsupported_objstores.end();
 }
 
+op_queue_type_t OSD::osd_op_queue_type() const
+{
+  /**
+   * All OSD shards employ the same scheduler type. Therefore, return
+   * the scheduler type set on the OSD shard with lowest id(0).
+   */
+  ceph_assert(shards.size());
+  return shards[0]->get_op_queue_type();
+}
+
 void OSD::update_log_config()
 {
   auto parsed_options = clog->parse_client_options(cct);
@@ -10695,17 +10856,17 @@ void OSDShard::update_scheduler_config()
   scheduler->update_configuration();
 }
 
-std::string OSDShard::get_scheduler_type()
+op_queue_type_t OSDShard::get_op_queue_type() const
 {
-  std::ostringstream scheduler_type;
-  scheduler_type << *scheduler;
-  return scheduler_type.str();
+  return scheduler->get_type();
 }
 
 OSDShard::OSDShard(
   int id,
   CephContext *cct,
-  OSD *osd)
+  OSD *osd,
+  op_queue_type_t osd_op_queue,
+  unsigned osd_op_queue_cut_off)
   : shard_id(id),
     cct(cct),
     osd(osd),
@@ -10717,7 +10878,7 @@ OSDShard::OSDShard(
     shard_lock{make_mutex(shard_lock_name)},
     scheduler(ceph::osd::scheduler::make_scheduler(
       cct, osd->whoami, osd->num_shards, id, osd->store->is_rotational(),
-      osd->store->get_type(), osd->monc)),
+      osd->store->get_type(), osd_op_queue, osd_op_queue_cut_off, osd->monc)),
     context_queue(sdata_wait_lock, sdata_cond)
 {
   dout(0) << "using op scheduler " << *scheduler << dendl;