#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include namespace fs = std::filesystem; namespace ProfileEvents { extern const Event ReplicatedPartFailedFetches; extern const Event ReplicatedPartFetchesOfMerged; extern const Event ObsoleteReplicatedParts; extern const Event ReplicatedPartFetches; extern const Event CreatedLogEntryForMerge; extern const Event NotCreatedLogEntryForMerge; extern const Event CreatedLogEntryForMutation; extern const Event NotCreatedLogEntryForMutation; extern const Event ReplicaPartialShutdown; } namespace CurrentMetrics { extern const Metric BackgroundFetchesPoolTask; } namespace DB { namespace ErrorCodes { extern const int CANNOT_READ_ALL_DATA; extern const int NOT_IMPLEMENTED; extern const int NO_ZOOKEEPER; extern const int INCORRECT_DATA; extern const int INCOMPATIBLE_COLUMNS; extern const int REPLICA_ALREADY_EXISTS; extern const int NO_REPLICA_HAS_PART; extern const int LOGICAL_ERROR; extern const int TOO_MANY_UNEXPECTED_DATA_PARTS; extern const int ABORTED; extern const int REPLICA_IS_NOT_IN_QUORUM; extern const int TABLE_IS_READ_ONLY; extern const int NOT_FOUND_NODE; extern const int NO_ACTIVE_REPLICAS; extern const int NOT_A_LEADER; extern const int TABLE_WAS_NOT_DROPPED; extern const int PARTITION_ALREADY_EXISTS; extern const int TOO_MANY_RETRIES_TO_FETCH_PARTS; extern const int RECEIVED_ERROR_FROM_REMOTE_IO_SERVER; extern const int PARTITION_DOESNT_EXIST; extern const int UNFINISHED; extern const int RECEIVED_ERROR_TOO_MANY_REQUESTS; extern const int PART_IS_TEMPORARILY_LOCKED; extern const int CANNOT_ASSIGN_OPTIMIZE; extern const int ALL_REPLICAS_LOST; extern const int REPLICA_STATUS_CHANGED; extern const int CANNOT_ASSIGN_ALTER; extern const int DIRECTORY_ALREADY_EXISTS; extern const int ILLEGAL_TYPE_OF_ARGUMENT; extern const int UNKNOWN_POLICY; extern const int NO_SUCH_DATA_PART; extern const int INTERSERVER_SCHEME_DOESNT_MATCH; extern const int DUPLICATE_DATA_PART; extern const int BAD_ARGUMENTS; extern const int CONCURRENT_ACCESS_NOT_SUPPORTED; extern const int CHECKSUM_DOESNT_MATCH; extern const int NOT_INITIALIZED; extern const int TOO_LARGE_DISTRIBUTED_DEPTH; } namespace ActionLocks { extern const StorageActionBlockType PartsMerge; extern const StorageActionBlockType PartsFetch; extern const StorageActionBlockType PartsSend; extern const StorageActionBlockType ReplicationQueue; extern const StorageActionBlockType PartsTTLMerge; extern const StorageActionBlockType PartsMove; } static const auto QUEUE_UPDATE_ERROR_SLEEP_MS = 1 * 1000; static const auto MUTATIONS_FINALIZING_SLEEP_MS = 1 * 1000; static const auto MUTATIONS_FINALIZING_IDLE_SLEEP_MS = 5 * 1000; const String StorageReplicatedMergeTree::default_zookeeper_name = "default"; void StorageReplicatedMergeTree::setZooKeeper() { /// Every ReplicatedMergeTree table is using only one ZooKeeper session. /// But if several ReplicatedMergeTree tables are using different /// ZooKeeper sessions, some queries like ATTACH PARTITION FROM may have /// strange effects. So we always use only one session for all tables. /// (excluding auxiliary zookeepers) std::lock_guard lock(current_zookeeper_mutex); if (zookeeper_name == default_zookeeper_name) { current_zookeeper = getContext()->getZooKeeper(); } else { current_zookeeper = getContext()->getAuxiliaryZooKeeper(zookeeper_name); } } zkutil::ZooKeeperPtr StorageReplicatedMergeTree::tryGetZooKeeper() const { std::lock_guard lock(current_zookeeper_mutex); return current_zookeeper; } zkutil::ZooKeeperPtr StorageReplicatedMergeTree::getZooKeeper() const { auto res = tryGetZooKeeper(); if (!res) throw Exception(ErrorCodes::NO_ZOOKEEPER, "Cannot get ZooKeeper"); return res; } zkutil::ZooKeeperPtr StorageReplicatedMergeTree::getZooKeeperAndAssertNotReadonly() const { /// There's a short period of time after connection loss when new session is created, /// but replication queue is not reinitialized. We must ensure that table is not readonly anymore /// before using new ZooKeeper session to write something (except maybe GET_PART) into replication log. auto res = getZooKeeper(); assertNotReadonly(); return res; } static ConnectionTimeouts getHTTPTimeouts(ContextPtr context) { return ConnectionTimeouts::getHTTPTimeouts(context->getSettingsRef(), {context->getConfigRef().getUInt("keep_alive_timeout", DEFAULT_HTTP_KEEP_ALIVE_TIMEOUT), 0}); } static MergeTreePartInfo makeDummyDropRangeForMovePartitionOrAttachPartitionFrom(const String & partition_id) { /// NOTE We don't have special log entry type for MOVE PARTITION/ATTACH PARTITION FROM, /// so we use REPLACE_RANGE with dummy range of one block, which means "attach, not replace". /// It's safe to fill drop range for MOVE PARTITION/ATTACH PARTITION FROM with zeros, /// because drop range for REPLACE PARTITION must contain at least 2 blocks, /// so we can distinguish dummy drop range from any real or virtual part. /// But we should never construct such part name, even for virtual part, /// because it can be confused with real part _0_0_0. /// TODO get rid of this. MergeTreePartInfo drop_range; drop_range.partition_id = partition_id; drop_range.min_block = 0; drop_range.max_block = 0; drop_range.level = 0; drop_range.mutation = 0; return drop_range; } StorageReplicatedMergeTree::StorageReplicatedMergeTree( const String & zookeeper_path_, const String & replica_name_, bool attach, const StorageID & table_id_, const String & relative_data_path_, const StorageInMemoryMetadata & metadata_, ContextMutablePtr context_, const String & date_column_name, const MergingParams & merging_params_, std::unique_ptr settings_, bool has_force_restore_data_flag, RenamingRestrictions renaming_restrictions_) : MergeTreeData(table_id_, metadata_, context_, date_column_name, merging_params_, std::move(settings_), true, /// require_part_metadata attach, [this] (const std::string & name) { enqueuePartForCheck(name); }) , zookeeper_name(zkutil::extractZooKeeperName(zookeeper_path_)) , zookeeper_path(zkutil::extractZooKeeperPath(zookeeper_path_, /* check_starts_with_slash */ !attach, log)) , replica_name(replica_name_) , replica_path(fs::path(zookeeper_path) / "replicas" / replica_name_) , reader(*this) , writer(*this) , merger_mutator(*this) , merge_strategy_picker(*this) , queue(*this, merge_strategy_picker) , fetcher(*this) , cleanup_thread(*this) , async_block_ids_cache(*this) , part_check_thread(*this) , restarting_thread(*this) , part_moves_between_shards_orchestrator(*this) , renaming_restrictions(renaming_restrictions_) , replicated_fetches_throttler(std::make_shared(getSettings()->max_replicated_fetches_network_bandwidth, getContext()->getReplicatedFetchesThrottler())) , replicated_sends_throttler(std::make_shared(getSettings()->max_replicated_sends_network_bandwidth, getContext()->getReplicatedSendsThrottler())) { initializeDirectoriesAndFormatVersion(relative_data_path_, attach, date_column_name); /// We create and deactivate all tasks for consistency. /// They all will be scheduled and activated by the restarting thread. queue_updating_task = getContext()->getSchedulePool().createTask( getStorageID().getFullTableName() + " (StorageReplicatedMergeTree::queueUpdatingTask)", [this]{ queueUpdatingTask(); }); queue_updating_task->deactivate(); mutations_updating_task = getContext()->getSchedulePool().createTask( getStorageID().getFullTableName() + " (StorageReplicatedMergeTree::mutationsUpdatingTask)", [this]{ mutationsUpdatingTask(); }); mutations_updating_task->deactivate(); merge_selecting_task = getContext()->getSchedulePool().createTask( getStorageID().getFullTableName() + " (StorageReplicatedMergeTree::mergeSelectingTask)", [this] { mergeSelectingTask(); }); /// Will be activated if we will achieve leader state. merge_selecting_task->deactivate(); mutations_finalizing_task = getContext()->getSchedulePool().createTask( getStorageID().getFullTableName() + " (StorageReplicatedMergeTree::mutationsFinalizingTask)", [this] { mutationsFinalizingTask(); }); /// This task can be scheduled by different parts of code even when storage is readonly. /// This can lead to redundant exceptions during startup. /// Will be activated by restarting thread. mutations_finalizing_task->deactivate(); bool has_zookeeper = getContext()->hasZooKeeper() || getContext()->hasAuxiliaryZooKeeper(zookeeper_name); if (has_zookeeper) { /// It's possible for getZooKeeper() to timeout if zookeeper host(s) can't /// be reached. In such cases Poco::Exception is thrown after a connection /// timeout - refer to src/Common/ZooKeeper/ZooKeeperImpl.cpp:866 for more info. /// /// Side effect of this is that the CreateQuery gets interrupted and it exits. /// But the data Directories for the tables being created aren't cleaned up. /// This unclean state will hinder table creation on any retries and will /// complain that the Directory for table already exists. /// /// To achieve a clean state on failed table creations, catch this error and /// call dropIfEmpty() method only if the operation isn't ATTACH then proceed /// throwing the exception. Without this, the Directory for the tables need /// to be manually deleted before retrying the CreateQuery. try { if (zookeeper_name == default_zookeeper_name) { current_zookeeper = getContext()->getZooKeeper(); } else { current_zookeeper = getContext()->getAuxiliaryZooKeeper(zookeeper_name); } } catch (...) { if (!attach) { dropIfEmpty(); throw; } else { current_zookeeper = nullptr; } } } bool skip_sanity_checks = false; /// It does not make sense for CREATE query if (attach) { try { if (current_zookeeper && current_zookeeper->exists(replica_path + "/host")) { /// Check it earlier if we can (we don't want incompatible version to start). /// If "/host" doesn't exist, then replica is probably dropped and there's nothing to check. ReplicatedMergeTreeAttachThread::checkHasReplicaMetadataInZooKeeper(current_zookeeper, replica_path); } if (current_zookeeper && current_zookeeper->exists(replica_path + "/flags/force_restore_data")) { skip_sanity_checks = true; current_zookeeper->remove(replica_path + "/flags/force_restore_data"); LOG_WARNING( log, "Skipping the limits on severity of changes to data parts and columns (flag {}/flags/force_restore_data).", replica_path); } else if (has_force_restore_data_flag) { skip_sanity_checks = true; LOG_WARNING(log, "Skipping the limits on severity of changes to data parts and columns (flag force_restore_data)."); } } catch (const Coordination::Exception & e) { if (!Coordination::isHardwareError(e.code)) throw; LOG_ERROR(log, "Caught exception while checking table metadata in ZooKeeper, will recheck later: {}", e.displayText()); } } loadDataParts(skip_sanity_checks); if (!current_zookeeper) { if (!attach) { dropIfEmpty(); throw Exception(ErrorCodes::NO_ZOOKEEPER, "Can't create replicated table without ZooKeeper"); } has_metadata_in_zookeeper = std::nullopt; if (!has_zookeeper) { /// Do not activate the replica. It will be readonly. LOG_ERROR(log, "No ZooKeeper defined: table will stay in readonly mode."); return; } } if (attach) { LOG_INFO(log, "Table will be in readonly mode until initialization is finished"); attach_thread.emplace(*this); attach_thread->setSkipSanityChecks(skip_sanity_checks); return; } auto metadata_snapshot = getInMemoryMetadataPtr(); has_metadata_in_zookeeper = true; if (!getDataPartsForInternalUsage().empty()) throw Exception(ErrorCodes::INCORRECT_DATA, "Data directory for table already contains data parts - probably it was unclean DROP table " "or manual intervention. You must either clear directory by hand " "or use ATTACH TABLE instead of CREATE TABLE if you need to use that parts."); try { bool is_first_replica = createTableIfNotExists(metadata_snapshot); try { /// NOTE If it's the first replica, these requests to ZooKeeper look redundant, we already know everything. /// We have to check granularity on other replicas. If it's fixed we /// must create our new replica with fixed granularity and store this /// information in /replica/metadata. other_replicas_fixed_granularity = checkFixedGranularityInZookeeper(); checkTableStructure(zookeeper_path, metadata_snapshot); Coordination::Stat metadata_stat; current_zookeeper->get(zookeeper_path + "/metadata", &metadata_stat); metadata_version = metadata_stat.version; } catch (Coordination::Exception & e) { if (!is_first_replica && e.code == Coordination::Error::ZNONODE) throw Exception(ErrorCodes::ALL_REPLICAS_LOST, "Table {} was suddenly removed.", zookeeper_path); else throw; } if (!is_first_replica) createReplica(metadata_snapshot); } catch (...) { /// If replica was not created, rollback creation of data directory. dropIfEmpty(); throw; } createNewZooKeeperNodes(); syncPinnedPartUUIDs(); if (!has_metadata_in_zookeeper.has_value() || *has_metadata_in_zookeeper) createTableSharedID(); initialization_done = true; } String StorageReplicatedMergeTree::getDefaultZooKeeperPath(const Poco::Util::AbstractConfiguration & config) { return config.getString("default_replica_path", "/clickhouse/tables/{uuid}/{shard}"); } String StorageReplicatedMergeTree::getDefaultReplicaName(const Poco::Util::AbstractConfiguration & config) { return config.getString("default_replica_name", "{replica}"); } bool StorageReplicatedMergeTree::checkFixedGranularityInZookeeper() { auto zookeeper = getZooKeeper(); String metadata_str = zookeeper->get(zookeeper_path + "/metadata"); auto metadata_from_zk = ReplicatedMergeTreeTableMetadata::parse(metadata_str); return metadata_from_zk.index_granularity_bytes == 0; } void StorageReplicatedMergeTree::waitMutationToFinishOnReplicas( const Strings & replicas, const String & mutation_id) const { if (replicas.empty()) return; /// Current replica must always be present in the list as the first element because we use local mutation status /// to check for mutation errors. So if it is not there, just add it. const Strings * all_required_replicas = &replicas; Strings extended_list_of_replicas; if (replicas.front() != replica_name) { extended_list_of_replicas.push_back(replica_name); extended_list_of_replicas.insert(extended_list_of_replicas.end(), replicas.begin(), replicas.end()); all_required_replicas = &extended_list_of_replicas; } std::set inactive_replicas; for (const String & replica : *all_required_replicas) { LOG_DEBUG(log, "Waiting for {} to apply mutation {}", replica, mutation_id); zkutil::EventPtr wait_event = std::make_shared(); while (!partial_shutdown_called) { /// Mutation maybe killed or whole replica was deleted. /// Wait event will unblock at this moment. Coordination::Stat exists_stat; if (!getZooKeeper()->exists(fs::path(zookeeper_path) / "mutations" / mutation_id, &exists_stat, wait_event)) { throw Exception(ErrorCodes::UNFINISHED, "Mutation {} was killed, manually removed or table was dropped", mutation_id); } auto zookeeper = getZooKeeper(); /// Replica could be inactive. if (!zookeeper->exists(fs::path(zookeeper_path) / "replicas" / replica / "is_active")) { LOG_WARNING(log, "Replica {} is not active during mutation. Mutation will be done asynchronously when replica becomes active.", replica); inactive_replicas.emplace(replica); break; } String mutation_pointer = fs::path(zookeeper_path) / "replicas" / replica / "mutation_pointer"; std::string mutation_pointer_value; /// Replica could be removed if (!zookeeper->tryGet(mutation_pointer, mutation_pointer_value, nullptr, wait_event)) { LOG_WARNING(log, "Replica {} was removed", replica); break; } else if (mutation_pointer_value >= mutation_id) /// Maybe we already processed more fresh mutation break; /// (numbers like 0000000000 and 0000000001) /// Replica can become inactive, so wait with timeout and recheck it if (wait_event->tryWait(1000)) continue; /// Here we check mutation for errors on local replica. If they happen on this replica /// they will happen on each replica, so we can check only in-memory info. auto mutation_status = queue.getIncompleteMutationsStatus(mutation_id); /// If mutation status is empty, than local replica may just not loaded it into memory. if (mutation_status && !mutation_status->latest_fail_reason.empty()) break; } /// This replica inactive, don't check anything if (!inactive_replicas.empty() && inactive_replicas.contains(replica)) break; /// It maybe already removed from zk, but local in-memory mutations /// state was not updated. if (!getZooKeeper()->exists(fs::path(zookeeper_path) / "mutations" / mutation_id)) { throw Exception(ErrorCodes::UNFINISHED, "Mutation {} was killed, manually removed or table was dropped", mutation_id); } if (partial_shutdown_called) throw Exception(ErrorCodes::UNFINISHED, "Mutation is not finished because table shutdown was called. " "It will be done after table restart."); /// Replica inactive, don't check mutation status if (!inactive_replicas.empty() && inactive_replicas.contains(replica)) continue; /// At least we have our current mutation std::set mutation_ids; mutation_ids.insert(mutation_id); /// Here we check mutation for errors or kill on local replica. If they happen on this replica /// they will happen on each replica, so we can check only in-memory info. auto mutation_status = queue.getIncompleteMutationsStatus(mutation_id, &mutation_ids); checkMutationStatus(mutation_status, mutation_ids); } if (!inactive_replicas.empty()) { throw Exception(ErrorCodes::UNFINISHED, "Mutation is not finished because some replicas are inactive right now: {}. Mutation will be done asynchronously", boost::algorithm::join(inactive_replicas, ", ")); } } void StorageReplicatedMergeTree::createNewZooKeeperNodes() { auto zookeeper = getZooKeeper(); std::vector futures; /// These 4 nodes used to be created in createNewZookeeperNodes() and they were moved to createTable() /// This means that if the first replica creating the table metadata has an older version of CH (22.3 or previous) /// there will be a time between its calls to `createTable` and `createNewZookeeperNodes` where the nodes won't exists /// and that will cause issues in newer replicas /// See https://github.com/ClickHouse/ClickHouse/issues/38600 for example futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/quorum", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/quorum/last_part", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/quorum/failed_parts", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/mutations", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/quorum/parallel", String(), zkutil::CreateMode::Persistent)); /// Nodes for remote fs zero-copy replication const auto settings = getSettings(); if (settings->allow_remote_fs_zero_copy_replication) { futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/zero_copy_s3", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/zero_copy_s3/shared", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/zero_copy_hdfs", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/zero_copy_hdfs/shared", String(), zkutil::CreateMode::Persistent)); } /// Part movement. futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/part_moves_shard", String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/pinned_part_uuids", getPinnedPartUUIDs()->toString(), zkutil::CreateMode::Persistent)); /// For ALTER PARTITION with multi-leaders futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/alter_partition_version", String(), zkutil::CreateMode::Persistent)); /// For deduplication of async inserts futures.push_back(zookeeper->asyncTryCreateNoThrow(zookeeper_path + "/async_blocks", String(), zkutil::CreateMode::Persistent)); /// As for now, "/temp" node must exist, but we want to be able to remove it in future if (zookeeper->exists(zookeeper_path + "/temp")) { /// For block numbers allocation (since 22.11) futures.push_back(zookeeper->asyncTryCreateNoThrow( zookeeper_path + "/temp/" + EphemeralLockInZooKeeper::LEGACY_LOCK_INSERT, String(), zkutil::CreateMode::Persistent)); futures.push_back(zookeeper->asyncTryCreateNoThrow( zookeeper_path + "/temp/" + EphemeralLockInZooKeeper::LEGACY_LOCK_OTHER, String(), zkutil::CreateMode::Persistent)); } for (auto & future : futures) { auto res = future.get(); if (res.error != Coordination::Error::ZOK && res.error != Coordination::Error::ZNODEEXISTS) throw Coordination::Exception(fmt::format("Failed to create new nodes at {}", zookeeper_path), res.error); } } bool StorageReplicatedMergeTree::createTableIfNotExists(const StorageMetadataPtr & metadata_snapshot) { auto zookeeper = getZooKeeper(); zookeeper->createAncestors(zookeeper_path); for (size_t i = 0; i < 1000; ++i) { /// Invariant: "replicas" does not exist if there is no table or if there are leftovers from incompletely dropped table. if (zookeeper->exists(zookeeper_path + "/replicas")) { LOG_DEBUG(log, "This table {} is already created, will add new replica", zookeeper_path); return false; } /// There are leftovers from incompletely dropped table. if (zookeeper->exists(zookeeper_path + "/dropped")) { /// This condition may happen when the previous drop attempt was not completed /// or when table is dropped by another replica right now. /// This is Ok because another replica is definitely going to drop the table. LOG_WARNING(log, "Removing leftovers from table {} (this might take several minutes)", zookeeper_path); String drop_lock_path = zookeeper_path + "/dropped/lock"; Coordination::Error code = zookeeper->tryCreate(drop_lock_path, "", zkutil::CreateMode::Ephemeral); if (code == Coordination::Error::ZNONODE || code == Coordination::Error::ZNODEEXISTS) { LOG_WARNING(log, "The leftovers from table {} were removed by another replica", zookeeper_path); } else if (code != Coordination::Error::ZOK) { throw Coordination::Exception(code, drop_lock_path); } else { auto metadata_drop_lock = zkutil::EphemeralNodeHolder::existing(drop_lock_path, *zookeeper); if (!removeTableNodesFromZooKeeper(zookeeper, zookeeper_path, metadata_drop_lock, log)) { /// Someone is recursively removing table right now, we cannot create new table until old one is removed continue; } } } LOG_DEBUG(log, "Creating table {}", zookeeper_path); /// We write metadata of table so that the replicas can check table parameters with them. String metadata_str = ReplicatedMergeTreeTableMetadata(*this, metadata_snapshot).toString(); Coordination::Requests ops; ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/metadata", metadata_str, zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/columns", metadata_snapshot->getColumns().toString(), zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/log", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/blocks", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/async_blocks", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/block_numbers", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/nonincrement_block_numbers", "", zkutil::CreateMode::Persistent)); /// /nonincrement_block_numbers dir is unused, but is created nonetheless for backwards compatibility. ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/leader_election", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/temp", "", zkutil::CreateMode::Persistent)); /// The following 2 nodes were added in 22.11 ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/temp/" + EphemeralLockInZooKeeper::LEGACY_LOCK_INSERT, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/temp/" + EphemeralLockInZooKeeper::LEGACY_LOCK_OTHER, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/replicas", "last added replica: " + replica_name, zkutil::CreateMode::Persistent)); /// The following 4 nodes were added in version 1.1.xxx, so we create them here, not in createNewZooKeeperNodes() ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/quorum", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/quorum/last_part", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/quorum/failed_parts", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/mutations", "", zkutil::CreateMode::Persistent)); /// And create first replica atomically. See also "createReplica" method that is used to create not the first replicas. ops.emplace_back(zkutil::makeCreateRequest(replica_path, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/host", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/log_pointer", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/queue", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/parts", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/flags", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/is_lost", "0", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/metadata", metadata_str, zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/columns", metadata_snapshot->getColumns().toString(), zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/metadata_version", std::to_string(metadata_version), zkutil::CreateMode::Persistent)); /// The following 3 nodes were added in version 1.1.xxx, so we create them here, not in createNewZooKeeperNodes() ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/min_unprocessed_insert_time", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/max_processed_insert_time", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/mutation_pointer", "", zkutil::CreateMode::Persistent)); Coordination::Responses responses; auto code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZNODEEXISTS) { LOG_INFO(log, "It looks like the table {} was created by another server at the same moment, will retry", zookeeper_path); continue; } else if (code != Coordination::Error::ZOK) { zkutil::KeeperMultiException::check(code, ops, responses); } return true; } /// Do not use LOGICAL_ERROR code, because it may happen if user has specified wrong zookeeper_path throw Exception(ErrorCodes::REPLICA_ALREADY_EXISTS, "Cannot create table, because it is created concurrently every time or because " "of wrong zookeeper_path or because of logical error"); } void StorageReplicatedMergeTree::createReplica(const StorageMetadataPtr & metadata_snapshot) { auto zookeeper = getZooKeeper(); LOG_DEBUG(log, "Creating replica {}", replica_path); Coordination::Error code; do { Coordination::Stat replicas_stat; String replicas_value; if (!zookeeper->tryGet(zookeeper_path + "/replicas", replicas_value, &replicas_stat)) throw Exception(ErrorCodes::ALL_REPLICAS_LOST, "Cannot create a replica of the table {}, because the last replica of the table was dropped right now", zookeeper_path); /// It is not the first replica, we will mark it as "lost", to immediately repair (clone) from existing replica. /// By the way, it's possible that the replica will be first, if all previous replicas were removed concurrently. const String is_lost_value = replicas_stat.numChildren ? "1" : "0"; Coordination::Requests ops; ops.emplace_back(zkutil::makeCreateRequest(replica_path, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/host", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/log_pointer", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/queue", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/parts", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/flags", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/is_lost", is_lost_value, zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/metadata", ReplicatedMergeTreeTableMetadata(*this, metadata_snapshot).toString(), zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/columns", metadata_snapshot->getColumns().toString(), zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/metadata_version", std::to_string(metadata_version), zkutil::CreateMode::Persistent)); /// The following 3 nodes were added in version 1.1.xxx, so we create them here, not in createNewZooKeeperNodes() ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/min_unprocessed_insert_time", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/max_processed_insert_time", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(replica_path + "/mutation_pointer", "", zkutil::CreateMode::Persistent)); /// Check version of /replicas to see if there are any replicas created at the same moment of time. ops.emplace_back(zkutil::makeSetRequest(zookeeper_path + "/replicas", "last added replica: " + replica_name, replicas_stat.version)); Coordination::Responses responses; code = zookeeper->tryMulti(ops, responses); switch (code) { case Coordination::Error::ZNODEEXISTS: throw Exception(ErrorCodes::REPLICA_ALREADY_EXISTS, "Replica {} already exists", replica_path); case Coordination::Error::ZBADVERSION: LOG_INFO(log, "Retrying createReplica(), because some other replicas were created at the same time"); break; case Coordination::Error::ZNONODE: throw Exception(ErrorCodes::ALL_REPLICAS_LOST, "Table {} was suddenly removed", zookeeper_path); default: zkutil::KeeperMultiException::check(code, ops, responses); } } while (code == Coordination::Error::ZBADVERSION); } zkutil::ZooKeeperPtr StorageReplicatedMergeTree::getZooKeeperIfTableShutDown() const { zkutil::ZooKeeperPtr maybe_new_zookeeper; if (zookeeper_name == default_zookeeper_name) maybe_new_zookeeper = getContext()->getZooKeeper(); else maybe_new_zookeeper = getContext()->getAuxiliaryZooKeeper(zookeeper_name); maybe_new_zookeeper->sync(zookeeper_path); return maybe_new_zookeeper; } void StorageReplicatedMergeTree::drop() { /// There is also the case when user has configured ClickHouse to wrong ZooKeeper cluster /// or metadata of staled replica were removed manually, /// in this case, has_metadata_in_zookeeper = false, and we also permit to drop the table. bool maybe_has_metadata_in_zookeeper = !has_metadata_in_zookeeper.has_value() || *has_metadata_in_zookeeper; zkutil::ZooKeeperPtr zookeeper; if (maybe_has_metadata_in_zookeeper) { /// Table can be shut down, restarting thread is not active /// and calling StorageReplicatedMergeTree::getZooKeeper()/getAuxiliaryZooKeeper() won't suffice. zookeeper = getZooKeeperIfTableShutDown(); /// If probably there is metadata in ZooKeeper, we don't allow to drop the table. if (!zookeeper) throw Exception(ErrorCodes::TABLE_IS_READ_ONLY, "Can't drop readonly replicated table (need to drop data in ZooKeeper as well)"); } /// Wait for loading of all outdated parts because /// in case of zero copy recursive removal of directory /// is not supported and table cannot be dropped. if (canUseZeroCopyReplication()) { /// Load remaining parts synchronously because task /// for loading is already cancelled in shutdown(). loadOutdatedDataParts(/*is_async=*/ false); } dropAllData(); if (maybe_has_metadata_in_zookeeper) { /// Session could expire, get it again zookeeper = getZooKeeperIfTableShutDown(); dropReplica(zookeeper, zookeeper_path, replica_name, log, getSettings(), &has_metadata_in_zookeeper); } } void StorageReplicatedMergeTree::dropReplica(zkutil::ZooKeeperPtr zookeeper, const String & zookeeper_path, const String & replica, Poco::Logger * logger, MergeTreeSettingsPtr table_settings, std::optional * has_metadata_out) { if (zookeeper->expired()) throw Exception(ErrorCodes::TABLE_WAS_NOT_DROPPED, "Table was not dropped because ZooKeeper session has expired."); auto remote_replica_path = zookeeper_path + "/replicas/" + replica; LOG_INFO(logger, "Removing replica {}, marking it as lost", remote_replica_path); /// Mark itself lost before removing, because the following recursive removal may fail /// and partially dropped replica may be considered as alive one (until someone will mark it lost) zookeeper->trySet(remote_replica_path + "/is_lost", "1"); /// NOTE: we should check for remote_replica_path existence, /// since otherwise DROP REPLICA will fail if the replica had been already removed. if (!zookeeper->exists(remote_replica_path)) { LOG_INFO(logger, "Removing replica {} does not exist", remote_replica_path); return; } { /// Remove "host" node first to mark replica as dropped (the choice is arbitrary, /// it could be any node without children that exists since ancient server versions and not re-created on startup) [[maybe_unused]] auto code = zookeeper->tryRemove(fs::path(remote_replica_path) / "host"); assert(code == Coordination::Error::ZOK || code == Coordination::Error::ZNONODE); /// Then try to remove paths that are known to be flat (all children are leafs) Strings flat_nodes = {"flags", "queue"}; if (table_settings && table_settings->use_minimalistic_part_header_in_zookeeper) flat_nodes.emplace_back("parts"); for (const auto & node : flat_nodes) { bool removed_quickly = zookeeper->tryRemoveChildrenRecursive(fs::path(remote_replica_path) / node, /* probably flat */ true); if (!removed_quickly) LOG_WARNING(logger, "Failed to quickly remove node '{}' and its children, fell back to recursive removal (replica: {})", node, remote_replica_path); } /// Then try to remove nodes that are known to have no children (and should always exist) Coordination::Requests ops; for (const auto & node : flat_nodes) ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/" + node, -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/columns", -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/is_lost", -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/log_pointer", -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/max_processed_insert_time", -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/min_unprocessed_insert_time", -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/metadata", -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/metadata_version", -1)); ops.emplace_back(zkutil::makeRemoveRequest(remote_replica_path + "/mutation_pointer", -1)); Coordination::Responses res; code = zookeeper->tryMulti(ops, res); if (code != Coordination::Error::ZOK) LOG_WARNING(logger, "Cannot quickly remove nodes without children: {} (replica: {}). Will remove recursively.", Coordination::errorMessage(code), remote_replica_path); /// And finally remove everything else recursively /// It may left some garbage if replica_path subtree is concurrently modified zookeeper->tryRemoveChildrenRecursive(remote_replica_path); /// Update has_metadata_in_zookeeper to avoid retries. Otherwise we can accidentally remove metadata of a new table on retries if (has_metadata_out) *has_metadata_out = false; if (zookeeper->tryRemove(remote_replica_path) != Coordination::Error::ZOK) LOG_ERROR(logger, "Replica was not completely removed from ZooKeeper, {} still exists and may contain some garbage.", remote_replica_path); } /// Check that `zookeeper_path` exists: it could have been deleted by another replica after execution of previous line. Strings replicas; if (Coordination::Error::ZOK != zookeeper->tryGetChildren(zookeeper_path + "/replicas", replicas) || !replicas.empty()) return; LOG_INFO(logger, "{} is the last replica, will remove table", remote_replica_path); /** At this moment, another replica can be created and we cannot remove the table. * Try to remove /replicas node first. If we successfully removed it, * it guarantees that we are the only replica that proceed to remove the table * and no new replicas can be created after that moment (it requires the existence of /replicas node). * and table cannot be recreated with new /replicas node on another servers while we are removing data, * because table creation is executed in single transaction that will conflict with remaining nodes. */ /// Node /dropped works like a lock that protects from concurrent removal of old table and creation of new table. /// But recursive removal may fail in the middle of operation leaving some garbage in zookeeper_path, so /// we remove it on table creation if there is /dropped node. Creating thread may remove /dropped node created by /// removing thread, and it causes race condition if removing thread is not finished yet. /// To avoid this we also create ephemeral child before starting recursive removal. /// (The existence of child node does not allow to remove parent node). Coordination::Requests ops; Coordination::Responses responses; String drop_lock_path = zookeeper_path + "/dropped/lock"; ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path + "/replicas", -1)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_path + "/dropped", "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest(drop_lock_path, "", zkutil::CreateMode::Ephemeral)); Coordination::Error code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZNONODE || code == Coordination::Error::ZNODEEXISTS) { LOG_WARNING(logger, "Table {} is already started to be removing by another replica right now", remote_replica_path); } else if (code == Coordination::Error::ZNOTEMPTY) { LOG_WARNING(logger, "Another replica was suddenly created, will keep the table {}", remote_replica_path); } else if (code != Coordination::Error::ZOK) { zkutil::KeeperMultiException::check(code, ops, responses); } else { auto metadata_drop_lock = zkutil::EphemeralNodeHolder::existing(drop_lock_path, *zookeeper); LOG_INFO(logger, "Removing table {} (this might take several minutes)", zookeeper_path); removeTableNodesFromZooKeeper(zookeeper, zookeeper_path, metadata_drop_lock, logger); } } bool StorageReplicatedMergeTree::removeTableNodesFromZooKeeper(zkutil::ZooKeeperPtr zookeeper, const String & zookeeper_path, const zkutil::EphemeralNodeHolder::Ptr & metadata_drop_lock, Poco::Logger * logger) { bool completely_removed = false; /// NOTE /block_numbers/ actually is not flat, because /block_numbers// may have ephemeral children, /// but we assume that all ephemeral block locks are already removed when table is being dropped. static constexpr std::array flat_nodes = {"block_numbers", "blocks", "async_blocks", "leader_election", "log", "mutations", "pinned_part_uuids"}; /// First try to remove paths that are known to be flat for (const auto * node : flat_nodes) { bool removed_quickly = zookeeper->tryRemoveChildrenRecursive(fs::path(zookeeper_path) / node, /* probably flat */ true); if (!removed_quickly) LOG_WARNING(logger, "Failed to quickly remove node '{}' and its children, fell back to recursive removal (table: {})", node, zookeeper_path); } /// Then try to remove nodes that are known to have no children (and should always exist) Coordination::Requests ops; for (const auto * node : flat_nodes) ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path + "/" + node, -1)); ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path + "/alter_partition_version", -1)); ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path + "/columns", -1)); ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path + "/metadata", -1)); ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path + "/table_shared_id", -1)); Coordination::Responses res; auto code = zookeeper->tryMulti(ops, res); if (code != Coordination::Error::ZOK) LOG_WARNING(logger, "Cannot quickly remove nodes without children: {} (table: {}). Will remove recursively.", Coordination::errorMessage(code), zookeeper_path); Strings children; code = zookeeper->tryGetChildren(zookeeper_path, children); if (code == Coordination::Error::ZNONODE) throw Exception(ErrorCodes::LOGICAL_ERROR, "There is a race condition between creation and removal of replicated table. It's a bug"); for (const auto & child : children) { if (child != "dropped") zookeeper->tryRemoveRecursive(fs::path(zookeeper_path) / child); } ops.clear(); Coordination::Responses responses; ops.emplace_back(zkutil::makeRemoveRequest(metadata_drop_lock->getPath(), -1)); ops.emplace_back(zkutil::makeRemoveRequest(fs::path(zookeeper_path) / "dropped", -1)); ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_path, -1)); code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZNONODE) { throw Exception(ErrorCodes::LOGICAL_ERROR, "There is a race condition between creation and removal of replicated table. It's a bug"); } else if (code == Coordination::Error::ZNOTEMPTY) { LOG_ERROR(logger, "Table was not completely removed from ZooKeeper, {} still exists and may contain some garbage," "but someone is removing it right now.", zookeeper_path); } else if (code != Coordination::Error::ZOK) { /// It is still possible that ZooKeeper session is expired or server is killed in the middle of the delete operation. zkutil::KeeperMultiException::check(code, ops, responses); } else { metadata_drop_lock->setAlreadyRemoved(); completely_removed = true; LOG_INFO(logger, "Table {} was successfully removed from ZooKeeper", zookeeper_path); } return completely_removed; } /** Verify that list of columns and table storage_settings_ptr match those specified in ZK (/metadata). * If not, throw an exception. */ void StorageReplicatedMergeTree::checkTableStructure(const String & zookeeper_prefix, const StorageMetadataPtr & metadata_snapshot) { auto zookeeper = getZooKeeper(); ReplicatedMergeTreeTableMetadata old_metadata(*this, metadata_snapshot); Coordination::Stat metadata_stat; String metadata_str = zookeeper->get(fs::path(zookeeper_prefix) / "metadata", &metadata_stat); auto metadata_from_zk = ReplicatedMergeTreeTableMetadata::parse(metadata_str); old_metadata.checkEquals(metadata_from_zk, metadata_snapshot->getColumns(), getContext()); Coordination::Stat columns_stat; auto columns_from_zk = ColumnsDescription::parse(zookeeper->get(fs::path(zookeeper_prefix) / "columns", &columns_stat)); const ColumnsDescription & old_columns = metadata_snapshot->getColumns(); if (columns_from_zk != old_columns) { throw Exception(ErrorCodes::INCOMPATIBLE_COLUMNS, "Table columns structure in ZooKeeper is different from local table structure. Local columns:\n" "{}\nZookeeper columns:\n{}", old_columns.toString(), columns_from_zk.toString()); } } void StorageReplicatedMergeTree::setTableStructure(const StorageID & table_id, const ContextPtr & local_context, ColumnsDescription new_columns, const ReplicatedMergeTreeTableMetadata::Diff & metadata_diff) { StorageInMemoryMetadata old_metadata = getInMemoryMetadata(); StorageInMemoryMetadata new_metadata = metadata_diff.getNewMetadata(new_columns, local_context, old_metadata); /// Even if the primary/sorting/partition keys didn't change we must reinitialize it /// because primary/partition key column types might have changed. checkTTLExpressions(new_metadata, old_metadata); setProperties(new_metadata, old_metadata); DatabaseCatalog::instance().getDatabase(table_id.database_name)->alterTable(local_context, table_id, new_metadata); } /** If necessary, restore a part, replica itself adds a record for its receipt. * What time should I put for this entry in the queue? Time is taken into account when calculating lag of replica. * For these purposes, it makes sense to use creation time of missing part * (that is, in calculating lag, it will be taken into account how old is the part we need to recover). */ static time_t tryGetPartCreateTime(zkutil::ZooKeeperPtr & zookeeper, const String & replica_path, const String & part_name) { time_t res = 0; /// We get creation time of part, if it still exists (was not merged, for example). Coordination::Stat stat; String unused; if (zookeeper->tryGet(fs::path(replica_path) / "parts" / part_name, unused, &stat)) res = stat.ctime / 1000; return res; } void StorageReplicatedMergeTree::checkParts(bool skip_sanity_checks) { auto zookeeper = getZooKeeper(); Strings expected_parts_vec = zookeeper->getChildren(fs::path(replica_path) / "parts"); /// Parts in ZK. NameSet expected_parts(expected_parts_vec.begin(), expected_parts_vec.end()); /// There are no PreActive parts at startup. auto parts = getDataParts({MergeTreeDataPartState::Active, MergeTreeDataPartState::Outdated}); /** Local parts that are not in ZK. * In very rare cases they may cover missing parts * and someone may think that pushing them to zookeeper is good idea. * But actually we can't precisely determine that ALL missing parts * covered by this unexpected part. So missing parts will be downloaded. */ DataParts unexpected_parts; /// Intersection of local parts and expected parts ActiveDataPartSet local_expected_parts_set(format_version); /// Collect unexpected parts for (const auto & part : parts) { if (expected_parts.contains(part->name)) local_expected_parts_set.add(part->name); else unexpected_parts.insert(part); /// this parts we will place to detached with ignored_ prefix } /// Which parts should be taken from other replicas. Strings parts_to_fetch; for (const String & missing_name : expected_parts) if (!getActiveContainingPart(missing_name)) parts_to_fetch.push_back(missing_name); /** To check the adequacy, for the parts that are in the FS, but not in ZK, we will only consider not the most recent parts. * Because unexpected new parts usually arise only because they did not have time to enroll in ZK with a rough restart of the server. * It also occurs from deduplicated parts that did not have time to retire. */ size_t unexpected_parts_nonnew = 0; UInt64 unexpected_parts_nonnew_rows = 0; UInt64 unexpected_parts_rows = 0; Strings covered_unexpected_parts; Strings uncovered_unexpected_parts; UInt64 uncovered_unexpected_parts_rows = 0; for (const auto & part : unexpected_parts) { unexpected_parts_rows += part->rows_count; /// This part may be covered by some expected part that is active and present locally /// Probably we just did not remove this part from disk before restart (but removed from ZooKeeper) String covering_local_part = local_expected_parts_set.getContainingPart(part->name); if (!covering_local_part.empty()) { covered_unexpected_parts.push_back(part->name); continue; } /// Part is unexpected and we don't have covering part: it's suspicious uncovered_unexpected_parts.push_back(part->name); uncovered_unexpected_parts_rows += part->rows_count; if (part->info.level > 0) { ++unexpected_parts_nonnew; unexpected_parts_nonnew_rows += part->rows_count; } } const UInt64 parts_to_fetch_blocks = std::accumulate(parts_to_fetch.cbegin(), parts_to_fetch.cend(), 0, [&](UInt64 acc, const String& part_name) { if (const auto part_info = MergeTreePartInfo::tryParsePartName(part_name, format_version)) return acc + part_info->getBlocksCount(); LOG_ERROR(log, "Unexpected part name: {}", part_name); return acc; }); /** We can automatically synchronize data, * if the ratio of the total number of errors to the total number of parts (minimum - on the local filesystem or in ZK) * is no more than some threshold (for example 50%). * * A large ratio of mismatches in the data on the filesystem and the expected data * may indicate a configuration error (the server accidentally connected as a replica not from right shard). * In this case, the protection mechanism does not allow the server to start. */ UInt64 total_rows_on_filesystem = 0; for (const auto & part : parts) total_rows_on_filesystem += part->rows_count; const auto storage_settings_ptr = getSettings(); bool insane = uncovered_unexpected_parts_rows > total_rows_on_filesystem * storage_settings_ptr->replicated_max_ratio_of_wrong_parts; constexpr auto sanity_report_fmt = "The local set of parts of table {} doesn't look like the set of parts in ZooKeeper: " "{} rows of {} total rows in filesystem are suspicious. " "There are {} uncovered unexpected parts with {} rows ({} of them is not just-written with {} rows), " "{} missing parts (with {} blocks), {} covered unexpected parts (with {} rows)."; constexpr auto sanity_report_debug_fmt = "Uncovered unexpected parts: {}. Missing parts: {}. Covered unexpected parts: {}. Expected parts: {}."; if (insane && !skip_sanity_checks) { LOG_DEBUG(log, sanity_report_debug_fmt, fmt::join(uncovered_unexpected_parts, ", "), fmt::join(parts_to_fetch, ", "), fmt::join(covered_unexpected_parts, ", "), fmt::join(expected_parts, ", ")); throw Exception(ErrorCodes::TOO_MANY_UNEXPECTED_DATA_PARTS, sanity_report_fmt, getStorageID().getNameForLogs(), formatReadableQuantity(uncovered_unexpected_parts_rows), formatReadableQuantity(total_rows_on_filesystem), uncovered_unexpected_parts.size(), uncovered_unexpected_parts_rows, unexpected_parts_nonnew, unexpected_parts_nonnew_rows, parts_to_fetch.size(), parts_to_fetch_blocks, covered_unexpected_parts.size(), unexpected_parts_rows - uncovered_unexpected_parts_rows); } if (unexpected_parts_nonnew_rows > 0 || uncovered_unexpected_parts_rows > 0) { LOG_DEBUG(log, sanity_report_debug_fmt, fmt::join(uncovered_unexpected_parts, ", "), fmt::join(parts_to_fetch, ", "), fmt::join(covered_unexpected_parts, ", "), fmt::join(expected_parts, ", ")); LOG_WARNING(log, sanity_report_fmt, getStorageID().getNameForLogs(), formatReadableQuantity(uncovered_unexpected_parts_rows), formatReadableQuantity(total_rows_on_filesystem), uncovered_unexpected_parts.size(), uncovered_unexpected_parts_rows, unexpected_parts_nonnew, unexpected_parts_nonnew_rows, parts_to_fetch.size(), parts_to_fetch_blocks, covered_unexpected_parts.size(), unexpected_parts_rows - uncovered_unexpected_parts_rows); } /// Add to the queue jobs to pick up the missing parts from other replicas and remove from ZK the information that we have them. queue.setBrokenPartsToEnqueueFetchesOnLoading(std::move(parts_to_fetch)); /// Remove extra local parts. for (const DataPartPtr & part : unexpected_parts) { LOG_ERROR(log, "Renaming unexpected part {} to ignored_{}", part->name, part->name); forcefullyMovePartToDetachedAndRemoveFromMemory(part, "ignored", true); } } void StorageReplicatedMergeTree::syncPinnedPartUUIDs() { auto zookeeper = getZooKeeper(); Coordination::Stat stat; String s = zookeeper->get(zookeeper_path + "/pinned_part_uuids", &stat); std::lock_guard lock(pinned_part_uuids_mutex); /// Unsure whether or not this can be called concurrently. if (pinned_part_uuids->stat.version < stat.version) { auto new_pinned_part_uuids = std::make_shared(); new_pinned_part_uuids->fromString(s); new_pinned_part_uuids->stat = stat; pinned_part_uuids = new_pinned_part_uuids; } } void StorageReplicatedMergeTree::checkPartChecksumsAndAddCommitOps(const zkutil::ZooKeeperPtr & zookeeper, const DataPartPtr & part, Coordination::Requests & ops, String part_name, NameSet * absent_replicas_paths) { if (part_name.empty()) part_name = part->name; auto local_part_header = ReplicatedMergeTreePartHeader::fromColumnsAndChecksums( part->getColumns(), part->checksums); Strings replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); std::shuffle(replicas.begin(), replicas.end(), thread_local_rng); bool has_been_already_added = false; for (const String & replica : replicas) { String current_part_path = fs::path(zookeeper_path) / "replicas" / replica / "parts" / part_name; String part_zk_str; if (!zookeeper->tryGet(current_part_path, part_zk_str)) { if (absent_replicas_paths) absent_replicas_paths->emplace(current_part_path); continue; } ReplicatedMergeTreePartHeader replica_part_header; if (part_zk_str.empty()) { String columns_str; String checksums_str; if (zookeeper->tryGet(fs::path(current_part_path) / "columns", columns_str) && zookeeper->tryGet(fs::path(current_part_path) / "checksums", checksums_str)) { replica_part_header = ReplicatedMergeTreePartHeader::fromColumnsAndChecksumsZNodes(columns_str, checksums_str); } else { if (zookeeper->exists(current_part_path)) throw Exception(ErrorCodes::LOGICAL_ERROR, "Part {} has empty header and does not have columns and checksums. " "Looks like a bug.", current_part_path); LOG_INFO(log, "Not checking checksums of part {} with replica {} because part was removed from ZooKeeper", part_name, replica); continue; } } else { replica_part_header = ReplicatedMergeTreePartHeader::fromString(part_zk_str); } if (replica_part_header.getColumnsHash() != local_part_header.getColumnsHash()) { /// Currently there are two (known) cases when it may happen: /// - KILL MUTATION query had removed mutation before all replicas have executed assigned MUTATE_PART entries. /// Some replicas may skip this mutation and update part version without actually applying any changes. /// It leads to mismatching checksum if changes were applied on other replicas. /// - ALTER_METADATA and MERGE_PARTS were reordered on some replicas. /// It may lead to different number of columns in merged parts on these replicas. throw Exception(ErrorCodes::CHECKSUM_DOESNT_MATCH, "Part {} from {} has different columns hash " "(it may rarely happen on race condition with KILL MUTATION or ALTER COLUMN).", part_name, replica); } replica_part_header.getChecksums().checkEqual(local_part_header.getChecksums(), true); if (replica == replica_name) has_been_already_added = true; /// If we verify checksums in "sequential manner" (i.e. recheck absence of checksums on other replicas when commit) /// then it is enough to verify checksums on at least one replica since checksums on other replicas must be the same. if (absent_replicas_paths) { absent_replicas_paths->clear(); break; } } if (!has_been_already_added) { const auto storage_settings_ptr = getSettings(); String part_path = fs::path(replica_path) / "parts" / part_name; if (storage_settings_ptr->use_minimalistic_part_header_in_zookeeper) { ops.emplace_back(zkutil::makeCreateRequest( part_path, local_part_header.toString(), zkutil::CreateMode::Persistent)); } else { ops.emplace_back(zkutil::makeCreateRequest( part_path, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest( fs::path(part_path) / "columns", part->getColumns().toString(), zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest( fs::path(part_path) / "checksums", getChecksumsForZooKeeper(part->checksums), zkutil::CreateMode::Persistent)); } } else { LOG_WARNING(log, "checkPartAndAddToZooKeeper: node {} already exists. Will not commit any nodes.", (fs::path(replica_path) / "parts" / part_name).string()); } } MergeTreeData::DataPartsVector StorageReplicatedMergeTree::checkPartChecksumsAndCommit(Transaction & transaction, const MutableDataPartPtr & part, std::optional hardlinked_files) { auto zookeeper = getZooKeeper(); while (true) { Coordination::Requests ops; NameSet absent_part_paths_on_replicas; lockSharedData(*part, false, hardlinked_files); /// Checksums are checked here and `ops` is filled. In fact, the part is added to ZK just below, when executing `multi`. checkPartChecksumsAndAddCommitOps(zookeeper, part, ops, part->name, &absent_part_paths_on_replicas); /// Do not commit if the part is obsolete, we have just briefly checked its checksums if (transaction.isEmpty()) return {}; /// Will check that the part did not suddenly appear on skipped replicas if (!absent_part_paths_on_replicas.empty()) { Coordination::Requests new_ops; for (const String & part_path : absent_part_paths_on_replicas) { /// NOTE Create request may fail with ZNONODE if replica is being dropped, we will throw an exception new_ops.emplace_back(zkutil::makeCreateRequest(part_path, "", zkutil::CreateMode::Persistent)); new_ops.emplace_back(zkutil::makeRemoveRequest(part_path, -1)); } /// Add check ops at the beginning new_ops.insert(new_ops.end(), ops.begin(), ops.end()); ops = std::move(new_ops); } Coordination::Responses responses; Coordination::Error e = zookeeper->tryMulti(ops, responses); if (e == Coordination::Error::ZOK) return transaction.commit(); if (e == Coordination::Error::ZNODEEXISTS) { size_t num_check_ops = 2 * absent_part_paths_on_replicas.size(); size_t failed_op_index = zkutil::getFailedOpIndex(e, responses); if (failed_op_index < num_check_ops) { LOG_INFO(log, "The part {} on a replica suddenly appeared, will recheck checksums", ops[failed_op_index]->getPath()); continue; } } throw zkutil::KeeperException(e); } } String StorageReplicatedMergeTree::getChecksumsForZooKeeper(const MergeTreeDataPartChecksums & checksums) const { return MinimalisticDataPartChecksums::getSerializedString(checksums, getSettings()->use_minimalistic_checksums_in_zookeeper); } MergeTreeData::MutableDataPartPtr StorageReplicatedMergeTree::attachPartHelperFoundValidPart(const LogEntry & entry) const { if (format_version != MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING) return {}; const MergeTreePartInfo actual_part_info = MergeTreePartInfo::fromPartName(entry.new_part_name, format_version); const String part_new_name = actual_part_info.getPartNameV1(); for (const DiskPtr & disk : getStoragePolicy()->getDisks()) { for (const auto it = disk->iterateDirectory(fs::path(relative_data_path) / "detached/"); it->isValid(); it->next()) { const auto part_info = MergeTreePartInfo::tryParsePartName(it->name(), format_version); if (!part_info || part_info->partition_id != actual_part_info.partition_id) continue; const auto part_old_name = part_info->getPartNameV1(); const auto volume = std::make_shared("volume_" + part_old_name, disk); auto part = getDataPartBuilder(entry.new_part_name, volume, fs::path("detached") / part_old_name) .withPartFormatFromDisk() .build(); try { part->loadColumnsChecksumsIndexes(true, true); } catch (const Exception&) { /// This method throws if the part data is corrupted or partly missing. In this case, we simply don't /// process the part. continue; } if (entry.part_checksum == part->checksums.getTotalChecksumHex()) { part->modification_time = part->getDataPartStorage().getLastModified().epochTime(); return part; } } } return {}; } bool StorageReplicatedMergeTree::executeLogEntry(LogEntry & entry) { if (entry.type == LogEntry::DROP_RANGE || entry.type == LogEntry::DROP_PART) { executeDropRange(entry); return true; } if (entry.type == LogEntry::REPLACE_RANGE) { executeReplaceRange(entry); return true; } const bool is_get_or_attach = entry.type == LogEntry::GET_PART || entry.type == LogEntry::ATTACH_PART; if (is_get_or_attach || entry.type == LogEntry::MERGE_PARTS || entry.type == LogEntry::MUTATE_PART) { /// If we already have this part or a part covering it, we do not need to do anything. /// The part may be still in the PreActive -> Active transition so we first search /// among PreActive parts to definitely find the desired part if it exists. DataPartPtr existing_part = getPartIfExists(entry.new_part_name, {MergeTreeDataPartState::PreActive}); if (!existing_part) existing_part = getActiveContainingPart(entry.new_part_name); /// Even if the part is local, it (in exceptional cases) may not be in ZooKeeper. Let's check that it is there. if (existing_part && getZooKeeper()->exists(fs::path(replica_path) / "parts" / existing_part->name)) { if (!is_get_or_attach || entry.source_replica != replica_name) LOG_DEBUG(log, "Skipping action for part {} because part {} already exists.", entry.new_part_name, existing_part->name); return true; } } if (entry.type == LogEntry::ATTACH_PART) { ProfileEventsScope profile_events_scope; if (MutableDataPartPtr part = attachPartHelperFoundValidPart(entry)) { LOG_TRACE(log, "Found valid local part for {}, preparing the transaction", part->name); Transaction transaction(*this, NO_TRANSACTION_RAW); part->version.setCreationTID(Tx::PrehistoricTID, nullptr); renameTempPartAndReplace(part, transaction); checkPartChecksumsAndCommit(transaction, part); writePartLog(PartLogElement::Type::NEW_PART, {}, 0 /** log entry is fake so we don't measure the time */, part->name, part, {} /** log entry is fake so there are no initial parts */, nullptr, profile_events_scope.getSnapshot()); return true; } LOG_TRACE(log, "Didn't find valid local part for {} ({}), will fetch it from other replica", entry.new_part_name, entry.actual_new_part_name); } if (is_get_or_attach && entry.source_replica == replica_name) LOG_WARNING(log, "Part {} from own log doesn't exist.", entry.new_part_name); /// Perhaps we don't need this part, because during write with quorum, the quorum has failed /// (see below about `/quorum/failed_parts`). if (entry.quorum && getZooKeeper()->exists(fs::path(zookeeper_path) / "quorum" / "failed_parts" / entry.new_part_name)) { LOG_DEBUG(log, "Skipping action for part {} because quorum for that part was failed.", entry.new_part_name); return true; /// NOTE Deletion from `virtual_parts` is not done, but it is only necessary for merge. } switch (entry.type) { case LogEntry::ATTACH_PART: /// We surely don't have this part locally as we've checked it before, so download it. [[fallthrough]]; case LogEntry::GET_PART: return executeFetch(entry); case LogEntry::MERGE_PARTS: throw Exception(ErrorCodes::LOGICAL_ERROR, "Merge has to be executed by another function"); case LogEntry::MUTATE_PART: throw Exception(ErrorCodes::LOGICAL_ERROR, "Mutation has to be executed by another function"); case LogEntry::ALTER_METADATA: return executeMetadataAlter(entry); case LogEntry::SYNC_PINNED_PART_UUIDS: syncPinnedPartUUIDs(); return true; case LogEntry::CLONE_PART_FROM_SHARD: executeClonePartFromShard(entry); return true; default: throw Exception(ErrorCodes::LOGICAL_ERROR, "Unexpected log entry type: {}", static_cast(entry.type)); } } bool StorageReplicatedMergeTree::executeFetch(LogEntry & entry, bool need_to_check_missing_part) { /// Looking for covering part. After that entry.actual_new_part_name may be filled. String replica = findReplicaHavingCoveringPart(entry, true); const auto storage_settings_ptr = getSettings(); auto metadata_snapshot = getInMemoryMetadataPtr(); try { if (replica.empty()) { /** If a part is to be written with a quorum and the quorum is not reached yet, * then (due to the fact that a part is impossible to download right now), * the quorum entry should be considered unsuccessful. * TODO Complex code, extract separately. */ if (entry.quorum) { if (entry.type != LogEntry::GET_PART) throw Exception(ErrorCodes::LOGICAL_ERROR, "Logical error: log entry with quorum but type is not GET_PART"); LOG_DEBUG(log, "No active replica has part {} which needs to be written with quorum. Will try to mark that quorum as failed.", entry.new_part_name); /** Atomically: * - if replicas do not become active; * - if there is a `quorum` node with this part; * - delete `quorum` node; * - add a part to the list `quorum/failed_parts`; * - if the part is not already removed from the list for deduplication `blocks/block_num`, then delete it; * * If something changes, then we will nothing - we'll get here again next time. */ /** We collect the `host` node versions from the replicas. * When the replica becomes active, it changes the value of host in the same transaction (with the creation of `is_active`). * This will ensure that the replicas do not become active. */ auto zookeeper = getZooKeeper(); Strings replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); Coordination::Requests ops; for (const auto & path_part : replicas) { Coordination::Stat stat; String path = fs::path(zookeeper_path) / "replicas" / path_part / "host"; zookeeper->get(path, &stat); ops.emplace_back(zkutil::makeCheckRequest(path, stat.version)); } /// We verify that while we were collecting versions, the replica with the necessary part did not come alive. replica = findReplicaHavingPart(entry.new_part_name, true); /// Also during this time a completely new replica could be created. /// But if a part does not appear on the old, then it can not be on the new one either. if (replica.empty()) { Coordination::Stat quorum_stat; const String quorum_unparallel_path = fs::path(zookeeper_path) / "quorum" / "status"; const String quorum_parallel_path = fs::path(zookeeper_path) / "quorum" / "parallel" / entry.new_part_name; String quorum_str, quorum_path; ReplicatedMergeTreeQuorumEntry quorum_entry; if (zookeeper->tryGet(quorum_unparallel_path, quorum_str, &quorum_stat)) quorum_path = quorum_unparallel_path; else { quorum_str = zookeeper->get(quorum_parallel_path, &quorum_stat); quorum_path = quorum_parallel_path; } quorum_entry.fromString(quorum_str); if (quorum_entry.part_name == entry.new_part_name) { ops.emplace_back(zkutil::makeRemoveRequest(quorum_path, quorum_stat.version)); auto part_info = MergeTreePartInfo::fromPartName(entry.new_part_name, format_version); if (part_info.min_block != part_info.max_block) throw Exception(ErrorCodes::LOGICAL_ERROR, "Logical error: log entry with quorum for part covering more than one block number"); ops.emplace_back(zkutil::makeCreateRequest( fs::path(zookeeper_path) / "quorum" / "failed_parts" / entry.new_part_name, "", zkutil::CreateMode::Persistent)); /// Deleting from `blocks`. if (!entry.block_id.empty() && zookeeper->exists(fs::path(zookeeper_path) / "blocks" / entry.block_id)) ops.emplace_back(zkutil::makeRemoveRequest(fs::path(zookeeper_path) / "blocks" / entry.block_id, -1)); Coordination::Responses responses; auto code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZOK) { LOG_DEBUG(log, "Marked quorum for part {} as failed.", entry.new_part_name); queue.removeFailedQuorumPart(part_info); return true; } else if (code == Coordination::Error::ZBADVERSION || code == Coordination::Error::ZNONODE || code == Coordination::Error::ZNODEEXISTS) { LOG_DEBUG(log, "State was changed or isn't expected when trying to mark quorum for part {} as failed. Code: {}", entry.new_part_name, Coordination::errorMessage(code)); } else throw Coordination::Exception(code); } else { LOG_WARNING(log, "No active replica has part {}, " "but that part needs quorum and /quorum/status contains entry about another part {}. " "It means that part was successfully written to {} replicas, but then all of them goes offline. " "Or it is a bug.", entry.new_part_name, quorum_entry.part_name, entry.quorum); } } } if (replica.empty()) { ProfileEvents::increment(ProfileEvents::ReplicatedPartFailedFetches); if (!need_to_check_missing_part) return false; throw Exception(ErrorCodes::NO_REPLICA_HAS_PART, "No active replica has part {} or covering part (cannot execute {}: {})", entry.new_part_name, entry.znode_name, entry.getDescriptionForLogs(format_version)); } } try { String part_name = entry.actual_new_part_name.empty() ? entry.new_part_name : entry.actual_new_part_name; if (!entry.actual_new_part_name.empty()) LOG_DEBUG(log, "Will fetch part {} instead of {}", entry.actual_new_part_name, entry.new_part_name); String source_replica_path = fs::path(zookeeper_path) / "replicas" / replica; if (!fetchPart(part_name, metadata_snapshot, source_replica_path, /* to_detached= */ false, entry.quorum, /* zookeeper_ */ nullptr, /* try_fetch_shared= */ true)) { return false; } } catch (Exception & e) { /// No stacktrace, just log message if (e.code() == ErrorCodes::RECEIVED_ERROR_TOO_MANY_REQUESTS) e.addMessage("Too busy replica. Will try later."); throw; } if (entry.type == LogEntry::MERGE_PARTS) ProfileEvents::increment(ProfileEvents::ReplicatedPartFetchesOfMerged); } catch (...) { /** If we can not download the part we need for some merge, it's better not to try to get other parts for this merge, * but try to get already merged part. To do this, move the action to get the remaining parts * for this merge at the end of the queue. */ try { auto parts_for_merge = queue.moveSiblingPartsForMergeToEndOfQueue(entry.new_part_name); if (!parts_for_merge.empty() && replica.empty()) { LOG_INFO(log, "No active replica has part {}. Will fetch merged part instead.", entry.new_part_name); /// We should enqueue it for check, because merged part may never appear if source part is lost enqueuePartForCheck(entry.new_part_name); return false; } /** If no active replica has a part, and there is no merge in the queue with its participation, * check to see if any (active or inactive) replica has such a part or covering it. */ if (replica.empty()) enqueuePartForCheck(entry.new_part_name); } catch (...) { tryLogCurrentException(log, __PRETTY_FUNCTION__); } throw; } return true; } MutableDataPartStoragePtr StorageReplicatedMergeTree::executeFetchShared( const String & source_replica, const String & new_part_name, const DiskPtr & disk, const String & path) { if (source_replica.empty()) { LOG_INFO(log, "No active replica has part {} on shared storage.", new_part_name); return nullptr; } const auto storage_settings_ptr = getSettings(); auto metadata_snapshot = getInMemoryMetadataPtr(); try { return fetchExistsPart(new_part_name, metadata_snapshot, fs::path(zookeeper_path) / "replicas" / source_replica, disk, path); } catch (Exception & e) { if (e.code() == ErrorCodes::RECEIVED_ERROR_TOO_MANY_REQUESTS) e.addMessage("Too busy replica. Will try later."); tryLogCurrentException(log, __PRETTY_FUNCTION__); throw; } } void StorageReplicatedMergeTree::executeDropRange(const LogEntry & entry) { LOG_TRACE(log, "Executing DROP_RANGE {}", entry.new_part_name); /// Wait for loading of outdated parts because DROP_RANGE /// command must be applied to all parts on disk. waitForOutdatedPartsToBeLoaded(); auto drop_range_info = MergeTreePartInfo::fromPartName(entry.new_part_name, format_version); getContext()->getMergeList().cancelInPartition(getStorageID(), drop_range_info.partition_id, drop_range_info.max_block); { auto pause_checking_parts = part_check_thread.pausePartsCheck(); queue.removePartProducingOpsInRange(getZooKeeper(), drop_range_info, entry); part_check_thread.cancelRemovedPartsCheck(drop_range_info); } /// Delete the parts contained in the range to be deleted. /// It's important that no old parts remain (after the merge), because otherwise, /// after adding a new replica, this new replica downloads them, but does not delete them. /// And, if you do not, the parts will come to life after the server is restarted. /// Therefore, we use all data parts. auto metadata_snapshot = getInMemoryMetadataPtr(); PartsToRemoveFromZooKeeper parts_to_remove; { auto data_parts_lock = lockParts(); parts_to_remove = removePartsInRangeFromWorkingSetAndGetPartsToRemoveFromZooKeeper(NO_TRANSACTION_RAW, drop_range_info, data_parts_lock); if (parts_to_remove.empty()) { if (!drop_range_info.isFakeDropRangePart()) LOG_INFO(log, "Log entry {} tried to drop single part {}, but part does not exist", entry.znode_name, entry.new_part_name); return; } } if (entry.detach) LOG_DEBUG(log, "Detaching parts."); else LOG_DEBUG(log, "Removing parts."); if (entry.detach) { /// If DETACH clone parts to detached/ directory for (const auto & part : parts_to_remove) { if (auto part_to_detach = part.getPartIfItWasActive()) { LOG_INFO(log, "Detaching {}", part_to_detach->getDataPartStorage().getPartDirectory()); part_to_detach->makeCloneInDetached("", metadata_snapshot); } } } /// Forcibly remove parts from ZooKeeper removePartsFromZooKeeperWithRetries(parts_to_remove); if (entry.detach) LOG_DEBUG(log, "Detached {} parts inside {}.", parts_to_remove.size(), entry.new_part_name); else LOG_DEBUG(log, "Removed {} parts inside {}.", parts_to_remove.size(), entry.new_part_name); /// We want to remove dropped parts from disk as soon as possible /// To be removed a partition should have zero refcount, therefore call the cleanup thread at exit parts_to_remove.clear(); cleanup_thread.wakeup(); } bool StorageReplicatedMergeTree::executeReplaceRange(const LogEntry & entry) { Stopwatch watch; ProfileEventsScope profile_events_scope; auto & entry_replace = *entry.replace_range_entry; LOG_DEBUG(log, "Executing log entry {} to replace parts range {} with {} parts from {}.{}", entry.znode_name, entry_replace.drop_range_part_name, entry_replace.new_part_names.size(), entry_replace.from_database, entry_replace.from_table); /// Wait for loading of outdated parts because REPLACE_RANGE /// command must be applied to all parts on disk. waitForOutdatedPartsToBeLoaded(); auto metadata_snapshot = getInMemoryMetadataPtr(); auto storage_settings_ptr = getSettings(); MergeTreePartInfo drop_range = MergeTreePartInfo::fromPartName(entry_replace.drop_range_part_name, format_version); /// Range with only one block has special meaning: it's ATTACH PARTITION or MOVE PARTITION, so there is no drop range bool replace = !LogEntry::ReplaceRangeEntry::isMovePartitionOrAttachFrom(drop_range); if (replace) { getContext()->getMergeList().cancelInPartition(getStorageID(), drop_range.partition_id, drop_range.max_block); auto pause_checking_parts = part_check_thread.pausePartsCheck(); queue.removePartProducingOpsInRange(getZooKeeper(), drop_range, entry); part_check_thread.cancelRemovedPartsCheck(drop_range); } else { drop_range = {}; } struct PartDescription { PartDescription( size_t index_, const String & src_part_name_, const String & new_part_name_, const String & checksum_hex_, MergeTreeDataFormatVersion format_version) : index(index_) , src_part_name(src_part_name_) , src_part_info(MergeTreePartInfo::fromPartName(src_part_name_, format_version)) , new_part_name(new_part_name_) , new_part_info(MergeTreePartInfo::fromPartName(new_part_name_, format_version)) , checksum_hex(checksum_hex_) { } size_t index; // in log entry arrays String src_part_name; MergeTreePartInfo src_part_info; String new_part_name; MergeTreePartInfo new_part_info; String checksum_hex; /// Part which will be committed MutableDataPartPtr res_part; /// We could find a covering part MergeTreePartInfo found_new_part_info; String found_new_part_name; /// Hold pointer to part in source table if will clone it from local table DataPartPtr src_table_part; /// A replica that will be used to fetch part String replica; MergeTreeData::HardlinkedFiles hardlinked_files; scope_guard temporary_part_lock; }; using PartDescriptionPtr = std::shared_ptr; using PartDescriptions = std::vector; PartDescriptions all_parts; PartDescriptions parts_to_add; PartsToRemoveFromZooKeeper parts_to_remove; auto table_lock_holder_dst_table = lockForShare( RWLockImpl::NO_QUERY, getSettings()->lock_acquire_timeout_for_background_operations); auto dst_metadata_snapshot = getInMemoryMetadataPtr(); for (size_t i = 0; i < entry_replace.new_part_names.size(); ++i) { all_parts.emplace_back(std::make_shared(i, entry_replace.src_part_names.at(i), entry_replace.new_part_names.at(i), entry_replace.part_names_checksums.at(i), format_version)); } /// What parts we should add? Or we have already added all required parts (we an replica-initializer) { auto data_parts_lock = lockParts(); for (const PartDescriptionPtr & part_desc : all_parts) { if (!getActiveContainingPart(part_desc->new_part_info, MergeTreeDataPartState::Active, data_parts_lock)) parts_to_add.emplace_back(part_desc); } if (parts_to_add.empty() && replace) { parts_to_remove = removePartsInRangeFromWorkingSetAndGetPartsToRemoveFromZooKeeper(NO_TRANSACTION_RAW, drop_range, data_parts_lock); String parts_to_remove_str; for (const auto & part : parts_to_remove) { parts_to_remove_str += part.getPartName(); parts_to_remove_str += " "; } LOG_TRACE(log, "Replacing {} parts {}with empty set", parts_to_remove.size(), parts_to_remove_str); } } if (parts_to_add.empty()) { LOG_INFO(log, "All parts from REPLACE PARTITION command have been already attached"); removePartsFromZooKeeperWithRetries(parts_to_remove); return true; } if (parts_to_add.size() < all_parts.size()) { LOG_WARNING(log, "Some (but not all) parts from REPLACE PARTITION command already exist. REPLACE PARTITION will not be atomic."); } StoragePtr source_table; TableLockHolder table_lock_holder_src_table; StorageID source_table_id{entry_replace.from_database, entry_replace.from_table}; auto clone_data_parts_from_source_table = [&] () -> size_t { source_table = DatabaseCatalog::instance().tryGetTable(source_table_id, getContext()); if (!source_table) { LOG_DEBUG(log, "Can't use {} as source table for REPLACE PARTITION command. It does not exist.", source_table_id.getNameForLogs()); return 0; } auto src_metadata_snapshot = source_table->getInMemoryMetadataPtr(); MergeTreeData * src_data = nullptr; try { src_data = &checkStructureAndGetMergeTreeData(source_table, src_metadata_snapshot, dst_metadata_snapshot); } catch (Exception &) { LOG_INFO(log, "Can't use {} as source table for REPLACE PARTITION command. Will fetch all parts. Reason: {}", source_table_id.getNameForLogs(), getCurrentExceptionMessage(false)); return 0; } table_lock_holder_src_table = source_table->lockForShare( RWLockImpl::NO_QUERY, getSettings()->lock_acquire_timeout_for_background_operations); DataPartStates valid_states{ MergeTreeDataPartState::PreActive, MergeTreeDataPartState::Active, MergeTreeDataPartState::Outdated}; size_t num_clonable_parts = 0; for (PartDescriptionPtr & part_desc : parts_to_add) { auto src_part = src_data->getPartIfExists(part_desc->src_part_info, valid_states); if (!src_part) { LOG_DEBUG(log, "There is no part {} in {}", part_desc->src_part_name, source_table_id.getNameForLogs()); continue; } bool avoid_copy_local_part = storage_settings_ptr->allow_remote_fs_zero_copy_replication && src_part->isStoredOnRemoteDiskWithZeroCopySupport(); if (avoid_copy_local_part) { LOG_DEBUG(log, "Avoid copy local part {} from table {} because of zero-copy replication", part_desc->src_part_name, source_table_id.getNameForLogs()); continue; } String checksum_hex = src_part->checksums.getTotalChecksumHex(); if (checksum_hex != part_desc->checksum_hex) { LOG_DEBUG(log, "Part {} of {} has inappropriate checksum", part_desc->src_part_name, source_table_id.getNameForLogs()); /// TODO: check version continue; } part_desc->found_new_part_name = part_desc->new_part_name; part_desc->found_new_part_info = part_desc->new_part_info; part_desc->src_table_part = src_part; ++num_clonable_parts; } return num_clonable_parts; }; size_t num_clonable_parts = clone_data_parts_from_source_table(); LOG_DEBUG(log, "Found {} parts that could be cloned (of {} required parts)", num_clonable_parts, parts_to_add.size()); ActiveDataPartSet adding_parts_active_set(format_version); std::unordered_map part_name_to_desc; for (PartDescriptionPtr & part_desc : parts_to_add) { if (part_desc->src_table_part) { /// It is clonable part adding_parts_active_set.add(part_desc->new_part_name); part_name_to_desc.emplace(part_desc->new_part_name, part_desc); continue; } /// Firstly, try find exact part to produce more accurate part set String replica = findReplicaHavingPart(part_desc->new_part_name, true); String found_part_name; /// TODO: check version if (replica.empty()) { LOG_DEBUG(log, "Part {} is not found on remote replicas", part_desc->new_part_name); /// Fallback to covering part replica = findReplicaHavingCoveringPart(part_desc->new_part_name, true, found_part_name); if (replica.empty()) { /// It is not fail, since adjacent parts could cover current part LOG_DEBUG(log, "Parts covering {} are not found on remote replicas", part_desc->new_part_name); continue; } } else { found_part_name = part_desc->new_part_name; } part_desc->found_new_part_name = found_part_name; part_desc->found_new_part_info = MergeTreePartInfo::fromPartName(found_part_name, format_version); part_desc->replica = replica; adding_parts_active_set.add(part_desc->found_new_part_name); part_name_to_desc.emplace(part_desc->found_new_part_name, part_desc); } /// Check that we could cover whole range for (PartDescriptionPtr & part_desc : parts_to_add) { if (adding_parts_active_set.getContainingPart(part_desc->new_part_info).empty()) { /// We should enqueue missing part for check, so it will be replaced with empty one (if needed) /// and we will be able to execute this REPLACE_RANGE. /// However, it's quite dangerous, because part may appear in source table. /// So we enqueue it for check only if no replicas of source table have part either. bool need_check = true; if (auto * replicated_src_table = typeid_cast(source_table.get())) { String src_replica = replicated_src_table->findReplicaHavingPart(part_desc->src_part_name, false); if (!src_replica.empty()) { LOG_DEBUG(log, "Found part {} on replica {} of source table, will not check part {} required for {}", part_desc->src_part_name, src_replica, part_desc->new_part_name, entry.znode_name); need_check = false; } } if (need_check) { LOG_DEBUG(log, "Will check part {} required for {}, because no replicas have it (including replicas of source table)", part_desc->new_part_name, entry.znode_name); enqueuePartForCheck(part_desc->new_part_name); } throw Exception(ErrorCodes::NO_REPLICA_HAS_PART, "Not found part {} (or part covering it) neither source table neither remote replicas", part_desc->new_part_name); } } /// Filter covered parts PartDescriptions final_parts; Strings final_part_names; { final_part_names = adding_parts_active_set.getParts(); for (const String & final_part_name : final_part_names) { auto part_desc = part_name_to_desc[final_part_name]; if (!part_desc) throw Exception(ErrorCodes::LOGICAL_ERROR, "There is no final part {}. This is a bug", final_part_name); final_parts.emplace_back(part_desc); if (final_parts.size() > 1) { auto & prev = *final_parts[final_parts.size() - 2]; auto & curr = *final_parts[final_parts.size() - 1]; if (!prev.found_new_part_info.isDisjoint(curr.found_new_part_info)) { throw Exception(ErrorCodes::LOGICAL_ERROR, "Intersected final parts detected: {} and {}. It should be investigated.", prev.found_new_part_name, curr.found_new_part_name); } } } } static const String TMP_PREFIX = "tmp_replace_from_"; std::vector hardlinked_files_for_parts; auto obtain_part = [&] (PartDescriptionPtr & part_desc) { if (part_desc->src_table_part) { if (part_desc->checksum_hex != part_desc->src_table_part->checksums.getTotalChecksumHex()) throw Exception(ErrorCodes::UNFINISHED, "Checksums of {} is suddenly changed", part_desc->src_table_part->name); auto [res_part, temporary_part_lock] = cloneAndLoadDataPartOnSameDisk( part_desc->src_table_part, TMP_PREFIX + "clone_", part_desc->new_part_info, metadata_snapshot, NO_TRANSACTION_PTR, &part_desc->hardlinked_files, false, {}); part_desc->res_part = std::move(res_part); part_desc->temporary_part_lock = std::move(temporary_part_lock); } else if (!part_desc->replica.empty()) { String source_replica_path = fs::path(zookeeper_path) / "replicas" / part_desc->replica; ReplicatedMergeTreeAddress address(getZooKeeper()->get(fs::path(source_replica_path) / "host")); auto timeouts = getFetchPartHTTPTimeouts(getContext()); auto credentials = getContext()->getInterserverCredentials(); String interserver_scheme = getContext()->getInterserverScheme(); if (interserver_scheme != address.scheme) throw Exception(ErrorCodes::LOGICAL_ERROR, "Interserver schemas are different '{}' != '{}', can't fetch part from {}", interserver_scheme, address.scheme, address.host); part_desc->res_part = fetcher.fetchSelectedPart( metadata_snapshot, getContext(), part_desc->found_new_part_name, source_replica_path, address.host, address.replication_port, timeouts, credentials->getUser(), credentials->getPassword(), interserver_scheme, replicated_fetches_throttler, false, TMP_PREFIX + "fetch_"); /// TODO: check columns_version of fetched part ProfileEvents::increment(ProfileEvents::ReplicatedPartFetches); } else throw Exception(ErrorCodes::LOGICAL_ERROR, "There is no receipt to produce part {}. This is bug", part_desc->new_part_name); }; /// Download or clone parts /// TODO: make it in parallel for (PartDescriptionPtr & part_desc : final_parts) obtain_part(part_desc); MutableDataPartsVector res_parts; for (PartDescriptionPtr & part_desc : final_parts) res_parts.emplace_back(part_desc->res_part); try { /// Commit parts auto zookeeper = getZooKeeper(); Transaction transaction(*this, NO_TRANSACTION_RAW); Coordination::Requests ops; for (PartDescriptionPtr & part_desc : final_parts) { renameTempPartAndReplace(part_desc->res_part, transaction); getCommitPartOps(ops, part_desc->res_part); lockSharedData(*part_desc->res_part, false, part_desc->hardlinked_files); } if (!ops.empty()) zookeeper->multi(ops); { auto data_parts_lock = lockParts(); transaction.commit(&data_parts_lock); if (replace) { parts_to_remove = removePartsInRangeFromWorkingSetAndGetPartsToRemoveFromZooKeeper(NO_TRANSACTION_RAW, drop_range, data_parts_lock); String parts_to_remove_str; for (const auto & part : parts_to_remove) { parts_to_remove_str += part.getPartName(); parts_to_remove_str += " "; } LOG_TRACE(log, "Replacing {} parts {}with {} parts {}", parts_to_remove.size(), parts_to_remove_str, final_parts.size(), boost::algorithm::join(final_part_names, ", ")); } } PartLog::addNewParts(getContext(), PartLog::createPartLogEntries(res_parts, watch.elapsed(), profile_events_scope.getSnapshot())); } catch (...) { PartLog::addNewParts(getContext(), PartLog::createPartLogEntries(res_parts, watch.elapsed()), ExecutionStatus::fromCurrentException("", true)); for (const auto & res_part : res_parts) unlockSharedData(*res_part); throw; } removePartsFromZooKeeperWithRetries(parts_to_remove); res_parts.clear(); parts_to_remove.clear(); cleanup_thread.wakeup(); return true; } void StorageReplicatedMergeTree::executeClonePartFromShard(const LogEntry & entry) { auto zookeeper = getZooKeeper(); Strings replicas = zookeeper->getChildren(entry.source_shard + "/replicas"); std::shuffle(replicas.begin(), replicas.end(), thread_local_rng); String replica; for (const String & candidate : replicas) { if (zookeeper->exists(entry.source_shard + "/replicas/" + candidate + "/is_active")) { replica = candidate; break; } } if (replica.empty()) throw Exception(ErrorCodes::NO_REPLICA_HAS_PART, "Not found active replica on shard {} to clone part {}", entry.source_shard, entry.new_part_name); LOG_INFO(log, "Will clone part from shard {} and replica {}", entry.source_shard, replica); MutableDataPartPtr part; { auto metadata_snapshot = getInMemoryMetadataPtr(); String source_replica_path = entry.source_shard + "/replicas/" + replica; ReplicatedMergeTreeAddress address(getZooKeeper()->get(source_replica_path + "/host")); auto timeouts = getHTTPTimeouts(getContext()); auto credentials = getContext()->getInterserverCredentials(); String interserver_scheme = getContext()->getInterserverScheme(); auto get_part = [&, address, timeouts, credentials, interserver_scheme]() { if (interserver_scheme != address.scheme) throw Exception(ErrorCodes::LOGICAL_ERROR, "Interserver schemes are different: '{}' != '{}', can't fetch part from {}", interserver_scheme, address.scheme, address.host); return fetcher.fetchSelectedPart( metadata_snapshot, getContext(), entry.new_part_name, source_replica_path, address.host, address.replication_port, timeouts, credentials->getUser(), credentials->getPassword(), interserver_scheme, replicated_fetches_throttler, true); }; part = get_part(); // The fetched part is valuable and should not be cleaned like a temp part. part->is_temp = false; part->renameTo("detached/" + entry.new_part_name, true); LOG_INFO(log, "Cloned part {} to detached directory", part->name); } } void StorageReplicatedMergeTree::cloneReplica(const String & source_replica, Coordination::Stat source_is_lost_stat, zkutil::ZooKeeperPtr & zookeeper) { String source_path = fs::path(zookeeper_path) / "replicas" / source_replica; /// The order of the following three actions is important. Strings source_queue_names; /// We are trying to get consistent /log_pointer and /queue state. Otherwise /// we can possibly duplicate entries in queue of cloned replica. while (true) { Coordination::Stat log_pointer_stat; String raw_log_pointer = zookeeper->get(fs::path(source_path) / "log_pointer", &log_pointer_stat); Coordination::Requests ops; ops.push_back(zkutil::makeSetRequest(fs::path(replica_path) / "log_pointer", raw_log_pointer, -1)); /// For support old versions CH. if (source_is_lost_stat.version == -1) { /// We check that it was not suddenly upgraded to new version. /// Otherwise it can be upgraded and instantly become lost, but we cannot notice that. ops.push_back(zkutil::makeCreateRequest(fs::path(source_path) / "is_lost", "0", zkutil::CreateMode::Persistent)); ops.push_back(zkutil::makeRemoveRequest(fs::path(source_path) / "is_lost", -1)); } else /// The replica we clone should not suddenly become lost. ops.push_back(zkutil::makeCheckRequest(fs::path(source_path) / "is_lost", source_is_lost_stat.version)); Coordination::Responses responses; /// Let's remember the queue of the reference/master replica. source_queue_names = zookeeper->getChildren(fs::path(source_path) / "queue"); /// Check that log pointer of source replica didn't changed while we read queue entries ops.push_back(zkutil::makeCheckRequest(fs::path(source_path) / "log_pointer", log_pointer_stat.version)); auto rc = zookeeper->tryMulti(ops, responses); if (rc == Coordination::Error::ZOK) { break; } else if (rc == Coordination::Error::ZNODEEXISTS) { throw Exception(ErrorCodes::REPLICA_STATUS_CHANGED, "Can not clone replica, because the {} updated to new ClickHouse version", source_replica); } else if (responses[1]->error == Coordination::Error::ZBADVERSION) { /// If is_lost node version changed than source replica also lost, /// so we cannot clone from it. throw Exception(ErrorCodes::REPLICA_STATUS_CHANGED, "Can not clone replica, because the {} became lost", source_replica); } else if (responses.back()->error == Coordination::Error::ZBADVERSION) { /// If source replica's log_pointer changed than we probably read /// stale state of /queue and have to try one more time. LOG_WARNING(log, "Log pointer of source replica {} changed while we loading queue nodes. Will retry.", source_replica); continue; } else { zkutil::KeeperMultiException::check(rc, ops, responses); } } ::sort(source_queue_names.begin(), source_queue_names.end()); struct QueueEntryInfo { String data = {}; Coordination::Stat stat = {}; LogEntryPtr parsed_entry = {}; }; /// We got log pointer and list of queue entries of source replica. /// At first we will get queue entries and then we will get list of active parts of source replica /// to enqueue fetches for missing parts. If source replica executes and removes some entry concurrently /// we will see produced part (or covering part) in replicas/source/parts and will enqueue fetch. /// We will try to parse queue entries before copying them /// to avoid creation of excessive and duplicating entries in our queue. /// See also removePartAndEnqueueFetch(...) std::vector source_queue; ActiveDataPartSet get_part_set{format_version}; ActiveDataPartSet drop_range_set{format_version}; std::unordered_set exact_part_names; { std::vector queue_get_futures; queue_get_futures.reserve(source_queue_names.size()); for (const String & entry_name : source_queue_names) queue_get_futures.push_back(zookeeper->asyncTryGet(fs::path(source_path) / "queue" / entry_name)); source_queue.reserve(source_queue_names.size()); for (size_t i = 0; i < source_queue_names.size(); ++i) { auto res = queue_get_futures[i].get(); /// It's ok if entry is already executed and removed: we also will get source parts set. if (res.error == Coordination::Error::ZNONODE) continue; assert(res.error == Coordination::Error::ZOK); source_queue.emplace_back(); auto & info = source_queue.back(); info.data = std::move(res.data); info.stat = std::move(res.stat); try { info.parsed_entry = LogEntry::parse(info.data, info.stat, format_version); } catch (...) { tryLogCurrentException(log, "Cannot parse source queue entry " + source_queue_names[i]); } /// It may be ok if source replica has newer version. We will copy entry as is. if (!info.parsed_entry) continue; info.parsed_entry->znode_name = source_queue_names[i]; if (info.parsed_entry->type == LogEntry::DROP_RANGE || info.parsed_entry->type == LogEntry::DROP_PART) { drop_range_set.add(info.parsed_entry->new_part_name); } else if (info.parsed_entry->type == LogEntry::GET_PART) { String maybe_covering_drop_range = drop_range_set.getContainingPart(info.parsed_entry->new_part_name); if (maybe_covering_drop_range.empty()) get_part_set.add(info.parsed_entry->new_part_name); } else { /// We should keep local parts if they present in the queue of source replica. /// There's a chance that we are the only replica that has these parts. Strings entry_virtual_parts = info.parsed_entry->getVirtualPartNames(format_version); std::move(entry_virtual_parts.begin(), entry_virtual_parts.end(), std::inserter(exact_part_names, exact_part_names.end())); } } } /// We should do it after copying queue, because some ALTER_METADATA entries can be lost otherwise. cloneMetadataIfNeeded(source_replica, source_path, zookeeper); /// Add to the queue jobs to receive all the active parts that the reference/master replica has. Strings source_replica_parts = zookeeper->getChildren(fs::path(source_path) / "parts"); for (const auto & active_part : source_replica_parts) get_part_set.add(active_part); Strings active_parts = get_part_set.getParts(); /// Remove local parts if source replica does not have them, because such parts will never be fetched by other replicas. Strings local_parts_in_zk = zookeeper->getChildren(fs::path(replica_path) / "parts"); Strings parts_to_remove_from_zk; for (const auto & part : local_parts_in_zk) { /// We look for exact match (and not for any covering part) /// because our part might be dropped and covering part might be merged though gap. /// (avoid resurrection of data that was removed a long time ago) if (get_part_set.getContainingPart(part) == part) continue; if (exact_part_names.contains(part)) continue; parts_to_remove_from_zk.emplace_back(part); LOG_WARNING(log, "Source replica does not have part {}. Removing it from ZooKeeper.", part); } { /// Check "is_lost" version after retrieving queue and parts. /// If version has changed, then replica most likely has been dropped and parts set is inconsistent, /// so throw exception and retry cloning. Coordination::Stat is_lost_stat_new; zookeeper->get(fs::path(source_path) / "is_lost", &is_lost_stat_new); if (is_lost_stat_new.version != source_is_lost_stat.version) throw Exception(ErrorCodes::REPLICA_STATUS_CHANGED, "Cannot clone {}, because it suddenly become lost " "or removed broken part from ZooKeeper", source_replica); } removePartsFromZooKeeperWithRetries(parts_to_remove_from_zk); auto local_active_parts = getDataPartsForInternalUsage(); DataPartsVector parts_to_remove_from_working_set; for (const auto & part : local_active_parts) { if (get_part_set.getContainingPart(part->name) == part->name) continue; if (exact_part_names.contains(part->name)) continue; parts_to_remove_from_working_set.emplace_back(part); LOG_WARNING(log, "Source replica does not have part {}. Removing it from working set.", part->name); } if (getSettings()->detach_old_local_parts_when_cloning_replica) { auto metadata_snapshot = getInMemoryMetadataPtr(); for (const auto & part : parts_to_remove_from_working_set) { LOG_INFO(log, "Detaching {}", part->getDataPartStorage().getPartDirectory()); part->makeCloneInDetached("clone", metadata_snapshot); } } removePartsFromWorkingSet(NO_TRANSACTION_RAW, parts_to_remove_from_working_set, true); std::unordered_set created_get_parts; /// Avoid creation of GET_PART entries which covered by another GET_PART or DROP_RANGE /// and creation of multiple entries with the same new_part_name. auto should_ignore_log_entry = [&drop_range_set, &get_part_set, this] (std::unordered_set & created_gets, const String & part_name, const String & log_msg_context) -> bool { /// We should not create entries covered by DROP_RANGE, because we will remove them anyway (kind of optimization). String covering_drop_range = drop_range_set.getContainingPart(part_name); if (!covering_drop_range.empty()) { LOG_TRACE(log, "{} {}: it's covered by drop range {}", log_msg_context, part_name, covering_drop_range); return true; } /// We should not create entries covered by GET_PART, /// because GET_PART entry has no source parts and we can execute it only by fetching. /// Parts covered by GET_PART are useless and may cause replication to stuck if covered part is lost. String covering_get_part_entry = get_part_set.getContainingPart(part_name); if (covering_get_part_entry.empty()) return false; if (covering_get_part_entry != part_name) { LOG_TRACE(log, "{} {}: it's covered by GET_PART {}", log_msg_context, part_name, covering_get_part_entry); return true; } /// NOTE: It does not completely avoids duplication of GET_PART entries, /// because it's possible that source replica has executed some GET_PART after we copied it's queue, /// but before we copied its active parts set. In this case we will GET_PART entry in our queue /// and later will pull the original GET_PART from replication log. /// It should not cause any issues, but it does not allow to get rid of duplicated entries and add an assertion. if (created_gets.contains(part_name)) { /// NOTE It would be better to copy log entry instead of creating GET_PART /// if there are GET_PART and log entry of other type with the same new_part_name. /// But it's a bit harder to implement, because it requires full-fledged virtual_parts set. LOG_TRACE(log, "{} {}: GET_PART for it is already created", log_msg_context, part_name); return true; } return false; }; for (const String & name : active_parts) { if (should_ignore_log_entry(created_get_parts, name, "Not fetching")) continue; LogEntry log_entry; if (are_restoring_replica) { LOG_DEBUG(log, "Obtaining checksum for path {}", name); // The part we want to fetch is probably present in detached/ folder. // However, we need to get part's checksum to check if it's not corrupt. log_entry.type = LogEntry::ATTACH_PART; MinimalisticDataPartChecksums desired_checksums; const fs::path part_path = fs::path(source_path) / "parts" / name; const String part_znode = zookeeper->get(part_path); if (!part_znode.empty()) desired_checksums = ReplicatedMergeTreePartHeader::fromString(part_znode).getChecksums(); else { String desired_checksums_str = zookeeper->get(part_path / "checksums"); desired_checksums = MinimalisticDataPartChecksums::deserializeFrom(desired_checksums_str); } const auto [lo, hi] = desired_checksums.hash_of_all_files; log_entry.part_checksum = getHexUIntUppercase(hi) + getHexUIntUppercase(lo); } else { log_entry.type = LogEntry::GET_PART; } log_entry.source_replica = ""; log_entry.new_part_name = name; log_entry.create_time = tryGetPartCreateTime(zookeeper, source_path, name); LOG_TEST(log, "Enqueueing {} for fetch", name); zookeeper->create(fs::path(replica_path) / "queue/queue-", log_entry.toString(), zkutil::CreateMode::PersistentSequential); created_get_parts.insert(name); } size_t total_parts_to_fetch = created_get_parts.size(); LOG_DEBUG(log, "Queued {} parts to be fetched, {} parts ignored", total_parts_to_fetch, active_parts.size() - total_parts_to_fetch); /// Add content of the reference/master replica queue to the queue. size_t total_entries_to_copy = 0; for (const auto & entry_info : source_queue) { assert(!entry_info.data.empty()); if (entry_info.parsed_entry && !entry_info.parsed_entry->new_part_name.empty()) { const String & part_name = entry_info.parsed_entry->new_part_name; const String & entry_name = entry_info.parsed_entry->znode_name; const auto & entry_type = entry_info.parsed_entry->type; if (should_ignore_log_entry(created_get_parts, part_name, fmt::format("Not copying {} {} ", entry_name, entry_type))) continue; if (entry_info.parsed_entry->type == LogEntry::GET_PART) created_get_parts.insert(part_name); } LOG_TEST(log, "Copying entry {}", entry_info.data); zookeeper->create(fs::path(replica_path) / "queue/queue-", entry_info.data, zkutil::CreateMode::PersistentSequential); ++total_entries_to_copy; } LOG_DEBUG(log, "Copied {} queue entries, {} entries ignored", total_entries_to_copy, source_queue.size() - total_entries_to_copy); } void StorageReplicatedMergeTree::cloneMetadataIfNeeded(const String & source_replica, const String & source_path, zkutil::ZooKeeperPtr & zookeeper) { String source_metadata_version_str; bool metadata_version_exists = zookeeper->tryGet(source_path + "/metadata_version", source_metadata_version_str); if (!metadata_version_exists) { /// For compatibility with version older than 20.3 /// TODO fix tests and delete it LOG_WARNING(log, "Node {} does not exist. " "Most likely it's because too old version of ClickHouse is running on replica {}. " "Will not check metadata consistency", source_path + "/metadata_version", source_replica); return; } Int32 source_metadata_version = parse(source_metadata_version_str); if (metadata_version == source_metadata_version) return; /// Our metadata it not up to date with source replica metadata. /// Metadata is updated by ALTER_METADATA entries, but some entries are probably cleaned up from the log. /// It's also possible that some newer ALTER_METADATA entries are present in source_queue list, /// and source replica are executing such entry right now (or had executed recently). /// More than that, /metadata_version update is not atomic with /columns and /metadata update... /// Fortunately, ALTER_METADATA seems to be idempotent, /// and older entries of such type can be replaced with newer entries. /// Let's try to get consistent values of source replica's /columns and /metadata /// and prepend dummy ALTER_METADATA to our replication queue. /// It should not break anything if source_queue already contains ALTER_METADATA entry /// with greater or equal metadata_version, but it will update our metadata /// if all such entries were cleaned up from the log and source_queue. LOG_WARNING(log, "Metadata version ({}) on replica is not up to date with metadata ({}) on source replica {}", metadata_version, source_metadata_version, source_replica); String source_metadata; String source_columns; while (true) { Coordination::Stat metadata_stat; Coordination::Stat columns_stat; source_metadata = zookeeper->get(source_path + "/metadata", &metadata_stat); source_columns = zookeeper->get(source_path + "/columns", &columns_stat); Coordination::Requests ops; Coordination::Responses responses; ops.emplace_back(zkutil::makeCheckRequest(source_path + "/metadata", metadata_stat.version)); ops.emplace_back(zkutil::makeCheckRequest(source_path + "/columns", columns_stat.version)); Coordination::Error code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZOK) break; else if (code == Coordination::Error::ZBADVERSION) LOG_WARNING(log, "Metadata of replica {} was changed", source_path); else zkutil::KeeperMultiException::check(code, ops, responses); } ReplicatedMergeTreeLogEntryData dummy_alter; dummy_alter.type = LogEntry::ALTER_METADATA; dummy_alter.source_replica = source_replica; dummy_alter.metadata_str = source_metadata; dummy_alter.columns_str = source_columns; dummy_alter.alter_version = source_metadata_version; dummy_alter.create_time = time(nullptr); zookeeper->create(replica_path + "/queue/queue-", dummy_alter.toString(), zkutil::CreateMode::PersistentSequential); /// We don't need to do anything with mutation_pointer, because mutation log cleanup process is different from /// replication log cleanup. A mutation is removed from ZooKeeper only if all replicas had executed the mutation, /// so all mutations which are greater or equal to our mutation pointer are still present in ZooKeeper. } void StorageReplicatedMergeTree::cloneReplicaIfNeeded(zkutil::ZooKeeperPtr zookeeper) { Coordination::Stat is_lost_stat; bool is_new_replica = true; String res; if (zookeeper->tryGet(fs::path(replica_path) / "is_lost", res, &is_lost_stat)) { if (res == "0") return; if (is_lost_stat.version) is_new_replica = false; } else { /// Replica was created by old version of CH, so me must create "/is_lost". /// Note that in old version of CH there was no "lost" replicas possible. /// TODO is_lost node should always exist since v18.12, maybe we can replace `tryGet` with `get` and remove old code? zookeeper->create(fs::path(replica_path) / "is_lost", "0", zkutil::CreateMode::Persistent); return; } /// is_lost is "1": it means that we are in repair mode. /// Try choose source replica to clone. /// Source replica must not be lost and should have minimal queue size and maximal log pointer. Strings replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); std::vector futures; for (const String & source_replica_name : replicas) { /// Do not clone from myself. if (source_replica_name == replica_name) continue; String source_replica_path = fs::path(zookeeper_path) / "replicas" / source_replica_name; /// Obviously the following get operations are not atomic, but it's ok to choose good enough replica, not the best one. /// NOTE: We may count some entries twice if log_pointer is moved. futures.emplace_back(zookeeper->asyncTryGet(fs::path(source_replica_path) / "is_lost")); futures.emplace_back(zookeeper->asyncTryGet(fs::path(source_replica_path) / "log_pointer")); futures.emplace_back(zookeeper->asyncTryGet(fs::path(source_replica_path) / "queue")); } /// Wait for results before getting log entries for (auto & future : futures) future.wait(); Strings log_entries = zookeeper->getChildren(fs::path(zookeeper_path) / "log"); size_t max_log_entry = 0; if (!log_entries.empty()) { String last_entry = *std::max_element(log_entries.begin(), log_entries.end()); max_log_entry = parse(last_entry.substr(strlen("log-"))); } /// log_pointer can point to future entry, which was not created yet ++max_log_entry; size_t min_replication_lag = std::numeric_limits::max(); String source_replica; Coordination::Stat source_is_lost_stat; size_t future_num = 0; for (const String & source_replica_name : replicas) { if (source_replica_name == replica_name) continue; auto get_is_lost = futures[future_num++].get(); auto get_log_pointer = futures[future_num++].get(); auto get_queue = futures[future_num++].get(); if (get_is_lost.error != Coordination::Error::ZOK) { LOG_INFO(log, "Not cloning {}, cannot get '/is_lost': {}", source_replica_name, Coordination::errorMessage(get_is_lost.error)); continue; } else if (get_is_lost.data != "0") { LOG_INFO(log, "Not cloning {}, it's lost", source_replica_name); continue; } if (get_log_pointer.error != Coordination::Error::ZOK) { LOG_INFO(log, "Not cloning {}, cannot get '/log_pointer': {}", source_replica_name, Coordination::errorMessage(get_log_pointer.error)); continue; } if (get_queue.error != Coordination::Error::ZOK) { LOG_INFO(log, "Not cloning {}, cannot get '/queue': {}", source_replica_name, Coordination::errorMessage(get_queue.error)); continue; } /// Replica is not lost and we can clone it. Let's calculate approx replication lag. size_t source_log_pointer = get_log_pointer.data.empty() ? 0 : parse(get_log_pointer.data); assert(source_log_pointer <= max_log_entry); size_t replica_queue_lag = max_log_entry - source_log_pointer; size_t replica_queue_size = get_queue.stat.numChildren; size_t replication_lag = replica_queue_lag + replica_queue_size; LOG_INFO(log, "Replica {} has log pointer '{}', approximate {} queue lag and {} queue size", source_replica_name, get_log_pointer.data, replica_queue_lag, replica_queue_size); if (replication_lag < min_replication_lag) { source_replica = source_replica_name; source_is_lost_stat = get_is_lost.stat; min_replication_lag = replication_lag; } } if (source_replica.empty()) throw Exception(ErrorCodes::ALL_REPLICAS_LOST, "All replicas are lost"); if (is_new_replica) LOG_INFO(log, "Will mimic {}", source_replica); else LOG_WARNING(log, "Will mimic {}", source_replica); /// Clear obsolete queue that we no longer need. zookeeper->removeChildren(fs::path(replica_path) / "queue"); queue.clear(); /// Will do repair from the selected replica. cloneReplica(source_replica, source_is_lost_stat, zookeeper); /// If repair fails to whatever reason, the exception is thrown, is_lost will remain "1" and the replica will be repaired later. /// If replica is repaired successfully, we remove is_lost flag. zookeeper->set(fs::path(replica_path) / "is_lost", "0"); } String StorageReplicatedMergeTree::getLastQueueUpdateException() const { std::lock_guard lock(last_queue_update_exception_lock); return last_queue_update_exception; } void StorageReplicatedMergeTree::queueUpdatingTask() { if (!queue_update_in_progress) { last_queue_update_start_time.store(time(nullptr)); queue_update_in_progress = true; } try { queue.pullLogsToQueue(getZooKeeperAndAssertNotReadonly(), queue_updating_task->getWatchCallback(), ReplicatedMergeTreeQueue::UPDATE); last_queue_update_finish_time.store(time(nullptr)); queue_update_in_progress = false; } catch (const Coordination::Exception & e) { tryLogCurrentException(log, __PRETTY_FUNCTION__); std::lock_guard lock(last_queue_update_exception_lock); last_queue_update_exception = getCurrentExceptionMessage(false); if (e.code == Coordination::Error::ZSESSIONEXPIRED) { restarting_thread.wakeup(); return; } queue_updating_task->scheduleAfter(QUEUE_UPDATE_ERROR_SLEEP_MS); } catch (...) { tryLogCurrentException(log, __PRETTY_FUNCTION__); std::lock_guard lock(last_queue_update_exception_lock); last_queue_update_exception = getCurrentExceptionMessage(false); queue_updating_task->scheduleAfter(QUEUE_UPDATE_ERROR_SLEEP_MS); } } void StorageReplicatedMergeTree::mutationsUpdatingTask() { try { queue.updateMutations(getZooKeeper(), mutations_updating_task->getWatchCallback()); } catch (const Coordination::Exception & e) { tryLogCurrentException(log, __PRETTY_FUNCTION__); if (e.code == Coordination::Error::ZSESSIONEXPIRED) return; mutations_updating_task->scheduleAfter(QUEUE_UPDATE_ERROR_SLEEP_MS); } catch (...) { tryLogCurrentException(log, __PRETTY_FUNCTION__); mutations_updating_task->scheduleAfter(QUEUE_UPDATE_ERROR_SLEEP_MS); } } ReplicatedMergeTreeQueue::SelectedEntryPtr StorageReplicatedMergeTree::selectQueueEntry() { /// This object will mark the element of the queue as running. ReplicatedMergeTreeQueue::SelectedEntryPtr selected; try { selected = queue.selectEntryToProcess(merger_mutator, *this); } catch (...) { tryLogCurrentException(log, __PRETTY_FUNCTION__); } return selected; } bool StorageReplicatedMergeTree::processQueueEntry(ReplicatedMergeTreeQueue::SelectedEntryPtr selected_entry) { LogEntryPtr & entry = selected_entry->log_entry; return queue.processEntry([this]{ return getZooKeeper(); }, entry, [&](LogEntryPtr & entry_to_process) { try { return executeLogEntry(*entry_to_process); } catch (const Exception & e) { if (e.code() == ErrorCodes::NO_REPLICA_HAS_PART) { /// If no one has the right part, probably not all replicas work; We will not write to log with Error level. LOG_INFO(log, getExceptionMessageAndPattern(e, /* with_stacktrace */ false)); } else if (e.code() == ErrorCodes::ABORTED) { /// Interrupted merge or downloading a part is not an error. LOG_INFO(log, getExceptionMessageAndPattern(e, /* with_stacktrace */ false)); } else if (e.code() == ErrorCodes::PART_IS_TEMPORARILY_LOCKED) { /// Part cannot be added temporarily LOG_INFO(log, getExceptionMessageAndPattern(e, /* with_stacktrace */ false)); cleanup_thread.wakeup(); } else tryLogCurrentException(log, __PRETTY_FUNCTION__); /** This exception will be written to the queue element, and it can be looked up using `system.replication_queue` table. * The thread that performs this action will sleep a few seconds after the exception. * See `queue.processEntry` function. */ throw; } catch (...) { tryLogCurrentException(log, __PRETTY_FUNCTION__); throw; } }); } bool StorageReplicatedMergeTree::scheduleDataProcessingJob(BackgroundJobsAssignee & assignee) { /// If replication queue is stopped exit immediately as we successfully executed the task if (queue.actions_blocker.isCancelled()) return false; /// This object will mark the element of the queue as running. ReplicatedMergeTreeQueue::SelectedEntryPtr selected_entry = selectQueueEntry(); if (!selected_entry) return false; auto job_type = selected_entry->log_entry->type; /// Depending on entry type execute in fetches (small) pool or big merge_mutate pool if (job_type == LogEntry::GET_PART) { assignee.scheduleFetchTask(std::make_shared( [this, selected_entry] () mutable { return processQueueEntry(selected_entry); }, common_assignee_trigger, getStorageID())); return true; } else if (job_type == LogEntry::MERGE_PARTS) { auto task = std::make_shared(selected_entry, *this, common_assignee_trigger); assignee.scheduleMergeMutateTask(task); return true; } else if (job_type == LogEntry::MUTATE_PART) { auto task = std::make_shared(selected_entry, *this, common_assignee_trigger); assignee.scheduleMergeMutateTask(task); return true; } else { assignee.scheduleCommonTask(std::make_shared( [this, selected_entry] () mutable { return processQueueEntry(selected_entry); }, common_assignee_trigger, getStorageID()), /* need_trigger */ true); return true; } } bool StorageReplicatedMergeTree::canExecuteFetch(const ReplicatedMergeTreeLogEntry & entry, String & disable_reason) const { if (fetcher.blocker.isCancelled()) { disable_reason = fmt::format("Not executing fetch of part {} because replicated fetches are cancelled now.", entry.new_part_name); return false; } auto replicated_fetches_pool_size = getContext()->getFetchesExecutor()->getMaxTasksCount(); size_t busy_threads_in_pool = CurrentMetrics::values[CurrentMetrics::BackgroundFetchesPoolTask].load(std::memory_order_relaxed); if (busy_threads_in_pool >= replicated_fetches_pool_size) { disable_reason = fmt::format("Not executing fetch of part {} because {} fetches already executing, max {}.", entry.new_part_name, busy_threads_in_pool, replicated_fetches_pool_size); return false; } if (replicated_fetches_throttler->isThrottling()) { disable_reason = fmt::format("Not executing fetch of part {} because fetches have already throttled by network settings " " or .", entry.new_part_name); return false; } return true; } bool StorageReplicatedMergeTree::partIsAssignedToBackgroundOperation(const DataPartPtr & part) const { return queue.isVirtualPart(part); } void StorageReplicatedMergeTree::mergeSelectingTask() { if (!is_leader) return; const auto storage_settings_ptr = getSettings(); const bool deduplicate = false; /// TODO: read deduplicate option from table config const Names deduplicate_by_columns = {}; const bool cleanup = (storage_settings_ptr->clean_deleted_rows != CleanDeletedRows::Never); CreateMergeEntryResult create_result = CreateMergeEntryResult::Other; try { /// We must select parts for merge under merge_selecting_mutex because other threads /// (OPTIMIZE queries) can assign new merges. std::lock_guard merge_selecting_lock(merge_selecting_mutex); auto zookeeper = getZooKeeperAndAssertNotReadonly(); ReplicatedMergeTreeMergePredicate merge_pred = queue.getMergePredicate(zookeeper, getAllPartitionIds()); /// If many merges is already queued, then will queue only small enough merges. /// Otherwise merge queue could be filled with only large merges, /// and in the same time, many small parts could be created and won't be merged. auto merges_and_mutations_queued = queue.countMergesAndPartMutations(); size_t merges_and_mutations_sum = merges_and_mutations_queued.merges + merges_and_mutations_queued.mutations; if (merges_and_mutations_sum >= storage_settings_ptr->max_replicated_merges_in_queue) { LOG_TRACE(log, "Number of queued merges ({}) and part mutations ({})" " is greater than max_replicated_merges_in_queue ({}), so won't select new parts to merge or mutate.", merges_and_mutations_queued.merges, merges_and_mutations_queued.mutations, storage_settings_ptr->max_replicated_merges_in_queue); } else { UInt64 max_source_parts_size_for_merge = merger_mutator.getMaxSourcePartsSizeForMerge( storage_settings_ptr->max_replicated_merges_in_queue, merges_and_mutations_sum); UInt64 max_source_part_size_for_mutation = merger_mutator.getMaxSourcePartSizeForMutation(); bool merge_with_ttl_allowed = merges_and_mutations_queued.merges_with_ttl < storage_settings_ptr->max_replicated_merges_with_ttl_in_queue && getTotalMergesWithTTLInMergeList() < storage_settings_ptr->max_number_of_merges_with_ttl_in_pool; auto future_merged_part = std::make_shared(); if (storage_settings.get()->assign_part_uuids) future_merged_part->uuid = UUIDHelpers::generateV4(); if (max_source_parts_size_for_merge > 0 && merger_mutator.selectPartsToMerge(future_merged_part, false, max_source_parts_size_for_merge, merge_pred, merge_with_ttl_allowed, NO_TRANSACTION_PTR, nullptr) == SelectPartsDecision::SELECTED) { create_result = createLogEntryToMergeParts( zookeeper, future_merged_part->parts, future_merged_part->name, future_merged_part->uuid, future_merged_part->part_format, deduplicate, deduplicate_by_columns, cleanup, nullptr, merge_pred.getVersion(), future_merged_part->merge_type); } /// If there are many mutations in queue, it may happen, that we cannot enqueue enough merges to merge all new parts else if (max_source_part_size_for_mutation > 0 && queue.countMutations() > 0 && merges_and_mutations_queued.mutations < storage_settings_ptr->max_replicated_mutations_in_queue) { /// Choose a part to mutate. DataPartsVector data_parts = getDataPartsVectorForInternalUsage(); for (const auto & part : data_parts) { if (part->getBytesOnDisk() > max_source_part_size_for_mutation) continue; std::optional> desired_mutation_version = merge_pred.getDesiredMutationVersion(part); if (!desired_mutation_version) continue; create_result = createLogEntryToMutatePart( *part, future_merged_part->uuid, desired_mutation_version->first, desired_mutation_version->second, merge_pred.getVersion()); if (create_result == CreateMergeEntryResult::Ok || create_result == CreateMergeEntryResult::LogUpdated) break; } } } } catch (...) { tryLogCurrentException(log, __PRETTY_FUNCTION__); } if (!is_leader) return; if (create_result != CreateMergeEntryResult::Ok && create_result != CreateMergeEntryResult::LogUpdated) { merge_selecting_task->scheduleAfter(storage_settings_ptr->merge_selecting_sleep_ms); } else { merge_selecting_task->schedule(); } } void StorageReplicatedMergeTree::mutationsFinalizingTask() { bool needs_reschedule = false; try { needs_reschedule = queue.tryFinalizeMutations(getZooKeeperAndAssertNotReadonly()); } catch (...) { tryLogCurrentException(log, __PRETTY_FUNCTION__); needs_reschedule = true; } if (needs_reschedule) { mutations_finalizing_task->scheduleAfter(MUTATIONS_FINALIZING_SLEEP_MS); } else { /// Even if no mutations seems to be done or appeared we are trying to /// finalize them in background because manual control the launch of /// this function is error prone. This can lead to mutations that /// processed all the parts but have is_done=0 state for a long time. Or /// killed mutations, which are also considered as undone. mutations_finalizing_task->scheduleAfter(MUTATIONS_FINALIZING_IDLE_SLEEP_MS); } } StorageReplicatedMergeTree::CreateMergeEntryResult StorageReplicatedMergeTree::createLogEntryToMergeParts( zkutil::ZooKeeperPtr & zookeeper, const DataPartsVector & parts, const String & merged_name, const UUID & merged_part_uuid, const MergeTreeDataPartFormat & merged_part_format, bool deduplicate, const Names & deduplicate_by_columns, bool cleanup, ReplicatedMergeTreeLogEntryData * out_log_entry, int32_t log_version, MergeType merge_type) { Strings exists_paths; exists_paths.reserve(parts.size()); for (const auto & part : parts) exists_paths.emplace_back(fs::path(replica_path) / "parts" / part->name); auto exists_results = zookeeper->exists(exists_paths); bool all_in_zk = true; for (size_t i = 0; i < parts.size(); ++i) { /// If there is no information about part in ZK, we will not merge it. if (exists_results[i].error == Coordination::Error::ZNONODE) { all_in_zk = false; const auto & part = parts[i]; if (part->modification_time + MAX_AGE_OF_LOCAL_PART_THAT_WASNT_ADDED_TO_ZOOKEEPER < time(nullptr)) { LOG_WARNING(log, "Part {} (that was selected for merge) with age {} seconds exists locally but not in ZooKeeper. Won't do merge with that part and will check it.", part->name, (time(nullptr) - part->modification_time)); enqueuePartForCheck(part->name); } } } if (!all_in_zk) return CreateMergeEntryResult::MissingPart; ReplicatedMergeTreeLogEntryData entry; entry.type = LogEntry::MERGE_PARTS; entry.source_replica = replica_name; entry.new_part_name = merged_name; entry.new_part_uuid = merged_part_uuid; entry.new_part_format = merged_part_format; entry.merge_type = merge_type; entry.deduplicate = deduplicate; entry.deduplicate_by_columns = deduplicate_by_columns; entry.cleanup = cleanup; entry.create_time = time(nullptr); for (const auto & part : parts) entry.source_parts.push_back(part->name); Coordination::Requests ops; Coordination::Responses responses; ops.emplace_back(zkutil::makeCreateRequest( fs::path(zookeeper_path) / "log/log-", entry.toString(), zkutil::CreateMode::PersistentSequential)); ops.emplace_back(zkutil::makeSetRequest( fs::path(zookeeper_path) / "log", "", log_version)); /// Check and update version. Coordination::Error code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZOK) { String path_created = dynamic_cast(*responses.front()).path_created; entry.znode_name = path_created.substr(path_created.find_last_of('/') + 1); ProfileEvents::increment(ProfileEvents::CreatedLogEntryForMerge); LOG_TRACE(log, "Created log entry {} for merge {}", path_created, merged_name); } else if (code == Coordination::Error::ZBADVERSION) { ProfileEvents::increment(ProfileEvents::NotCreatedLogEntryForMerge); LOG_TRACE(log, "Log entry is not created for merge {} because log was updated", merged_name); return CreateMergeEntryResult::LogUpdated; } else { zkutil::KeeperMultiException::check(code, ops, responses); } if (out_log_entry) *out_log_entry = entry; return CreateMergeEntryResult::Ok; } StorageReplicatedMergeTree::CreateMergeEntryResult StorageReplicatedMergeTree::createLogEntryToMutatePart( const IMergeTreeDataPart & part, const UUID & new_part_uuid, Int64 mutation_version, int32_t alter_version, int32_t log_version) { auto zookeeper = getZooKeeper(); /// If there is no information about part in ZK, we will not mutate it. if (!zookeeper->exists(fs::path(replica_path) / "parts" / part.name)) { if (part.modification_time + MAX_AGE_OF_LOCAL_PART_THAT_WASNT_ADDED_TO_ZOOKEEPER < time(nullptr)) { LOG_WARNING(log, "Part {} (that was selected for mutation) with age {} seconds exists locally but not in ZooKeeper." " Won't mutate that part and will check it.", part.name, (time(nullptr) - part.modification_time)); enqueuePartForCheck(part.name); } return CreateMergeEntryResult::MissingPart; } MergeTreePartInfo new_part_info = part.info; new_part_info.mutation = mutation_version; String new_part_name = part.getNewName(new_part_info); ReplicatedMergeTreeLogEntryData entry; entry.type = LogEntry::MUTATE_PART; entry.source_replica = replica_name; entry.source_parts.push_back(part.name); entry.new_part_name = new_part_name; entry.new_part_uuid = new_part_uuid; entry.create_time = time(nullptr); entry.alter_version = alter_version; Coordination::Requests ops; Coordination::Responses responses; ops.emplace_back(zkutil::makeCreateRequest( fs::path(zookeeper_path) / "log/log-", entry.toString(), zkutil::CreateMode::PersistentSequential)); ops.emplace_back(zkutil::makeSetRequest( fs::path(zookeeper_path) / "log", "", log_version)); /// Check and update version. Coordination::Error code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZBADVERSION) { ProfileEvents::increment(ProfileEvents::NotCreatedLogEntryForMutation); LOG_TRACE(log, "Log entry is not created for mutation {} because log was updated", new_part_name); return CreateMergeEntryResult::LogUpdated; } zkutil::KeeperMultiException::check(code, ops, responses); ProfileEvents::increment(ProfileEvents::CreatedLogEntryForMutation); LOG_TRACE(log, "Created log entry for mutation {}", new_part_name); return CreateMergeEntryResult::Ok; } void StorageReplicatedMergeTree::getRemovePartFromZooKeeperOps(const String & part_name, Coordination::Requests & ops, bool has_children) { String part_path = fs::path(replica_path) / "parts" / part_name; if (has_children) { ops.emplace_back(zkutil::makeRemoveRequest(fs::path(part_path) / "checksums", -1)); ops.emplace_back(zkutil::makeRemoveRequest(fs::path(part_path) / "columns", -1)); } ops.emplace_back(zkutil::makeRemoveRequest(part_path, -1)); } void StorageReplicatedMergeTree::removePartFromZooKeeper(const String & part_name) { auto zookeeper = getZooKeeper(); String part_path = fs::path(replica_path) / "parts" / part_name; Coordination::Stat stat; /// Part doesn't exist, nothing to remove if (!zookeeper->exists(part_path, &stat)) return; Coordination::Requests ops; getRemovePartFromZooKeeperOps(part_name, ops, stat.numChildren > 0); zookeeper->multi(ops); } void StorageReplicatedMergeTree::removePartAndEnqueueFetch(const String & part_name, bool storage_init) { auto zookeeper = getZooKeeper(); /// We don't know exactly what happened to broken part /// and we are going to remove all covered log entries. /// It's quite dangerous, so clone covered parts to detached. auto broken_part_info = MergeTreePartInfo::fromPartName(part_name, format_version); auto partition_range = getVisibleDataPartsVectorInPartition(getContext(), broken_part_info.partition_id); for (const auto & part : partition_range) { if (!broken_part_info.contains(part->info)) continue; /// Broken part itself either already moved to detached or does not exist. assert(broken_part_info != part->info); part->makeCloneInDetached("covered-by-broken", getInMemoryMetadataPtr()); } /// It's possible that queue contains entries covered by part_name. /// For example, we had GET_PART all_1_42_5 and MUTATE_PART all_1_42_5_63, /// then all_1_42_5_63 was executed by fetching, but part was written to disk incorrectly. /// In this case we have to remove it as broken and create GET_PART all_1_42_5_63 to fetch it again, /// but GET_PART all_1_42_5 may be still in the queue. /// We should remove all covered entries before creating GET_PART entry, because: /// 1. In the situation described above, we do not know how to merge/mutate all_1_42_5_63 from all_1_42_5, /// so GET_PART all_1_42_5 (and all source parts) is useless. The only thing we can do is to fetch all_1_42_5_63. /// 2. If all_1_42_5_63 is lost, then replication may stuck waiting for all_1_42_5_63 to appear, /// because we may have some covered parts (more precisely, parts with the same min and max blocks) queue.removePartProducingOpsInRange(zookeeper, broken_part_info, /* covering_entry= */ {}); String part_path = fs::path(replica_path) / "parts" / part_name; while (true) { Coordination::Requests ops; time_t part_create_time = 0; Coordination::Stat stat; if (zookeeper->exists(part_path, &stat)) { /// Update version of /is_lost node to avoid race condition with cloneReplica(...). /// cloneReplica(...) expects that if some entry was executed, then its new_part_name is added to /parts, /// but we are going to remove it from /parts and add to queue again. Coordination::Stat is_lost_stat; String is_lost_value = zookeeper->get(replica_path + "/is_lost", &is_lost_stat); assert(is_lost_value == "0"); ops.emplace_back(zkutil::makeSetRequest(replica_path + "/is_lost", is_lost_value, is_lost_stat.version)); part_create_time = stat.ctime / 1000; getRemovePartFromZooKeeperOps(part_name, ops, stat.numChildren > 0); } LogEntryPtr log_entry = std::make_shared(); log_entry->type = LogEntry::GET_PART; log_entry->create_time = part_create_time; log_entry->source_replica = ""; log_entry->new_part_name = part_name; if (!storage_init) { /// We use merge predicate + version check here, because DROP RANGE update log version and we are trying to avoid race with it. We must be sure, that our part /// was not dropped, otherwise we will have fetch entry, but no virtual part for it (DROP RANGE will remove it). So bad sequence is the following: /// 1) Create DROP PART in log for broken_part (for example because it's not only broken, but also empty) /// 1.5) Pull to log and start execution (removePartProducingOpsInRange) /// 2) Execute this method (create GET PART for broken part) /// 3) Finish DROP PART execution (remove part from virtual parts) /// 4) Now we get GET PART for broken part which is absent in virtual parts == bug /// /// Because of version check this method will never create FETCH if drop part exists ReplicatedMergeTreeMergePredicate merge_pred = queue.getMergePredicate(zookeeper, PartitionIdsHint{broken_part_info.partition_id}); if (merge_pred.isGoingToBeDropped(broken_part_info)) { LOG_INFO(log, "Broken part {} is covered by drop range, don't need to fetch it", part_name); return; } /// Check that our version of log (and queue) is the most fresh. Otherwise don't create new entry fetch entry. ops.emplace_back(zkutil::makeCheckRequest(fs::path(zookeeper_path) / "log", merge_pred.getVersion())); } ops.emplace_back(zkutil::makeCreateRequest( fs::path(replica_path) / "queue/queue-", log_entry->toString(), zkutil::CreateMode::PersistentSequential)); Coordination::Responses results; auto rc = zookeeper->tryMulti(ops, results); if (rc == Coordination::Error::ZBADVERSION) { LOG_TRACE(log, "Log updated, cannot create fetch entry for part {}, will retry.", part_name); continue; } zkutil::KeeperMultiException::check(rc, ops, results); String path_created = dynamic_cast(*results.back()).path_created; log_entry->znode_name = path_created.substr(path_created.find_last_of('/') + 1); queue.insert(zookeeper, log_entry); break; } } void StorageReplicatedMergeTree::startBeingLeader() { auto zookeeper = getZooKeeper(); if (!getSettings()->replicated_can_become_leader) { LOG_INFO(log, "Will not enter leader election because replicated_can_become_leader=0"); return; } zkutil::checkNoOldLeaders(log, *zookeeper, fs::path(zookeeper_path) / "leader_election"); LOG_INFO(log, "Became leader"); is_leader = true; } void StorageReplicatedMergeTree::stopBeingLeader() { if (!is_leader) return; LOG_INFO(log, "Stopped being leader"); is_leader = false; } ConnectionTimeouts StorageReplicatedMergeTree::getFetchPartHTTPTimeouts(ContextPtr local_context) { auto timeouts = getHTTPTimeouts(local_context); auto settings = getSettings(); if (settings->replicated_fetches_http_connection_timeout.changed) timeouts.connection_timeout = settings->replicated_fetches_http_connection_timeout; if (settings->replicated_fetches_http_send_timeout.changed) timeouts.send_timeout = settings->replicated_fetches_http_send_timeout; if (settings->replicated_fetches_http_receive_timeout.changed) timeouts.receive_timeout = settings->replicated_fetches_http_receive_timeout; return timeouts; } bool StorageReplicatedMergeTree::checkReplicaHavePart(const String & replica, const String & part_name) { auto zookeeper = getZooKeeper(); return zookeeper->exists(fs::path(zookeeper_path) / "replicas" / replica / "parts" / part_name); } String StorageReplicatedMergeTree::findReplicaHavingPart(const String & part_name, bool active) { auto zookeeper = getZooKeeper(); Strings replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); /// Select replicas in uniformly random order. std::shuffle(replicas.begin(), replicas.end(), thread_local_rng); LOG_TRACE(log, "Candidate replicas: {}", replicas.size()); for (const String & replica : replicas) { /// We aren't interested in ourself. if (replica == replica_name) continue; LOG_TRACE(log, "Candidate replica: {}", replica); if (checkReplicaHavePart(replica, part_name) && (!active || zookeeper->exists(fs::path(zookeeper_path) / "replicas" / replica / "is_active"))) return replica; /// Obviously, replica could become inactive or even vanish after return from this method. } return {}; } String StorageReplicatedMergeTree::findReplicaHavingCoveringPart(LogEntry & entry, bool active) { auto zookeeper = getZooKeeper(); Strings replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); /// Select replicas in uniformly random order. std::shuffle(replicas.begin(), replicas.end(), thread_local_rng); for (const String & replica : replicas) { if (replica == replica_name) continue; if (active && !zookeeper->exists(fs::path(zookeeper_path) / "replicas" / replica / "is_active")) continue; String largest_part_found; Strings parts = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas" / replica / "parts"); for (const String & part_on_replica : parts) { if (part_on_replica == entry.new_part_name || MergeTreePartInfo::contains(part_on_replica, entry.new_part_name, format_version)) { if (largest_part_found.empty() || MergeTreePartInfo::contains(part_on_replica, largest_part_found, format_version)) { largest_part_found = part_on_replica; } } } if (!largest_part_found.empty()) { bool the_same_part = largest_part_found == entry.new_part_name; /// Make a check in case if selected part differs from source part if (!the_same_part) { String reject_reason; if (!queue.addFuturePartIfNotCoveredByThem(largest_part_found, entry, reject_reason)) { LOG_INFO(log, "Will not fetch part {} covering {}. {}", largest_part_found, entry.new_part_name, reject_reason); return {}; } } return replica; } } return {}; } String StorageReplicatedMergeTree::findReplicaHavingCoveringPart( const String & part_name, bool active, String & found_part_name) { auto zookeeper = getZooKeeper(); Strings replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); /// Select replicas in uniformly random order. std::shuffle(replicas.begin(), replicas.end(), thread_local_rng); String largest_part_found; String largest_replica_found; for (const String & replica : replicas) { if (replica == replica_name) continue; if (active && !zookeeper->exists(fs::path(zookeeper_path) / "replicas" / replica / "is_active")) continue; Strings parts = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas" / replica / "parts"); for (const String & part_on_replica : parts) { if (part_on_replica == part_name || MergeTreePartInfo::contains(part_on_replica, part_name, format_version)) { if (largest_part_found.empty() || MergeTreePartInfo::contains(part_on_replica, largest_part_found, format_version)) { largest_part_found = part_on_replica; largest_replica_found = replica; } } } } found_part_name = largest_part_found; return largest_replica_found; } /** If a quorum is tracked for a part, update information about it in ZK. */ void StorageReplicatedMergeTree::updateQuorum(const String & part_name, bool is_parallel) { if (is_parallel && format_version < MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING) throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Parallel quorum inserts are not compatible with the deprecated syntax of *MergeTree engines"); auto zookeeper = getZooKeeper(); /// Information on which replicas a part has been added, if the quorum has not yet been reached. String quorum_status_path = fs::path(zookeeper_path) / "quorum" / "status"; if (is_parallel) quorum_status_path = fs::path(zookeeper_path) / "quorum" / "parallel" / part_name; /// The name of the previous part for which the quorum was reached. const String quorum_last_part_path = fs::path(zookeeper_path) / "quorum" / "last_part"; String value; Coordination::Stat stat; /// If there is no node, then all quorum INSERTs have already reached the quorum, and nothing is needed. while (zookeeper->tryGet(quorum_status_path, value, &stat)) { ReplicatedMergeTreeQuorumEntry quorum_entry(value); if (quorum_entry.part_name != part_name) { LOG_TRACE(log, "Quorum {}, already achieved for part {} current part {}", quorum_status_path, part_name, quorum_entry.part_name); /// The quorum has already been achieved. Moreover, another INSERT with a quorum has already started. break; } quorum_entry.replicas.insert(replica_name); if (quorum_entry.replicas.size() >= quorum_entry.required_number_of_replicas) { /// The quorum is reached. Delete the node, and update information about the last part that was successfully written with quorum. LOG_TRACE(log, "Got {} (of {} required) replicas confirmed quorum {}, going to remove node", quorum_entry.replicas.size(), quorum_entry.required_number_of_replicas, quorum_status_path); Coordination::Requests ops; Coordination::Responses responses; if (!is_parallel) { Coordination::Stat added_parts_stat; String old_added_parts = zookeeper->get(quorum_last_part_path, &added_parts_stat); ReplicatedMergeTreeQuorumAddedParts parts_with_quorum(format_version); if (!old_added_parts.empty()) parts_with_quorum.fromString(old_added_parts); auto part_info = MergeTreePartInfo::fromPartName(part_name, format_version); /// We store one last part which reached quorum for each partition. parts_with_quorum.added_parts[part_info.partition_id] = part_name; String new_added_parts = parts_with_quorum.toString(); ops.emplace_back(zkutil::makeRemoveRequest(quorum_status_path, stat.version)); ops.emplace_back(zkutil::makeSetRequest(quorum_last_part_path, new_added_parts, added_parts_stat.version)); } else ops.emplace_back(zkutil::makeRemoveRequest(quorum_status_path, stat.version)); auto code = zookeeper->tryMulti(ops, responses); if (code == Coordination::Error::ZOK) { break; } else if (code == Coordination::Error::ZNONODE) { /// The quorum has already been achieved. break; } else if (code == Coordination::Error::ZBADVERSION) { /// Node was updated meanwhile. We must re-read it and repeat all the actions. continue; } else throw Coordination::Exception(code, quorum_status_path); } else { LOG_TRACE(log, "Quorum {} still not satisfied (have only {} of {} replicas), updating node", quorum_status_path, quorum_entry.replicas.size(), quorum_entry.required_number_of_replicas); /// We update the node, registering there one more replica. auto code = zookeeper->trySet(quorum_status_path, quorum_entry.toString(), stat.version); if (code == Coordination::Error::ZOK) { break; } else if (code == Coordination::Error::ZNONODE) { /// The quorum has already been achieved. break; } else if (code == Coordination::Error::ZBADVERSION) { /// Node was updated meanwhile. We must re-read it and repeat all the actions. continue; } else throw Coordination::Exception(code, quorum_status_path); } } } void StorageReplicatedMergeTree::cleanLastPartNode(const String & partition_id) { auto zookeeper = getZooKeeper(); LOG_DEBUG(log, "Cleaning up last parent node for partition {}", partition_id); /// The name of the previous part for which the quorum was reached. const String quorum_last_part_path = fs::path(zookeeper_path) / "quorum" / "last_part"; /// Delete information from "last_part" node. while (true) { Coordination::Stat added_parts_stat; String old_added_parts = zookeeper->get(quorum_last_part_path, &added_parts_stat); ReplicatedMergeTreeQuorumAddedParts parts_with_quorum(format_version); if (!old_added_parts.empty()) parts_with_quorum.fromString(old_added_parts); /// Delete information about particular partition. if (!parts_with_quorum.added_parts.contains(partition_id)) { /// There is no information about interested part. break; } parts_with_quorum.added_parts.erase(partition_id); String new_added_parts = parts_with_quorum.toString(); auto code = zookeeper->trySet(quorum_last_part_path, new_added_parts, added_parts_stat.version); if (code == Coordination::Error::ZOK) { LOG_DEBUG(log, "Last parent node for partition {} is cleaned up", partition_id); break; } else if (code == Coordination::Error::ZNONODE) { /// Node is deleted. It is impossible, but it is Ok. break; } else if (code == Coordination::Error::ZBADVERSION) { /// Node was updated meanwhile. We must re-read it and repeat all the actions. continue; } else throw Coordination::Exception(code, quorum_last_part_path); } } bool StorageReplicatedMergeTree::partIsInsertingWithParallelQuorum(const MergeTreePartInfo & part_info) const { auto zookeeper = getZooKeeper(); return zookeeper->exists(fs::path(zookeeper_path) / "quorum" / "parallel" / part_info.getPartNameV1()); } bool StorageReplicatedMergeTree::partIsLastQuorumPart(const MergeTreePartInfo & part_info) const { auto zookeeper = getZooKeeper(); const String parts_with_quorum_path = fs::path(zookeeper_path) / "quorum" / "last_part"; String parts_with_quorum_str = zookeeper->get(parts_with_quorum_path); if (parts_with_quorum_str.empty()) return false; ReplicatedMergeTreeQuorumAddedParts parts_with_quorum(format_version); parts_with_quorum.fromString(parts_with_quorum_str); auto partition_it = parts_with_quorum.added_parts.find(part_info.partition_id); if (partition_it == parts_with_quorum.added_parts.end()) return false; return partition_it->second == part_info.getPartNameAndCheckFormat(format_version); } bool StorageReplicatedMergeTree::fetchPart( const String & part_name, const StorageMetadataPtr & metadata_snapshot, const String & source_replica_path, bool to_detached, size_t quorum, zkutil::ZooKeeper::Ptr zookeeper_, bool try_fetch_shared) { auto zookeeper = zookeeper_ ? zookeeper_ : getZooKeeper(); const auto part_info = MergeTreePartInfo::fromPartName(part_name, format_version); if (!to_detached) { if (auto part = getPartIfExists(part_info, {MergeTreeDataPartState::Outdated, MergeTreeDataPartState::Deleting})) { LOG_DEBUG(log, "Part {} should be deleted after previous attempt before fetch", part->name); /// Force immediate parts cleanup to delete the part that was left from the previous fetch attempt. cleanup_thread.wakeup(); return false; } } { std::lock_guard lock(currently_fetching_parts_mutex); if (!currently_fetching_parts.insert(part_name).second) { LOG_DEBUG(log, "Part {} is already fetching right now", part_name); return false; } } SCOPE_EXIT_MEMORY ({ std::lock_guard lock(currently_fetching_parts_mutex); currently_fetching_parts.erase(part_name); }); LOG_DEBUG(log, "Fetching part {} from {}", part_name, source_replica_path); auto settings_ptr = getSettings(); TableLockHolder table_lock_holder; if (!to_detached) table_lock_holder = lockForShare(RWLockImpl::NO_QUERY, settings_ptr->lock_acquire_timeout_for_background_operations); /// Logging Stopwatch stopwatch; MutableDataPartPtr part; DataPartsVector replaced_parts; ProfileEventsScope profile_events_scope; auto write_part_log = [&] (const ExecutionStatus & execution_status) { writePartLog( PartLogElement::DOWNLOAD_PART, execution_status, stopwatch.elapsed(), part_name, part, replaced_parts, nullptr, profile_events_scope.getSnapshot()); }; DataPartPtr part_to_clone; { /// If the desired part is a result of a part mutation, try to find the source part and compare /// its checksums to the checksums of the desired part. If they match, we can just clone the local part. /// If we have the source part, its part_info will contain covered_part_info. auto covered_part_info = part_info; covered_part_info.mutation = 0; auto source_part = getActiveContainingPart(covered_part_info); /// Fetch for zero-copy replication is cheap and straightforward, so we don't use local clone here if (source_part && (!settings_ptr->allow_remote_fs_zero_copy_replication || !source_part->getDataPartStorage().supportZeroCopyReplication())) { auto source_part_header = ReplicatedMergeTreePartHeader::fromColumnsAndChecksums( source_part->getColumns(), source_part->checksums); String part_path = fs::path(source_replica_path) / "parts" / part_name; String part_znode = zookeeper->get(part_path); std::optional desired_part_header; if (!part_znode.empty()) { desired_part_header = ReplicatedMergeTreePartHeader::fromString(part_znode); } else { String columns_str; String checksums_str; if (zookeeper->tryGet(fs::path(part_path) / "columns", columns_str) && zookeeper->tryGet(fs::path(part_path) / "checksums", checksums_str)) { desired_part_header = ReplicatedMergeTreePartHeader::fromColumnsAndChecksumsZNodes(columns_str, checksums_str); } else { LOG_INFO(log, "Not checking checksums of part {} with replica {} because part was removed from ZooKeeper", part_name, source_replica_path); } } /// Checking both checksums and columns hash. For example we can have empty part /// with same checksums but different columns. And we attaching it exception will /// be thrown. if (desired_part_header && source_part_header.getColumnsHash() == desired_part_header->getColumnsHash() && source_part_header.getChecksums() == desired_part_header->getChecksums()) { LOG_TRACE(log, "Found local part {} with the same checksums and columns hash as {}", source_part->name, part_name); part_to_clone = source_part; } } } ReplicatedMergeTreeAddress address; ConnectionTimeouts timeouts; String interserver_scheme; InterserverCredentialsPtr credentials; std::optional tagger_ptr; std::function get_part; MergeTreeData::HardlinkedFiles hardlinked_files; scope_guard part_to_clone_lock; if (part_to_clone) { get_part = [&, part_to_clone]() { auto [cloned_part, lock] = cloneAndLoadDataPartOnSameDisk(part_to_clone, "tmp_clone_", part_info, metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, false, {}); part_to_clone_lock = std::move(lock); return cloned_part; }; } else { address.fromString(zookeeper->get(fs::path(source_replica_path) / "host")); timeouts = getFetchPartHTTPTimeouts(getContext()); credentials = getContext()->getInterserverCredentials(); interserver_scheme = getContext()->getInterserverScheme(); get_part = [&, address, timeouts, credentials, interserver_scheme]() { if (interserver_scheme != address.scheme) throw Exception(ErrorCodes::INTERSERVER_SCHEME_DOESNT_MATCH, "Interserver schemes are different: " "'{}' != '{}', can't fetch part from {}", interserver_scheme, address.scheme, address.host); return fetcher.fetchSelectedPart( metadata_snapshot, getContext(), part_name, source_replica_path, address.host, address.replication_port, timeouts, credentials->getUser(), credentials->getPassword(), interserver_scheme, replicated_fetches_throttler, to_detached, "", &tagger_ptr, try_fetch_shared); }; } try { part = get_part(); if (!to_detached) { Transaction transaction(*this, NO_TRANSACTION_RAW); renameTempPartAndReplace(part, transaction); replaced_parts = checkPartChecksumsAndCommit(transaction, part, hardlinked_files); /** If a quorum is tracked for this part, you must update it. * If you do not have time, in case of losing the session, when you restart the server - see the `ReplicatedMergeTreeRestartingThread::updateQuorumIfWeHavePart` method. */ if (quorum) { /// Check if this quorum insert is parallel or not if (zookeeper->exists(fs::path(zookeeper_path) / "quorum" / "parallel" / part_name)) updateQuorum(part_name, true); else if (zookeeper->exists(fs::path(zookeeper_path) / "quorum" / "status")) updateQuorum(part_name, false); } /// merged parts that are still inserted with quorum. if it only contains one block, it hasn't been merged before if (part_info.level != 0 || part_info.mutation != 0) { Strings quorum_parts = zookeeper->getChildren(fs::path(zookeeper_path) / "quorum" / "parallel"); for (const String & quorum_part : quorum_parts) { auto quorum_part_info = MergeTreePartInfo::fromPartName(quorum_part, format_version); if (part_info.contains(quorum_part_info)) updateQuorum(quorum_part, true); } } merge_selecting_task->schedule(); for (const auto & replaced_part : replaced_parts) { LOG_DEBUG(log, "Part {} is rendered obsolete by fetching part {}", replaced_part->name, part_name); ProfileEvents::increment(ProfileEvents::ObsoleteReplicatedParts); } write_part_log({}); } else { // The fetched part is valuable and should not be cleaned like a temp part. part->is_temp = false; part->renameTo(fs::path("detached") / part_name, true); } } catch (const Exception & e) { /// The same part is being written right now (but probably it's not committed yet). /// We will check the need for fetch later. if (e.code() == ErrorCodes::DIRECTORY_ALREADY_EXISTS) { LOG_TRACE(log, "Not fetching part: {}", e.message()); return false; } throw; } catch (...) { if (!to_detached) write_part_log(ExecutionStatus::fromCurrentException("", true)); throw; } ProfileEvents::increment(ProfileEvents::ReplicatedPartFetches); if (part_to_clone) LOG_DEBUG(log, "Cloned part {} from {}{}", part_name, part_to_clone->name, to_detached ? " (to 'detached' directory)" : ""); else LOG_DEBUG(log, "Fetched part {} from {}{}", part_name, source_replica_path, to_detached ? " (to 'detached' directory)" : ""); return true; } MutableDataPartStoragePtr StorageReplicatedMergeTree::fetchExistsPart( const String & part_name, const StorageMetadataPtr & metadata_snapshot, const String & source_replica_path, DiskPtr replaced_disk, String replaced_part_path) { auto zookeeper = getZooKeeper(); const auto part_info = MergeTreePartInfo::fromPartName(part_name, format_version); if (auto part = getPartIfExists(part_info, {MergeTreeDataPartState::Outdated, MergeTreeDataPartState::Deleting})) { LOG_DEBUG(log, "Part {} should be deleted after previous attempt before fetch", part->name); /// Force immediate parts cleanup to delete the part that was left from the previous fetch attempt. cleanup_thread.wakeup(); return nullptr; } { std::lock_guard lock(currently_fetching_parts_mutex); if (!currently_fetching_parts.insert(part_name).second) { LOG_DEBUG(log, "Part {} is already fetching right now", part_name); return nullptr; } } SCOPE_EXIT_MEMORY ({ std::lock_guard lock(currently_fetching_parts_mutex); currently_fetching_parts.erase(part_name); }); LOG_DEBUG(log, "Fetching already known part {} from {}", part_name, source_replica_path); TableLockHolder table_lock_holder = lockForShare(RWLockImpl::NO_QUERY, getSettings()->lock_acquire_timeout_for_background_operations); /// Logging Stopwatch stopwatch; MutableDataPartPtr part; DataPartsVector replaced_parts; ProfileEventsScope profile_events_scope; auto write_part_log = [&] (const ExecutionStatus & execution_status) { writePartLog( PartLogElement::DOWNLOAD_PART, execution_status, stopwatch.elapsed(), part_name, part, replaced_parts, nullptr, profile_events_scope.getSnapshot()); }; std::function get_part; ReplicatedMergeTreeAddress address(zookeeper->get(fs::path(source_replica_path) / "host")); auto timeouts = getHTTPTimeouts(getContext()); auto credentials = getContext()->getInterserverCredentials(); String interserver_scheme = getContext()->getInterserverScheme(); get_part = [&, address, timeouts, interserver_scheme, credentials]() { if (interserver_scheme != address.scheme) throw Exception(ErrorCodes::INTERSERVER_SCHEME_DOESNT_MATCH, "Interserver schemes are different: " "'{}' != '{}', can't fetch part from {}", interserver_scheme, address.scheme, address.host); return fetcher.fetchSelectedPart( metadata_snapshot, getContext(), part_name, source_replica_path, address.host, address.replication_port, timeouts, credentials->getUser(), credentials->getPassword(), interserver_scheme, replicated_fetches_throttler, false, "", nullptr, true, replaced_disk); }; try { part = get_part(); if (part->getDataPartStorage().getDiskName() != replaced_disk->getName()) throw Exception(ErrorCodes::LOGICAL_ERROR, "Part {} fetched on wrong disk {}", part->name, part->getDataPartStorage().getDiskName()); auto replaced_path = fs::path(replaced_part_path); part->getDataPartStorage().rename(replaced_path.parent_path(), replaced_path.filename(), nullptr, true, false); } catch (const Exception & e) { /// The same part is being written right now (but probably it's not committed yet). /// We will check the need for fetch later. if (e.code() == ErrorCodes::DIRECTORY_ALREADY_EXISTS) { LOG_TRACE(log, "Not fetching part: {}", e.message()); return nullptr; } throw; } catch (...) { write_part_log(ExecutionStatus::fromCurrentException("", true)); throw; } ProfileEvents::increment(ProfileEvents::ReplicatedPartFetches); LOG_DEBUG(log, "Fetched part {} from {}", part_name, source_replica_path); return part->getDataPartStoragePtr(); } void StorageReplicatedMergeTree::startup() { startOutdatedDataPartsLoadingTask(); if (attach_thread) { attach_thread->start(); attach_thread->waitFirstTry(); return; } startupImpl(/* from_attach_thread */ false); } void StorageReplicatedMergeTree::startupImpl(bool from_attach_thread) { /// Do not start replication if ZooKeeper is not configured or there is no metadata in zookeeper if (!has_metadata_in_zookeeper.has_value() || !*has_metadata_in_zookeeper) return; try { auto zookeeper = getZooKeeper(); InterserverIOEndpointPtr data_parts_exchange_ptr = std::make_shared(*this); [[maybe_unused]] auto prev_ptr = std::atomic_exchange(&data_parts_exchange_endpoint, data_parts_exchange_ptr); assert(prev_ptr == nullptr); getContext()->getInterserverIOHandler().addEndpoint(data_parts_exchange_ptr->getId(replica_path), data_parts_exchange_ptr); startBeingLeader(); /// In this thread replica will be activated. restarting_thread.start(); /// And this is just a callback session_expired_callback_handler = EventNotifier::instance().subscribe(Coordination::Error::ZSESSIONEXPIRED, [this]() { LOG_TEST(log, "Received event for expired session. Waking up restarting thread"); restarting_thread.start(); }); /// Wait while restarting_thread finishing initialization. /// NOTE It does not mean that replication is actually started after receiving this event. /// It only means that an attempt to startup replication was made. /// Table may be still in readonly mode if this attempt failed for any reason. startup_event.wait(); startBackgroundMovesIfNeeded(); part_moves_between_shards_orchestrator.start(); } catch (...) { /// Exception safety: failed "startup" does not require a call to "shutdown" from the caller. /// And it should be able to safely destroy table after exception in "startup" method. /// It means that failed "startup" must not create any background tasks that we will have to wait. try { /// it's important to avoid full shutdown here, because it even tries to shutdown attach thread which was /// designed exactly for this: try to start table if no zookeeper connection available. if (from_attach_thread) { restarting_thread.shutdown(/* part_of_full_shutdown */false); } else { shutdown(); } } catch (...) { std::terminate(); } /// Note: after failed "startup", the table will be in a state that only allows to destroy the object. throw; } } void StorageReplicatedMergeTree::flush() { if (flush_called.exchange(true)) return; flushAllInMemoryPartsIfNeeded(); } void StorageReplicatedMergeTree::partialShutdown() { ProfileEvents::increment(ProfileEvents::ReplicaPartialShutdown); partial_shutdown_called = true; partial_shutdown_event.set(); replica_is_active_node = nullptr; LOG_TRACE(log, "Waiting for threads to finish"); merge_selecting_task->deactivate(); queue_updating_task->deactivate(); mutations_updating_task->deactivate(); mutations_finalizing_task->deactivate(); cleanup_thread.stop(); async_block_ids_cache.stop(); part_check_thread.stop(); /// Stop queue processing { auto fetch_lock = fetcher.blocker.cancel(); auto merge_lock = merger_mutator.merges_blocker.cancel(); auto move_lock = parts_mover.moves_blocker.cancel(); background_operations_assignee.finish(); } LOG_TRACE(log, "Threads finished"); } void StorageReplicatedMergeTree::shutdown() { if (shutdown_called.exchange(true)) return; session_expired_callback_handler.reset(); stopOutdatedDataPartsLoadingTask(); /// Cancel fetches, merges and mutations to force the queue_task to finish ASAP. fetcher.blocker.cancelForever(); merger_mutator.merges_blocker.cancelForever(); parts_mover.moves_blocker.cancelForever(); mutations_finalizing_task->deactivate(); stopBeingLeader(); if (attach_thread) attach_thread->shutdown(); restarting_thread.shutdown(/* part_of_full_shutdown */true); background_operations_assignee.finish(); part_moves_between_shards_orchestrator.shutdown(); { auto lock = queue.lockQueue(); /// Cancel logs pulling after background task were cancelled. It's still /// required because we can trigger pullLogsToQueue during manual OPTIMIZE, /// MUTATE, etc. query. queue.pull_log_blocker.cancelForever(); } background_moves_assignee.finish(); auto data_parts_exchange_ptr = std::atomic_exchange(&data_parts_exchange_endpoint, InterserverIOEndpointPtr{}); if (data_parts_exchange_ptr) { getContext()->getInterserverIOHandler().removeEndpointIfExists(data_parts_exchange_ptr->getId(replica_path)); /// Ask all parts exchange handlers to finish asap. New ones will fail to start data_parts_exchange_ptr->blocker.cancelForever(); /// Wait for all of them std::lock_guard lock(data_parts_exchange_ptr->rwlock); } } StorageReplicatedMergeTree::~StorageReplicatedMergeTree() { try { shutdown(); } catch (...) { tryLogCurrentException(__PRETTY_FUNCTION__); } } ReplicatedMergeTreeQuorumAddedParts::PartitionIdToMaxBlock StorageReplicatedMergeTree::getMaxAddedBlocks() const { ReplicatedMergeTreeQuorumAddedParts::PartitionIdToMaxBlock max_added_blocks; for (const auto & data_part : getDataPartsForInternalUsage()) { max_added_blocks[data_part->info.partition_id] = std::max(max_added_blocks[data_part->info.partition_id], data_part->info.max_block); } auto zookeeper = getZooKeeper(); const String quorum_status_path = fs::path(zookeeper_path) / "quorum" / "status"; String value; Coordination::Stat stat; if (zookeeper->tryGet(quorum_status_path, value, &stat)) { ReplicatedMergeTreeQuorumEntry quorum_entry; quorum_entry.fromString(value); auto part_info = MergeTreePartInfo::fromPartName(quorum_entry.part_name, format_version); max_added_blocks[part_info.partition_id] = part_info.max_block - 1; } String added_parts_str; if (zookeeper->tryGet(fs::path(zookeeper_path) / "quorum" / "last_part", added_parts_str)) { if (!added_parts_str.empty()) { ReplicatedMergeTreeQuorumAddedParts part_with_quorum(format_version); part_with_quorum.fromString(added_parts_str); auto added_parts = part_with_quorum.added_parts; for (const auto & added_part : added_parts) { if (!getActiveContainingPart(added_part.second)) throw Exception(ErrorCodes::REPLICA_IS_NOT_IN_QUORUM, "Replica doesn't have part '{}' which was successfully written to quorum of other replicas. " "Send query to another replica or disable 'select_sequential_consistency' setting", added_part.second); } for (const auto & max_block : part_with_quorum.getMaxInsertedBlocks()) max_added_blocks[max_block.first] = max_block.second; } } return max_added_blocks; } void StorageReplicatedMergeTree::read( QueryPlan & query_plan, const Names & column_names, const StorageSnapshotPtr & storage_snapshot, SelectQueryInfo & query_info, ContextPtr local_context, QueryProcessingStage::Enum processed_stage, const size_t max_block_size, const size_t num_streams) { SCOPE_EXIT({ /// Now, copy of parts that is required for the query, stored in the processors, /// while snapshot_data.parts includes all parts, even one that had been filtered out with partition pruning, /// reset them to avoid holding them. auto & snapshot_data = assert_cast(*storage_snapshot->data); snapshot_data.parts = {}; }); /** The `select_sequential_consistency` setting has two meanings: * 1. To throw an exception if on a replica there are not all parts which have been written down on quorum of remaining replicas. * 2. Do not read parts that have not yet been written to the quorum of the replicas. * For this you have to synchronously go to ZooKeeper. */ if (local_context->getSettingsRef().select_sequential_consistency) { auto max_added_blocks = std::make_shared(getMaxAddedBlocks()); if (auto plan = reader.read( column_names, storage_snapshot, query_info, local_context, max_block_size, num_streams, processed_stage, std::move(max_added_blocks), /*enable_parallel_reading*/false)) query_plan = std::move(*plan); return; } if (local_context->canUseParallelReplicasOnInitiator()) { auto table_id = getStorageID(); const auto & modified_query_ast = ClusterProxy::rewriteSelectQuery( local_context, query_info.query, table_id.database_name, table_id.table_name, /*remote_table_function_ptr*/nullptr); auto cluster = local_context->getCluster(local_context->getSettingsRef().cluster_for_parallel_replicas); Block header = InterpreterSelectQuery(modified_query_ast, local_context, SelectQueryOptions(processed_stage).analyze()).getSampleBlock(); ClusterProxy::SelectStreamFactory select_stream_factory = ClusterProxy::SelectStreamFactory( header, {}, storage_snapshot, processed_stage); ClusterProxy::executeQueryWithParallelReplicas( query_plan, getStorageID(), /*remove_table_function_ptr*/ nullptr, select_stream_factory, modified_query_ast, local_context, query_info, cluster); } else { if (auto plan = reader.read( column_names, storage_snapshot, query_info, local_context, max_block_size, num_streams, processed_stage, nullptr, /*enable_parallel_reading*/local_context->canUseParallelReplicasOnFollower())) query_plan = std::move(*plan); } } template void StorageReplicatedMergeTree::foreachActiveParts(Func && func, bool select_sequential_consistency) const { std::optional max_added_blocks = {}; /** * Synchronously go to ZooKeeper when select_sequential_consistency enabled */ if (select_sequential_consistency) max_added_blocks = getMaxAddedBlocks(); auto lock = lockParts(); /// TODO Transactions: should we count visible parts only? for (const auto & part : getDataPartsStateRange(DataPartState::Active)) { if (part->isEmpty()) continue; if (max_added_blocks) { auto blocks_iterator = max_added_blocks->find(part->info.partition_id); if (blocks_iterator == max_added_blocks->end() || part->info.max_block > blocks_iterator->second) continue; } func(part); } } std::optional StorageReplicatedMergeTree::totalRows(const Settings & settings) const { UInt64 res = 0; foreachActiveParts([&res](auto & part) { res += part->rows_count; }, settings.select_sequential_consistency); return res; } std::optional StorageReplicatedMergeTree::totalRowsByPartitionPredicate(const SelectQueryInfo & query_info, ContextPtr local_context) const { DataPartsVector parts; foreachActiveParts([&](auto & part) { parts.push_back(part); }, local_context->getSettingsRef().select_sequential_consistency); return totalRowsByPartitionPredicateImpl(query_info, local_context, parts); } std::optional StorageReplicatedMergeTree::totalBytes(const Settings & settings) const { UInt64 res = 0; foreachActiveParts([&res](auto & part) { res += part->getBytesOnDisk(); }, settings.select_sequential_consistency); return res; } void StorageReplicatedMergeTree::assertNotReadonly() const { if (is_readonly) throw Exception(ErrorCodes::TABLE_IS_READ_ONLY, "Table is in readonly mode (replica path: {})", replica_path); } SinkToStoragePtr StorageReplicatedMergeTree::write(const ASTPtr & /*query*/, const StorageMetadataPtr & metadata_snapshot, ContextPtr local_context) { if (!initialization_done) throw Exception(ErrorCodes::NOT_INITIALIZED, "Table is not initialized yet"); /// If table is read-only because it doesn't have metadata in zk yet, then it's not possible to insert into it /// Without this check, we'll write data parts on disk, and afterwards will remove them since we'll fail to commit them into zk /// In case of remote storage like s3, it'll generate unnecessary PUT requests if (is_readonly && (!has_metadata_in_zookeeper.has_value() || false == has_metadata_in_zookeeper.value())) throw Exception( ErrorCodes::TABLE_IS_READ_ONLY, "Table is in readonly mode since table metadata was not found in zookeeper: replica_path={}", replica_path); const auto storage_settings_ptr = getSettings(); const Settings & query_settings = local_context->getSettingsRef(); bool deduplicate = storage_settings_ptr->replicated_deduplication_window != 0 && query_settings.insert_deduplicate; bool async_deduplicate = query_settings.async_insert && query_settings.async_insert_deduplicate && storage_settings_ptr->replicated_deduplication_window_for_async_inserts != 0 && query_settings.insert_deduplicate; if (async_deduplicate) return std::make_shared( *this, metadata_snapshot, query_settings.insert_quorum.valueOr(0), query_settings.insert_quorum_timeout.totalMilliseconds(), query_settings.max_partitions_per_insert_block, query_settings.insert_quorum_parallel, deduplicate, query_settings.insert_quorum.is_auto, local_context); // TODO: should we also somehow pass list of columns to deduplicate on to the ReplicatedMergeTreeSink? return std::make_shared( *this, metadata_snapshot, query_settings.insert_quorum.valueOr(0), query_settings.insert_quorum_timeout.totalMilliseconds(), query_settings.max_partitions_per_insert_block, query_settings.insert_quorum_parallel, deduplicate, query_settings.insert_quorum.is_auto, local_context); } std::optional StorageReplicatedMergeTree::distributedWriteFromClusterStorage(const std::shared_ptr & src_storage_cluster, const ASTInsertQuery & query, ContextPtr local_context) { const auto & settings = local_context->getSettingsRef(); auto extension = src_storage_cluster->getTaskIteratorExtension(nullptr, local_context); /// Here we won't check that the cluster formed from table replicas is a subset of a cluster specified in s3Cluster/hdfsCluster table function auto src_cluster = src_storage_cluster->getCluster(local_context); /// Actually the query doesn't change, we just serialize it to string String query_str; { WriteBufferFromOwnString buf; IAST::FormatSettings ast_format_settings(buf, /*one_line*/ true); ast_format_settings.always_quote_identifiers = true; query.IAST::format(ast_format_settings); query_str = buf.str(); } QueryPipeline pipeline; ContextMutablePtr query_context = Context::createCopy(local_context); ++query_context->getClientInfo().distributed_depth; for (const auto & replicas : src_cluster->getShardsAddresses()) { /// There will be only one replica, because we consider each replica as a shard for (const auto & node : replicas) { auto connection = std::make_shared( node.host_name, node.port, query_context->getGlobalContext()->getCurrentDatabase(), node.user, node.password, node.quota_key, node.cluster, node.cluster_secret, "ParallelInsertSelectInititiator", node.compression, node.secure ); auto remote_query_executor = std::make_shared( connection, query_str, Block{}, query_context, /*throttler=*/nullptr, Scalars{}, Tables{}, QueryProcessingStage::Complete, extension); QueryPipeline remote_pipeline(std::make_shared(remote_query_executor, false, settings.async_socket_for_remote)); remote_pipeline.complete(std::make_shared(remote_query_executor->getHeader())); pipeline.addCompletedPipeline(std::move(remote_pipeline)); } } return pipeline; } std::optional StorageReplicatedMergeTree::distributedWrite(const ASTInsertQuery & query, ContextPtr local_context) { /// Do not enable parallel distributed INSERT SELECT in case when query probably comes from another server if (local_context->getClientInfo().query_kind != ClientInfo::QueryKind::INITIAL_QUERY) return {}; const Settings & settings = local_context->getSettingsRef(); if (settings.max_distributed_depth && local_context->getClientInfo().distributed_depth >= settings.max_distributed_depth) throw Exception(ErrorCodes::TOO_LARGE_DISTRIBUTED_DEPTH, "Maximum distributed depth exceeded"); auto & select = query.select->as(); StoragePtr src_storage; if (select.list_of_selects->children.size() == 1) { if (auto * select_query = select.list_of_selects->children.at(0)->as()) { JoinedTables joined_tables(Context::createCopy(local_context), *select_query); if (joined_tables.tablesCount() == 1) { src_storage = joined_tables.getLeftTableStorage(); } } } if (!src_storage) return {}; if (auto src_distributed = std::dynamic_pointer_cast(src_storage)) { return distributedWriteFromClusterStorage(src_distributed, query, local_context); } else if (local_context->getClientInfo().distributed_depth == 0) { throw Exception(ErrorCodes::BAD_ARGUMENTS, "Parallel distributed INSERT SELECT is not possible. Reason: distributed " "reading into Replicated table is supported only from *Cluster table functions, but got {} storage", src_storage->getName()); } return {}; } bool StorageReplicatedMergeTree::optimize( const ASTPtr &, const StorageMetadataPtr &, const ASTPtr & partition, bool final, bool deduplicate, const Names & deduplicate_by_columns, bool cleanup, ContextPtr query_context) { /// NOTE: exclusive lock cannot be used here, since this may lead to deadlock (see comments below), /// but it should be safe to use non-exclusive to avoid dropping parts that may be required for processing queue. auto table_lock = lockForShare(query_context->getCurrentQueryId(), query_context->getSettingsRef().lock_acquire_timeout); assertNotReadonly(); if (!is_leader) throw Exception(ErrorCodes::NOT_A_LEADER, "OPTIMIZE cannot be done on this replica because it is not a leader"); if (cleanup) LOG_DEBUG(log, "Cleanup the ReplicatedMergeTree."); auto handle_noop = [&](FormatStringHelper fmt_string, Args && ...args) { PreformattedMessage message = fmt_string.format(std::forward(args)...); LOG_DEBUG(log, message); if (query_context->getSettingsRef().optimize_throw_if_noop) throw Exception(std::move(message), ErrorCodes::CANNOT_ASSIGN_OPTIMIZE); return false; }; auto zookeeper = getZooKeeperAndAssertNotReadonly(); const auto storage_settings_ptr = getSettings(); auto metadata_snapshot = getInMemoryMetadataPtr(); std::vector merge_entries; auto try_assign_merge = [&](const String & partition_id) -> bool { constexpr size_t max_retries = 10; size_t try_no = 0; for (; try_no < max_retries; ++try_no) { /// We must select parts for merge under merge_selecting_mutex because other threads /// (merge_selecting_thread or OPTIMIZE queries) could assign new merges. std::lock_guard merge_selecting_lock(merge_selecting_mutex); PartitionIdsHint partition_ids_hint; if (partition_id.empty()) { partition_ids_hint = getAllPartitionIds(); } else { auto parts_lock = lockParts(); if (!getAnyPartInPartition(partition_id, parts_lock)) handle_noop("Cannot select parts for optimization: there are no parts in partition {}", partition_id); partition_ids_hint.insert(partition_id); } ReplicatedMergeTreeMergePredicate can_merge = queue.getMergePredicate(zookeeper, std::move(partition_ids_hint)); auto future_merged_part = std::make_shared(); if (storage_settings.get()->assign_part_uuids) future_merged_part->uuid = UUIDHelpers::generateV4(); constexpr const char * unknown_disable_reason = "unknown reason"; String disable_reason = unknown_disable_reason; SelectPartsDecision select_decision = SelectPartsDecision::CANNOT_SELECT; if (partition_id.empty()) { select_decision = merger_mutator.selectPartsToMerge( future_merged_part, /* aggressive */ true, storage_settings_ptr->max_bytes_to_merge_at_max_space_in_pool, can_merge, /* merge_with_ttl_allowed */ false, NO_TRANSACTION_PTR, &disable_reason); } else { select_decision = merger_mutator.selectAllPartsToMergeWithinPartition( future_merged_part, can_merge, partition_id, final, metadata_snapshot, NO_TRANSACTION_PTR, &disable_reason, query_context->getSettingsRef().optimize_skip_merged_partitions); } /// If there is nothing to merge then we treat this merge as successful (needed for optimize final optimization) if (select_decision == SelectPartsDecision::NOTHING_TO_MERGE) return false; if (select_decision != SelectPartsDecision::SELECTED) { constexpr const char * message_fmt = "Cannot select parts for optimization: {}"; assert(disable_reason != unknown_disable_reason); if (!partition_id.empty()) disable_reason += fmt::format(" (in partition {})", partition_id); return handle_noop(message_fmt, disable_reason); } ReplicatedMergeTreeLogEntryData merge_entry; CreateMergeEntryResult create_result = createLogEntryToMergeParts( zookeeper, future_merged_part->parts, future_merged_part->name, future_merged_part->uuid, future_merged_part->part_format, deduplicate, deduplicate_by_columns, cleanup, &merge_entry, can_merge.getVersion(), future_merged_part->merge_type); if (create_result == CreateMergeEntryResult::MissingPart) { static constexpr const char * message_fmt = "Can't create merge queue node in ZooKeeper, because some parts are missing"; return handle_noop(message_fmt); } if (create_result == CreateMergeEntryResult::LogUpdated) continue; merge_entries.push_back(std::move(merge_entry)); return true; } assert(try_no == max_retries); static constexpr const char * message_fmt = "Can't create merge queue node in ZooKeeper, because log was updated in every of {} tries"; return handle_noop(message_fmt, try_no); }; bool assigned = false; if (!partition && final) { if (cleanup && this->merging_params.mode != MergingParams::Mode::Replacing) { constexpr const char * message = "Cannot OPTIMIZE with CLEANUP table: {}"; String disable_reason = "only ReplacingMergeTree can be CLEANUP"; throw Exception(ErrorCodes::CANNOT_ASSIGN_OPTIMIZE, message, disable_reason); } DataPartsVector data_parts = getVisibleDataPartsVector(query_context); std::unordered_set partition_ids; for (const DataPartPtr & part : data_parts) partition_ids.emplace(part->info.partition_id); for (const String & partition_id : partition_ids) { assigned = try_assign_merge(partition_id); if (!assigned) break; } } else { String partition_id; if (partition) partition_id = getPartitionIDFromQuery(partition, query_context); assigned = try_assign_merge(partition_id); } table_lock.reset(); for (auto & merge_entry : merge_entries) waitForLogEntryToBeProcessedIfNecessary(merge_entry, query_context); return assigned; } bool StorageReplicatedMergeTree::executeMetadataAlter(const StorageReplicatedMergeTree::LogEntry & entry) { if (entry.alter_version < metadata_version) { /// TODO Can we replace it with LOGICAL_ERROR? /// As for now, it may rarely happen due to reordering of ALTER_METADATA entries in the queue of /// non-initial replica and also may happen after stale replica recovery. LOG_WARNING(log, "Attempt to update metadata of version {} " "to older version {} when processing log entry {}: {}", metadata_version, entry.alter_version, entry.znode_name, entry.toString()); return true; } auto zookeeper = getZooKeeper(); auto columns_from_entry = ColumnsDescription::parse(entry.columns_str); auto metadata_from_entry = ReplicatedMergeTreeTableMetadata::parse(entry.metadata_str); MergeTreeData::DataParts parts; /// If metadata nodes have changed, we will update table structure locally. Coordination::Requests requests; requests.emplace_back(zkutil::makeSetRequest(fs::path(replica_path) / "columns", entry.columns_str, -1)); requests.emplace_back(zkutil::makeSetRequest(fs::path(replica_path) / "metadata", entry.metadata_str, -1)); auto table_id = getStorageID(); auto alter_context = getContext(); auto database = DatabaseCatalog::instance().getDatabase(table_id.database_name); bool is_in_replicated_database = database->getEngineName() == "Replicated"; if (is_in_replicated_database) { auto mutable_alter_context = Context::createCopy(getContext()); const auto * replicated = dynamic_cast(database.get()); mutable_alter_context->makeQueryContext(); auto alter_txn = std::make_shared(zookeeper, replicated->getZooKeeperPath(), /* is_initial_query */ false, /* task_zk_path */ ""); mutable_alter_context->initZooKeeperMetadataTransaction(alter_txn); alter_context = mutable_alter_context; for (auto & op : requests) alter_txn->addOp(std::move(op)); requests.clear(); /// Requests will be executed by database in setTableStructure } else { zookeeper->multi(requests); } { auto table_lock_holder = lockForShare(RWLockImpl::NO_QUERY, getSettings()->lock_acquire_timeout_for_background_operations); auto alter_lock_holder = lockForAlter(getSettings()->lock_acquire_timeout_for_background_operations); LOG_INFO(log, "Metadata changed in ZooKeeper. Applying changes locally."); auto metadata_diff = ReplicatedMergeTreeTableMetadata(*this, getInMemoryMetadataPtr()).checkAndFindDiff(metadata_from_entry, getInMemoryMetadataPtr()->getColumns(), getContext()); setTableStructure(table_id, alter_context, std::move(columns_from_entry), metadata_diff); metadata_version = entry.alter_version; LOG_INFO(log, "Applied changes to the metadata of the table. Current metadata version: {}", metadata_version); } { /// Reset Object columns, because column of type /// Object may be added or dropped by alter. auto parts_lock = lockParts(); resetObjectColumnsFromActiveParts(parts_lock); } /// This transaction may not happen, but it's OK, because on the next retry we will eventually create/update this node /// TODO Maybe do in in one transaction for Replicated database? zookeeper->createOrUpdate(fs::path(replica_path) / "metadata_version", std::to_string(metadata_version), zkutil::CreateMode::Persistent); return true; } PartitionBlockNumbersHolder StorageReplicatedMergeTree::allocateBlockNumbersInAffectedPartitions( const MutationCommands & commands, ContextPtr query_context, const zkutil::ZooKeeperPtr & zookeeper) const { const std::set mutation_affected_partition_ids = getPartitionIdsAffectedByCommands(commands, query_context); if (mutation_affected_partition_ids.size() == 1) { const auto & affected_partition_id = *mutation_affected_partition_ids.cbegin(); auto block_number_holder = allocateBlockNumber(affected_partition_id, zookeeper); if (!block_number_holder.has_value()) return {}; auto block_number = block_number_holder->getNumber(); /// Avoid possible UB due to std::move return {{{affected_partition_id, block_number}}, std::move(block_number_holder)}; } else { /// TODO: Implement optimal block number acquisition algorithm in multiple (but not all) partitions EphemeralLocksInAllPartitions lock_holder( fs::path(zookeeper_path) / "block_numbers", "block-", fs::path(zookeeper_path) / "temp", *zookeeper); PartitionBlockNumbersHolder::BlockNumbersType block_numbers; for (const auto & lock : lock_holder.getLocks()) { if (mutation_affected_partition_ids.empty() || mutation_affected_partition_ids.contains(lock.partition_id)) block_numbers[lock.partition_id] = lock.number; } return {std::move(block_numbers), std::move(lock_holder)}; } } void StorageReplicatedMergeTree::alter( const AlterCommands & commands, ContextPtr query_context, AlterLockHolder & table_lock_holder) { assertNotReadonly(); auto table_id = getStorageID(); if (commands.isSettingsAlter()) { /// We don't replicate storage_settings_ptr ALTER. It's local operation. /// Also we don't upgrade alter lock to table structure lock. StorageInMemoryMetadata future_metadata = getInMemoryMetadata(); commands.apply(future_metadata, query_context); merge_strategy_picker.refreshState(); changeSettings(future_metadata.settings_changes, table_lock_holder); DatabaseCatalog::instance().getDatabase(table_id.database_name)->alterTable(query_context, table_id, future_metadata); return; } auto ast_to_str = [](ASTPtr query) -> String { if (!query) return ""; return queryToString(query); }; const auto zookeeper = getZooKeeperAndAssertNotReadonly(); std::optional alter_entry; std::optional mutation_znode; while (true) { /// Clear nodes from previous iteration alter_entry.emplace(); mutation_znode.reset(); auto current_metadata = getInMemoryMetadataPtr(); StorageInMemoryMetadata future_metadata = *current_metadata; commands.apply(future_metadata, query_context); ReplicatedMergeTreeTableMetadata future_metadata_in_zk(*this, current_metadata); if (ast_to_str(future_metadata.sorting_key.definition_ast) != ast_to_str(current_metadata->sorting_key.definition_ast)) { /// We serialize definition_ast as list, because code which apply ALTER (setTableStructure) expect serialized non empty expression /// list here and we cannot change this representation for compatibility. Also we have preparsed AST `sorting_key.expression_list_ast` /// in KeyDescription, but it contain version column for VersionedCollapsingMergeTree, which shouldn't be defined as a part of key definition AST. /// So the best compatible way is just to convert definition_ast to list and serialize it. In all other places key.expression_list_ast should be used. future_metadata_in_zk.sorting_key = serializeAST(*extractKeyExpressionList(future_metadata.sorting_key.definition_ast)); } if (ast_to_str(future_metadata.sampling_key.definition_ast) != ast_to_str(current_metadata->sampling_key.definition_ast)) future_metadata_in_zk.sampling_expression = serializeAST(*extractKeyExpressionList(future_metadata.sampling_key.definition_ast)); if (ast_to_str(future_metadata.partition_key.definition_ast) != ast_to_str(current_metadata->partition_key.definition_ast)) future_metadata_in_zk.partition_key = serializeAST(*extractKeyExpressionList(future_metadata.partition_key.definition_ast)); if (ast_to_str(future_metadata.table_ttl.definition_ast) != ast_to_str(current_metadata->table_ttl.definition_ast)) { if (future_metadata.table_ttl.definition_ast) future_metadata_in_zk.ttl_table = serializeAST(*future_metadata.table_ttl.definition_ast); else /// TTL was removed future_metadata_in_zk.ttl_table = ""; } String new_indices_str = future_metadata.secondary_indices.toString(); if (new_indices_str != current_metadata->secondary_indices.toString()) future_metadata_in_zk.skip_indices = new_indices_str; String new_projections_str = future_metadata.projections.toString(); if (new_projections_str != current_metadata->projections.toString()) future_metadata_in_zk.projections = new_projections_str; String new_constraints_str = future_metadata.constraints.toString(); if (new_constraints_str != current_metadata->constraints.toString()) future_metadata_in_zk.constraints = new_constraints_str; Coordination::Requests ops; size_t alter_path_idx = std::numeric_limits::max(); size_t mutation_path_idx = std::numeric_limits::max(); String new_metadata_str = future_metadata_in_zk.toString(); ops.emplace_back(zkutil::makeSetRequest(fs::path(zookeeper_path) / "metadata", new_metadata_str, metadata_version)); String new_columns_str = future_metadata.columns.toString(); ops.emplace_back(zkutil::makeSetRequest(fs::path(zookeeper_path) / "columns", new_columns_str, -1)); if (ast_to_str(current_metadata->settings_changes) != ast_to_str(future_metadata.settings_changes)) { /// Just change settings StorageInMemoryMetadata metadata_copy = *current_metadata; metadata_copy.settings_changes = future_metadata.settings_changes; changeSettings(metadata_copy.settings_changes, table_lock_holder); DatabaseCatalog::instance().getDatabase(table_id.database_name)->alterTable(query_context, table_id, metadata_copy); } /// We can be sure, that in case of successful commit in zookeeper our /// version will increments by 1. Because we update with version check. int new_metadata_version = metadata_version + 1; alter_entry->type = LogEntry::ALTER_METADATA; alter_entry->source_replica = replica_name; alter_entry->metadata_str = new_metadata_str; alter_entry->columns_str = new_columns_str; alter_entry->alter_version = new_metadata_version; alter_entry->create_time = time(nullptr); auto maybe_mutation_commands = commands.getMutationCommands( *current_metadata, query_context->getSettingsRef().materialize_ttl_after_modify, query_context); bool have_mutation = !maybe_mutation_commands.empty(); alter_entry->have_mutation = have_mutation; alter_path_idx = ops.size(); ops.emplace_back(zkutil::makeCreateRequest( fs::path(zookeeper_path) / "log/log-", alter_entry->toString(), zkutil::CreateMode::PersistentSequential)); PartitionBlockNumbersHolder partition_block_numbers_holder; if (have_mutation) { const String mutations_path(fs::path(zookeeper_path) / "mutations"); ReplicatedMergeTreeMutationEntry mutation_entry; mutation_entry.alter_version = new_metadata_version; mutation_entry.source_replica = replica_name; mutation_entry.commands = std::move(maybe_mutation_commands); Coordination::Stat mutations_stat; zookeeper->get(mutations_path, &mutations_stat); partition_block_numbers_holder = allocateBlockNumbersInAffectedPartitions(mutation_entry.commands, query_context, zookeeper); mutation_entry.block_numbers = partition_block_numbers_holder.getBlockNumbers(); mutation_entry.create_time = time(nullptr); ops.emplace_back(zkutil::makeSetRequest(mutations_path, String(), mutations_stat.version)); mutation_path_idx = ops.size(); ops.emplace_back( zkutil::makeCreateRequest(fs::path(mutations_path) / "", mutation_entry.toString(), zkutil::CreateMode::PersistentSequential)); } if (auto txn = query_context->getZooKeeperMetadataTransaction()) { /// It would be better to clone ops instead of moving, so we could retry on ZBADVERSION, /// but clone() is not implemented for Coordination::Request. txn->moveOpsTo(ops); /// NOTE: IDatabase::alterTable(...) is called when executing ALTER_METADATA queue entry without query context, /// so we have to update metadata of DatabaseReplicated here. String metadata_zk_path = fs::path(txn->getDatabaseZooKeeperPath()) / "metadata" / escapeForFileName(table_id.table_name); auto ast = DatabaseCatalog::instance().getDatabase(table_id.database_name)->getCreateTableQuery(table_id.table_name, query_context); applyMetadataChangesToCreateQuery(ast, future_metadata); ops.emplace_back(zkutil::makeSetRequest(metadata_zk_path, getObjectDefinitionFromCreateQuery(ast), -1)); } Coordination::Responses results; Coordination::Error rc = zookeeper->tryMulti(ops, results); /// For the sake of consistency with mechanics of concurrent background process of assigning parts merge tasks /// this placeholder must be held up until the moment of committing into ZK of the mutation entry /// See ReplicatedMergeTreeMergePredicate::canMergeTwoParts() method partition_block_numbers_holder.reset(); if (rc == Coordination::Error::ZOK) { if (have_mutation) { /// ALTER_METADATA record in replication /log String alter_path = dynamic_cast(*results[alter_path_idx]).path_created; alter_entry->znode_name = alter_path.substr(alter_path.find_last_of('/') + 1); /// ReplicatedMergeTreeMutationEntry record in /mutations String mutation_path = dynamic_cast(*results[mutation_path_idx]).path_created; mutation_znode = mutation_path.substr(mutation_path.find_last_of('/') + 1); } else { /// ALTER_METADATA record in replication /log String alter_path = dynamic_cast(*results[alter_path_idx]).path_created; alter_entry->znode_name = alter_path.substr(alter_path.find_last_of('/') + 1); } break; } else if (rc == Coordination::Error::ZBADVERSION) { if (results[0]->error != Coordination::Error::ZOK) throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Metadata on replica is not up to date with common metadata in Zookeeper. " "It means that this replica still not applied some of previous alters." " Probably too many alters executing concurrently (highly not recommended). " "You can retry this error"); /// Cannot retry automatically, because some zookeeper ops were lost on the first attempt. Will retry on DDLWorker-level. if (query_context->getZooKeeperMetadataTransaction()) throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot execute alter, because mutations version was suddenly changed due " "to concurrent alter"); continue; } else { throw Coordination::Exception("Alter cannot be assigned because of Zookeeper error", rc); } } table_lock_holder.unlock(); LOG_DEBUG(log, "Updated shared metadata nodes in ZooKeeper. Waiting for replicas to apply changes."); waitForLogEntryToBeProcessedIfNecessary(*alter_entry, query_context, "Some replicas doesn't finish metadata alter: "); if (mutation_znode) { LOG_DEBUG(log, "Metadata changes applied. Will wait for data changes."); waitMutation(*mutation_znode, query_context->getSettingsRef().alter_sync); LOG_DEBUG(log, "Data changes applied."); } } /// If new version returns ordinary name, else returns part name containing the first and last month of the month /// NOTE: use it in pair with getFakePartCoveringAllPartsInPartition(...) String getPartNamePossiblyFake(MergeTreeDataFormatVersion format_version, const MergeTreePartInfo & part_info) { if (format_version < MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING) { /// The date range is all month long. const auto & lut = DateLUT::instance(); time_t start_time = lut.YYYYMMDDToDate(parse(part_info.partition_id + "01")); DayNum left_date = DayNum(lut.toDayNum(start_time).toUnderType()); DayNum right_date = DayNum(static_cast(left_date) + lut.daysInMonth(start_time) - 1); return part_info.getPartNameV0(left_date, right_date); } return part_info.getPartNameV1(); } bool StorageReplicatedMergeTree::getFakePartCoveringAllPartsInPartition( const String & partition_id, MergeTreePartInfo & part_info, std::optional & delimiting_block_lock, bool for_replace_range) { /// Even if there is no data in the partition, you still need to mark the range for deletion. /// - Because before executing DETACH, tasks for downloading parts to this partition can be executed. Int64 left = 0; /** Let's skip one number in `block_numbers` for the partition being deleted, and we will only delete parts until this number. * This prohibits merges of deleted parts with the new inserted * Invariant: merges of deleted parts with other parts do not appear in the log. * NOTE: If you need to similarly support a `DROP PART` request, you will have to think of some new mechanism for it, * to guarantee this invariant. */ Int64 right; Int64 mutation_version; { delimiting_block_lock = allocateBlockNumber(partition_id, getZooKeeper()); right = delimiting_block_lock->getNumber(); /// Make sure we cover all parts in drop range. /// There might be parts with mutation version greater than current block number /// if some part mutation has been assigned after block number allocation, but before creation of DROP_RANGE entry. mutation_version = MergeTreePartInfo::MAX_BLOCK_NUMBER; } if (for_replace_range) { /// NOTE Do not decrement max block number for REPLACE_RANGE, because there are invariants: /// - drop range for REPLACE PARTITION must contain at least 2 blocks (1 skipped block and at least 1 real block) /// - drop range for MOVE PARTITION/ATTACH PARTITION FROM always contains 1 block /// NOTE UINT_MAX was previously used as max level for REPLACE/MOVE PARTITION (it was incorrect) part_info = MergeTreePartInfo(partition_id, left, right, MergeTreePartInfo::MAX_LEVEL, mutation_version); return right != 0; } /// Empty partition. if (right == 0) return false; --right; /// Artificial high level is chosen, to make this part "covering" all parts inside. part_info = MergeTreePartInfo(partition_id, left, right, MergeTreePartInfo::MAX_LEVEL, mutation_version); return true; } void StorageReplicatedMergeTree::restoreMetadataInZooKeeper() { LOG_INFO(log, "Restoring replica metadata"); if (!initialization_done) throw Exception(ErrorCodes::NOT_INITIALIZED, "Table is not initialized yet"); if (!is_readonly) throw Exception(ErrorCodes::BAD_ARGUMENTS, "Replica must be readonly"); if (getZooKeeper()->exists(replica_path)) throw Exception(ErrorCodes::BAD_ARGUMENTS, "Replica path is present at {} - nothing to restore. " "If you are sure that metadata is lost and that replica path contains some garbage, " "then use SYSTEM DROP REPLICA query first.", replica_path); if (has_metadata_in_zookeeper.has_value() && *has_metadata_in_zookeeper) throw Exception(ErrorCodes::LOGICAL_ERROR, "Replica has metadata in ZooKeeper: " "it's either a bug or it's a result of manual intervention to ZooKeeper"); if (are_restoring_replica.exchange(true)) throw Exception(ErrorCodes::CONCURRENT_ACCESS_NOT_SUPPORTED, "Replica restoration in progress"); SCOPE_EXIT({ are_restoring_replica.store(false); }); auto metadata_snapshot = getInMemoryMetadataPtr(); waitForOutdatedPartsToBeLoaded(); const DataPartsVector all_parts = getAllDataPartsVector(); Strings active_parts_names; /// Why all parts (not only Active) are moved to detached/: /// After ZK metadata restoration ZK resets sequential counters (including block number counters), so one may /// potentially encounter a situation that a part we want to attach already exists. for (const auto & part : all_parts) { if (part->getState() == DataPartState::Active) active_parts_names.push_back(part->name); forcefullyMovePartToDetachedAndRemoveFromMemory(part); } LOG_INFO(log, "Moved all parts to detached/"); const bool is_first_replica = createTableIfNotExists(metadata_snapshot); LOG_INFO(log, "Created initial ZK nodes, replica is first: {}", is_first_replica); if (!is_first_replica) createReplica(metadata_snapshot); createNewZooKeeperNodes(); LOG_INFO(log, "Created ZK nodes for table"); has_metadata_in_zookeeper = true; if (is_first_replica) for (const String& part_name : active_parts_names) attachPartition(std::make_shared(part_name), metadata_snapshot, true, getContext()); LOG_INFO(log, "Attached all partitions, starting table"); startupImpl(/* from_attach_thread */ false); } void StorageReplicatedMergeTree::dropPartNoWaitNoThrow(const String & part_name) { assertNotReadonly(); if (!is_leader) throw Exception(ErrorCodes::NOT_A_LEADER, "DROP PART cannot be done on this replica because it is not a leader"); zkutil::ZooKeeperPtr zookeeper = getZooKeeperAndAssertNotReadonly(); LogEntry entry; dropPartImpl(zookeeper, part_name, entry, /*detach=*/ false, /*throw_if_noop=*/ false); } void StorageReplicatedMergeTree::dropPart(const String & part_name, bool detach, ContextPtr query_context) { assertNotReadonly(); if (!is_leader) throw Exception(ErrorCodes::NOT_A_LEADER, "DROP PART cannot be done on this replica because it is not a leader"); zkutil::ZooKeeperPtr zookeeper = getZooKeeperAndAssertNotReadonly(); LogEntry entry; dropPartImpl(zookeeper, part_name, entry, detach, /*throw_if_noop=*/ true); waitForLogEntryToBeProcessedIfNecessary(entry, query_context); } void StorageReplicatedMergeTree::dropAllPartitionsImpl(const zkutil::ZooKeeperPtr & zookeeper, bool detach, ContextPtr query_context) { Strings partitions = zookeeper->getChildren(fs::path(zookeeper_path) / "block_numbers"); std::vector entries; dropAllPartsInPartitions(*zookeeper, partitions, entries, query_context, detach); for (const auto & entry : entries) { waitForLogEntryToBeProcessedIfNecessary(*entry, query_context); auto drop_range_info = MergeTreePartInfo::fromPartName(entry->new_part_name, format_version); cleanLastPartNode(drop_range_info.partition_id); } } void StorageReplicatedMergeTree::dropPartition(const ASTPtr & partition, bool detach, ContextPtr query_context) { assertNotReadonly(); if (!is_leader) throw Exception(ErrorCodes::NOT_A_LEADER, "DROP PARTITION cannot be done on this replica because it is not a leader"); zkutil::ZooKeeperPtr zookeeper = getZooKeeperAndAssertNotReadonly(); const auto * partition_ast = partition->as(); if (partition_ast && partition_ast->all) { dropAllPartitionsImpl(zookeeper, detach, query_context); } else { String partition_id = getPartitionIDFromQuery(partition, query_context); auto entry = dropAllPartsInPartition(*zookeeper, partition_id, query_context, detach); if (entry) { waitForLogEntryToBeProcessedIfNecessary(*entry, query_context); cleanLastPartNode(partition_id); } } } void StorageReplicatedMergeTree::truncate( const ASTPtr &, const StorageMetadataPtr &, ContextPtr query_context, TableExclusiveLockHolder & table_lock) { table_lock.release(); /// Truncate is done asynchronously. assertNotReadonly(); if (!is_leader) throw Exception(ErrorCodes::NOT_A_LEADER, "TRUNCATE cannot be done on this replica because it is not a leader"); waitForOutdatedPartsToBeLoaded(); zkutil::ZooKeeperPtr zookeeper = getZooKeeperAndAssertNotReadonly(); dropAllPartitionsImpl(zookeeper, /* detach */ false, query_context); } PartitionCommandsResultInfo StorageReplicatedMergeTree::attachPartition( const ASTPtr & partition, const StorageMetadataPtr & metadata_snapshot, bool attach_part, ContextPtr query_context) { /// Allow ATTACH PARTITION on readonly replica when restoring it. if (!are_restoring_replica) assertNotReadonly(); PartitionCommandsResultInfo results; PartsTemporaryRename renamed_parts(*this, "detached/"); MutableDataPartsVector loaded_parts = tryLoadPartsToAttach(partition, attach_part, query_context, renamed_parts); /// TODO Allow to use quorum here. ReplicatedMergeTreeSink output(*this, metadata_snapshot, 0, 0, 0, false, false, false, query_context, /*is_attach*/true); for (size_t i = 0; i < loaded_parts.size(); ++i) { const String old_name = loaded_parts[i]->name; output.writeExistingPart(loaded_parts[i]); renamed_parts.old_and_new_names[i].old_name.clear(); LOG_DEBUG(log, "Attached part {} as {}", old_name, loaded_parts[i]->name); results.push_back(PartitionCommandResultInfo{ .partition_id = loaded_parts[i]->info.partition_id, .part_name = loaded_parts[i]->name, .old_part_name = old_name, }); } return results; } void StorageReplicatedMergeTree::checkTableCanBeDropped() const { auto table_id = getStorageID(); getContext()->checkTableCanBeDropped(table_id.database_name, table_id.table_name, getTotalActiveSizeInBytes()); } void StorageReplicatedMergeTree::checkTableCanBeRenamed(const StorageID & new_name) const { if (renaming_restrictions == RenamingRestrictions::ALLOW_ANY) return; if (renaming_restrictions == RenamingRestrictions::DO_NOT_ALLOW) { auto old_name = getStorageID(); bool is_server_startup = Context::getGlobalContextInstance()->getApplicationType() == Context::ApplicationType::SERVER && !Context::getGlobalContextInstance()->isServerCompletelyStarted(); bool move_to_atomic = old_name.uuid == UUIDHelpers::Nil && new_name.uuid != UUIDHelpers::Nil; bool likely_converting_ordinary_to_atomic = is_server_startup && move_to_atomic; if (likely_converting_ordinary_to_atomic) { LOG_INFO(log, "Table {} should not be renamed, because zookeeper_path contains implicit 'database' or 'table' macro. " "We cannot rename path in ZooKeeper, so path may become inconsistent with table name. " "However, we allow renaming while converting Ordinary database to Atomic, because all tables will be renamed back", old_name.getNameForLogs()); return; } throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Cannot rename Replicated table, because zookeeper_path contains implicit 'database' " "or 'table' macro. We cannot rename path " "in ZooKeeper, so path may become inconsistent with table name. " "If you really want to rename table, you should edit metadata file first and restart server or reattach the table."); } assert(renaming_restrictions == RenamingRestrictions::ALLOW_PRESERVING_UUID); if (!new_name.hasUUID() && getStorageID().hasUUID()) throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Cannot move Replicated table to Ordinary database, because zookeeper_path contains implicit " "'uuid' macro. If you really want to rename table, you should edit metadata file first " "and restart server or reattach the table."); } void StorageReplicatedMergeTree::rename(const String & new_path_to_table_data, const StorageID & new_table_id) { checkTableCanBeRenamed(new_table_id); MergeTreeData::rename(new_path_to_table_data, new_table_id); /// Update table name in zookeeper if (!is_readonly) { /// We don't do it for readonly tables, because it will be updated on next table startup. /// It is also Ok to skip ZK error for the same reason. try { auto zookeeper = getZooKeeper(); zookeeper->set(fs::path(replica_path) / "host", getReplicatedMergeTreeAddress().toString()); } catch (Coordination::Exception & e) { LOG_WARNING(log, "Cannot update the value of 'host' node (replica address) in ZooKeeper: {}", e.displayText()); } } /// TODO: You can update names of loggers. } bool StorageReplicatedMergeTree::existsNodeCached(const ZooKeeperWithFaultInjectionPtr & zookeeper, const std::string & path) const { { std::lock_guard lock(existing_nodes_cache_mutex); if (existing_nodes_cache.contains(path)) return true; } bool res = zookeeper->exists(path); if (res) { std::lock_guard lock(existing_nodes_cache_mutex); existing_nodes_cache.insert(path); } return res; } std::optional StorageReplicatedMergeTree::allocateBlockNumber( const String & partition_id, const zkutil::ZooKeeperPtr & zookeeper, const String & zookeeper_block_id_path, const String & zookeeper_path_prefix) const { return allocateBlockNumber( partition_id, std::make_shared(zookeeper), zookeeper_block_id_path, zookeeper_path_prefix); } template std::optional StorageReplicatedMergeTree::allocateBlockNumber( const String & partition_id, const ZooKeeperWithFaultInjectionPtr & zookeeper, const T & zookeeper_block_id_path, const String & zookeeper_path_prefix) const { String zookeeper_table_path; if (zookeeper_path_prefix.empty()) zookeeper_table_path = zookeeper_path; else zookeeper_table_path = zookeeper_path_prefix; String block_numbers_path = fs::path(zookeeper_table_path) / "block_numbers"; String partition_path = fs::path(block_numbers_path) / partition_id; if (!existsNodeCached(zookeeper, partition_path)) { Coordination::Requests ops; /// Check that table is not being dropped ("host" is the first node that is removed on replica drop) ops.push_back(zkutil::makeCheckRequest(fs::path(replica_path) / "host", -1)); ops.push_back(zkutil::makeCreateRequest(partition_path, "", zkutil::CreateMode::Persistent)); /// We increment data version of the block_numbers node so that it becomes possible /// to check in a ZK transaction that the set of partitions didn't change /// (unfortunately there is no CheckChildren op). ops.push_back(zkutil::makeSetRequest(block_numbers_path, "", -1)); Coordination::Responses responses; Coordination::Error code = zookeeper->tryMulti(ops, responses); if (code != Coordination::Error::ZOK && code != Coordination::Error::ZNODEEXISTS) zkutil::KeeperMultiException::check(code, ops, responses); } return createEphemeralLockInZooKeeper( fs::path(partition_path) / "block-", fs::path(zookeeper_table_path) / "temp", zookeeper, zookeeper_block_id_path); } Strings StorageReplicatedMergeTree::tryWaitForAllReplicasToProcessLogEntry( const String & table_zookeeper_path, const ReplicatedMergeTreeLogEntryData & entry, Int64 wait_for_inactive_timeout) { LOG_DEBUG(log, "Waiting for all replicas to process {}", entry.znode_name); auto zookeeper = getZooKeeper(); Strings replicas = zookeeper->getChildren(fs::path(table_zookeeper_path) / "replicas"); Strings unwaited; bool wait_for_inactive = wait_for_inactive_timeout != 0; for (const String & replica : replicas) { if (wait_for_inactive || zookeeper->exists(fs::path(table_zookeeper_path) / "replicas" / replica / "is_active")) { if (!tryWaitForReplicaToProcessLogEntry(table_zookeeper_path, replica, entry, wait_for_inactive_timeout)) unwaited.push_back(replica); } else { unwaited.push_back(replica); } } LOG_DEBUG(log, "Finished waiting for all replicas to process {}", entry.znode_name); return unwaited; } void StorageReplicatedMergeTree::waitForAllReplicasToProcessLogEntry( const String & table_zookeeper_path, const ReplicatedMergeTreeLogEntryData & entry, Int64 wait_for_inactive_timeout, const String & error_context) { Strings unfinished_replicas = tryWaitForAllReplicasToProcessLogEntry(table_zookeeper_path, entry, wait_for_inactive_timeout); if (unfinished_replicas.empty()) return; throw Exception(ErrorCodes::UNFINISHED, "{}Timeout exceeded while waiting for replicas {} to process entry {}. " "Probably some replicas are inactive", error_context, fmt::join(unfinished_replicas, ", "), entry.znode_name); } void StorageReplicatedMergeTree::waitForLogEntryToBeProcessedIfNecessary(const ReplicatedMergeTreeLogEntryData & entry, ContextPtr query_context, const String & error_context) { /// If necessary, wait until the operation is performed on itself or on all replicas. Int64 wait_for_inactive_timeout = query_context->getSettingsRef().replication_wait_for_inactive_replica_timeout; if (query_context->getSettingsRef().alter_sync == 1) { bool finished = tryWaitForReplicaToProcessLogEntry(zookeeper_path, replica_name, entry, wait_for_inactive_timeout); if (!finished) { throw Exception(ErrorCodes::UNFINISHED, "{}Log entry {} is not precessed on local replica, " "most likely because the replica was shut down.", error_context, entry.znode_name); } } else if (query_context->getSettingsRef().alter_sync == 2) { waitForAllReplicasToProcessLogEntry(zookeeper_path, entry, wait_for_inactive_timeout, error_context); } } bool StorageReplicatedMergeTree::tryWaitForReplicaToProcessLogEntry( const String & table_zookeeper_path, const String & replica, const ReplicatedMergeTreeLogEntryData & entry, Int64 wait_for_inactive_timeout) { String entry_str = entry.toString(); String log_node_name; /** Wait for entries from `log` directory (a common log, from where replicas copy entries to their queue) to be processed. * * The problem is that the numbers (`sequential` node) of the queue elements in `log` and in `queue` do not match. * (And the numbers of the same log element for different replicas do not match in the `queue`.) */ /** First, you need to wait until replica takes `queue` element from the `log` to its queue, * if it has not been done already (see the `pullLogsToQueue` function). * * To do this, check its node `log_pointer` - the maximum number of the element taken from `log` + 1. */ bool waiting_itself = replica == replica_name; /// Do not wait if timeout is zero bool wait_for_inactive = wait_for_inactive_timeout != 0; /// Wait for unlimited time if timeout is negative bool check_timeout = wait_for_inactive_timeout > 0; Stopwatch time_waiting; const auto & stop_waiting = [&]() { bool stop_waiting_itself = waiting_itself && partial_shutdown_called; bool timeout_exceeded = check_timeout && wait_for_inactive_timeout < time_waiting.elapsedSeconds(); bool stop_waiting_inactive = (!wait_for_inactive || timeout_exceeded) && !getZooKeeper()->exists(fs::path(table_zookeeper_path) / "replicas" / replica / "is_active"); return is_dropped || stop_waiting_itself || stop_waiting_inactive; }; /// Don't recheck ZooKeeper too often constexpr auto event_wait_timeout_ms = 3000; LOG_DEBUG(log, "Waiting for {} to process log entry", replica); if (startsWith(entry.znode_name, "log-")) { /// Take the number from the node name `log-xxxxxxxxxx`. UInt64 log_index = parse(entry.znode_name.substr(entry.znode_name.size() - 10)); log_node_name = entry.znode_name; LOG_DEBUG(log, "Waiting for {} to pull {} to queue", replica, log_node_name); /// Let's wait until entry gets into the replica queue. bool pulled_to_queue = false; do { zkutil::EventPtr event = std::make_shared(); String log_pointer = getZooKeeper()->get(fs::path(table_zookeeper_path) / "replicas" / replica / "log_pointer", nullptr, event); if (!log_pointer.empty() && parse(log_pointer) > log_index) { pulled_to_queue = true; break; } /// Wait with timeout because we can be already shut down, but not dropped. /// So log_pointer node will exist, but we will never update it because all background threads already stopped. /// It can lead to query hung because table drop query can wait for some query (alter, optimize, etc) which called this method, /// but the query will never finish because the drop already shut down the table. if (!stop_waiting()) event->tryWait(event_wait_timeout_ms); } while (!stop_waiting()); if (!pulled_to_queue) return false; LOG_DEBUG(log, "Looking for node corresponding to {} in {} queue", log_node_name, replica); } else if (!entry.log_entry_id.empty()) { /// First pass, check the table log. /// If found in the log, wait for replica to fetch it to the queue. /// If not found in the log, it is already in the queue. LOG_DEBUG(log, "Looking for log entry with id `{}` in the log", entry.log_entry_id); String log_pointer = getZooKeeper()->get(fs::path(table_zookeeper_path) / "replicas" / replica / "log_pointer"); Strings log_entries = getZooKeeper()->getChildren(fs::path(table_zookeeper_path) / "log"); UInt64 log_index = 0; bool found = false; for (const String & log_entry_name : log_entries) { log_index = parse(log_entry_name.substr(log_entry_name.size() - 10)); if (!log_pointer.empty() && log_index < parse(log_pointer)) continue; String log_entry_str; Coordination::Stat log_entry_stat; bool exists = getZooKeeper()->tryGet(fs::path(table_zookeeper_path) / "log" / log_entry_name, log_entry_str, &log_entry_stat); ReplicatedMergeTreeLogEntryData log_entry = *ReplicatedMergeTreeLogEntry::parse(log_entry_str, log_entry_stat, format_version); if (exists && entry.log_entry_id == log_entry.log_entry_id) { LOG_DEBUG(log, "Found log entry with id `{}` in the log", entry.log_entry_id); found = true; log_node_name = log_entry_name; break; } } if (found) { LOG_DEBUG(log, "Waiting for {} to pull {} to queue", replica, log_node_name); /// Let's wait until entry gets into the replica queue. bool pulled_to_queue = false; do { zkutil::EventPtr event = std::make_shared(); log_pointer = getZooKeeper()->get(fs::path(table_zookeeper_path) / "replicas" / replica / "log_pointer", nullptr, event); if (!log_pointer.empty() && parse(log_pointer) > log_index) { pulled_to_queue = true; break; } /// Wait with timeout because we can be already shut down, but not dropped. /// So log_pointer node will exist, but we will never update it because all background threads already stopped. /// It can lead to query hung because table drop query can wait for some query (alter, optimize, etc) which called this method, /// but the query will never finish because the drop already shut down the table. if (!stop_waiting()) event->tryWait(event_wait_timeout_ms); } while (!stop_waiting()); if (!pulled_to_queue) return false; } } else { throw Exception(ErrorCodes::LOGICAL_ERROR, "Logical error: unexpected name of log node: {}", entry.znode_name); } /** Second - find the corresponding entry in the queue of the specified replica. * Its number may not match the `log` node. Therefore, we search by comparing the content. */ Strings queue_entries = getZooKeeper()->getChildren(fs::path(table_zookeeper_path) / "replicas" / replica / "queue"); String queue_entry_to_wait_for; for (const String & entry_name : queue_entries) { String queue_entry_str; Coordination::Stat queue_entry_stat; bool exists = getZooKeeper()->tryGet(fs::path(table_zookeeper_path) / "replicas" / replica / "queue" / entry_name, queue_entry_str, &queue_entry_stat); if (exists && queue_entry_str == entry_str) { queue_entry_to_wait_for = entry_name; break; } else if (!entry.log_entry_id.empty()) { /// Check if the id matches rather than just contents. This entry /// might have been written by different ClickHouse versions and /// it is hard to guarantee same text representation. ReplicatedMergeTreeLogEntryData queue_entry = *ReplicatedMergeTreeLogEntry::parse(queue_entry_str, queue_entry_stat, format_version); if (entry.log_entry_id == queue_entry.log_entry_id) { queue_entry_to_wait_for = entry_name; break; } } } /// While looking for the record, it has already been executed and deleted. if (queue_entry_to_wait_for.empty()) { LOG_DEBUG(log, "No corresponding node found. Assuming it has been already processed. Found {} nodes", queue_entries.size()); return true; } LOG_DEBUG(log, "Waiting for {} to disappear from {} queue", queue_entry_to_wait_for, replica); /// Third - wait until the entry disappears from the replica queue or replica become inactive. String path_to_wait_on = fs::path(table_zookeeper_path) / "replicas" / replica / "queue" / queue_entry_to_wait_for; return getZooKeeper()->waitForDisappear(path_to_wait_on, stop_waiting); } void StorageReplicatedMergeTree::getStatus(ReplicatedTableStatus & res, bool with_zk_fields) { auto zookeeper = tryGetZooKeeper(); const auto storage_settings_ptr = getSettings(); res.is_leader = is_leader; res.can_become_leader = storage_settings_ptr->replicated_can_become_leader; res.is_readonly = is_readonly; res.is_session_expired = !zookeeper || zookeeper->expired(); res.queue = queue.getStatus(); res.absolute_delay = getAbsoluteDelay(); /// NOTE: may be slightly inconsistent with queue status. /// NOTE: consider convert to UInt64 res.parts_to_check = static_cast(part_check_thread.size()); res.zookeeper_path = zookeeper_path; res.replica_name = replica_name; res.replica_path = replica_path; res.columns_version = -1; res.log_max_index = 0; res.log_pointer = 0; res.total_replicas = 0; res.active_replicas = 0; res.last_queue_update_exception = getLastQueueUpdateException(); if (with_zk_fields && !res.is_session_expired) { try { std::vector paths; paths.push_back(fs::path(zookeeper_path) / "log"); paths.push_back(fs::path(zookeeper_path) / "replicas"); auto children_result = zookeeper->getChildren(paths); const auto & log_entries = children_result[0].names; const auto & all_replicas = children_result[1].names; paths.clear(); paths.push_back(fs::path(replica_path) / "log_pointer"); for (const String & replica : all_replicas) paths.push_back(fs::path(zookeeper_path) / "replicas" / replica / "is_active"); auto get_result = zookeeper->tryGet(paths); const auto & log_pointer_str = get_result[0].data; if (get_result[0].error == Coordination::Error::ZNONODE) throw zkutil::KeeperException(get_result[0].error); if (!log_entries.empty()) { const String & last_log_entry = *std::max_element(log_entries.begin(), log_entries.end()); res.log_max_index = parse(last_log_entry.substr(strlen("log-"))); } res.log_pointer = log_pointer_str.empty() ? 0 : parse(log_pointer_str); res.total_replicas = all_replicas.size(); for (size_t i = 0, size = all_replicas.size(); i < size; ++i) { bool is_replica_active = get_result[i + 1].error != Coordination::Error::ZNONODE; res.active_replicas += static_cast(is_replica_active); res.replica_is_active.emplace(all_replicas[i], is_replica_active); } } catch (const Coordination::Exception &) { res.zookeeper_exception = getCurrentExceptionMessage(false); } } } void StorageReplicatedMergeTree::getQueue(LogEntriesData & res, String & replica_name_) { replica_name_ = replica_name; queue.getEntries(res); } std::vector StorageReplicatedMergeTree::getPartMovesBetweenShardsEntries() { return part_moves_between_shards_orchestrator.getEntries(); } time_t StorageReplicatedMergeTree::getAbsoluteDelay() const { time_t min_unprocessed_insert_time = 0; time_t max_processed_insert_time = 0; queue.getInsertTimes(min_unprocessed_insert_time, max_processed_insert_time); /// Load start time, then finish time to avoid reporting false delay when start time is updated /// between loading of two variables. time_t queue_update_start_time = last_queue_update_start_time.load(); time_t queue_update_finish_time = last_queue_update_finish_time.load(); time_t current_time = time(nullptr); if (!queue_update_finish_time) { /// We have not updated queue even once yet (perhaps replica is readonly). /// As we have no info about the current state of replication log, return effectively infinite delay. return current_time; } else if (min_unprocessed_insert_time) { /// There are some unprocessed insert entries in queue. return (current_time > min_unprocessed_insert_time) ? (current_time - min_unprocessed_insert_time) : 0; } else if (queue_update_start_time > queue_update_finish_time) { /// Queue is empty, but there are some in-flight or failed queue update attempts /// (likely because of problems with connecting to ZooKeeper). /// Return the time passed since last attempt. return (current_time > queue_update_start_time) ? (current_time - queue_update_start_time) : 0; } else { /// Everything is up-to-date. return 0; } } void StorageReplicatedMergeTree::getReplicaDelays(time_t & out_absolute_delay, time_t & out_relative_delay) { assertNotReadonly(); time_t current_time = time(nullptr); out_absolute_delay = getAbsoluteDelay(); out_relative_delay = 0; const auto storage_settings_ptr = getSettings(); /** Relative delay is the maximum difference of absolute delay from any other replica, * (if this replica lags behind any other live replica, or zero, otherwise). * Calculated only if the absolute delay is large enough. */ if (out_absolute_delay < static_cast(storage_settings_ptr->min_relative_delay_to_measure)) return; auto zookeeper = getZooKeeper(); time_t max_replicas_unprocessed_insert_time = 0; bool have_replica_with_nothing_unprocessed = false; Strings replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); for (const auto & replica : replicas) { if (replica == replica_name) continue; /// Skip dead replicas. if (!zookeeper->exists(fs::path(zookeeper_path) / "replicas" / replica / "is_active")) continue; String value; if (!zookeeper->tryGet(fs::path(zookeeper_path) / "replicas" / replica / "min_unprocessed_insert_time", value)) continue; time_t replica_time = value.empty() ? 0 : parse(value); if (replica_time == 0) { /** Note * The conclusion that the replica does not lag may be incorrect, * because the information about `min_unprocessed_insert_time` is taken * only from that part of the log that has been moved to the queue. * If the replica for some reason has stalled `queueUpdatingTask`, * then `min_unprocessed_insert_time` will be incorrect. */ have_replica_with_nothing_unprocessed = true; break; } if (replica_time > max_replicas_unprocessed_insert_time) max_replicas_unprocessed_insert_time = replica_time; } if (have_replica_with_nothing_unprocessed) out_relative_delay = out_absolute_delay; else { max_replicas_unprocessed_insert_time = std::min(current_time, max_replicas_unprocessed_insert_time); time_t min_replicas_delay = current_time - max_replicas_unprocessed_insert_time; if (out_absolute_delay > min_replicas_delay) out_relative_delay = out_absolute_delay - min_replicas_delay; } } void StorageReplicatedMergeTree::fetchPartition( const ASTPtr & partition, const StorageMetadataPtr & metadata_snapshot, const String & from_, bool fetch_part, ContextPtr query_context) { Macros::MacroExpansionInfo info; info.expand_special_macros_only = false; info.table_id = getStorageID(); info.table_id.uuid = UUIDHelpers::Nil; auto expand_from = query_context->getMacros()->expand(from_, info); String auxiliary_zookeeper_name = zkutil::extractZooKeeperName(expand_from); String from = zkutil::extractZooKeeperPath(expand_from, /* check_starts_with_slash */ true); if (from.empty()) throw Exception(ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT, "ZooKeeper path should not be empty"); zkutil::ZooKeeperPtr zookeeper; if (auxiliary_zookeeper_name != default_zookeeper_name) zookeeper = getContext()->getAuxiliaryZooKeeper(auxiliary_zookeeper_name); else zookeeper = getZooKeeper(); if (from.back() == '/') from.resize(from.size() - 1); if (fetch_part) { String part_name = partition->as().value.safeGet(); auto part_path = findReplicaHavingPart(part_name, from, zookeeper); if (part_path.empty()) throw Exception(ErrorCodes::NO_REPLICA_HAS_PART, "Part {} does not exist on any replica", part_name); /** Let's check that there is no such part in the `detached` directory (where we will write the downloaded parts). * Unreliable (there is a race condition) - such a part may appear a little later. */ if (checkIfDetachedPartExists(part_name)) throw Exception(ErrorCodes::DUPLICATE_DATA_PART, "Detached part {} already exists.", part_name); LOG_INFO(log, "Will fetch part {} from shard {} (zookeeper '{}')", part_name, from_, auxiliary_zookeeper_name); try { /// part name , metadata, part_path , true, 0, zookeeper if (!fetchPart(part_name, metadata_snapshot, part_path, true, 0, zookeeper, /* try_fetch_shared = */ false)) throw Exception(ErrorCodes::UNFINISHED, "Failed to fetch part {} from {}", part_name, from_); } catch (const DB::Exception & e) { if (e.code() != ErrorCodes::RECEIVED_ERROR_FROM_REMOTE_IO_SERVER && e.code() != ErrorCodes::RECEIVED_ERROR_TOO_MANY_REQUESTS && e.code() != ErrorCodes::CANNOT_READ_ALL_DATA) throw; LOG_INFO(log, getExceptionMessageAndPattern(e, /* with_stacktrace */ false)); } return; } String partition_id = getPartitionIDFromQuery(partition, query_context); LOG_INFO(log, "Will fetch partition {} from shard {} (zookeeper '{}')", partition_id, from_, auxiliary_zookeeper_name); /** Let's check that there is no such partition in the `detached` directory (where we will write the downloaded parts). * Unreliable (there is a race condition) - such a partition may appear a little later. */ if (checkIfDetachedPartitionExists(partition_id)) throw Exception(ErrorCodes::PARTITION_ALREADY_EXISTS, "Detached partition {} already exists.", partition_id); zkutil::Strings replicas; zkutil::Strings active_replicas; String best_replica; { /// List of replicas of source shard. replicas = zookeeper->getChildren(fs::path(from) / "replicas"); /// Leave only active replicas. active_replicas.reserve(replicas.size()); for (const String & replica : replicas) if (zookeeper->exists(fs::path(from) / "replicas" / replica / "is_active")) active_replicas.push_back(replica); if (active_replicas.empty()) throw Exception(ErrorCodes::NO_ACTIVE_REPLICAS, "No active replicas for shard {}", from); /** You must select the best (most relevant) replica. * This is a replica with the maximum `log_pointer`, then with the minimum `queue` size. * NOTE This is not exactly the best criteria. It does not make sense to download old partitions, * and it would be nice to be able to choose the replica closest by network. * NOTE Of course, there are data races here. You can solve it by retrying. */ Int64 max_log_pointer = -1; UInt64 min_queue_size = std::numeric_limits::max(); for (const String & replica : active_replicas) { String current_replica_path = fs::path(from) / "replicas" / replica; String log_pointer_str = zookeeper->get(fs::path(current_replica_path) / "log_pointer"); Int64 log_pointer = log_pointer_str.empty() ? 0 : parse(log_pointer_str); Coordination::Stat stat; zookeeper->get(fs::path(current_replica_path) / "queue", &stat); size_t queue_size = stat.numChildren; if (log_pointer > max_log_pointer || (log_pointer == max_log_pointer && queue_size < min_queue_size)) { max_log_pointer = log_pointer; min_queue_size = queue_size; best_replica = replica; } } } if (best_replica.empty()) throw Exception(ErrorCodes::LOGICAL_ERROR, "Logical error: cannot choose best replica."); LOG_INFO(log, "Found {} replicas, {} of them are active. Selected {} to fetch from.", replicas.size(), active_replicas.size(), best_replica); String best_replica_path = fs::path(from) / "replicas" / best_replica; /// Let's find out which parts are on the best replica. /** Trying to download these parts. * Some of them could be deleted due to the merge. * In this case, update the information about the available parts and try again. */ unsigned try_no = 0; Strings missing_parts; do { if (try_no) LOG_INFO(log, "Some of parts ({}) are missing. Will try to fetch covering parts.", missing_parts.size()); if (try_no >= query_context->getSettings().max_fetch_partition_retries_count) throw Exception(ErrorCodes::TOO_MANY_RETRIES_TO_FETCH_PARTS, "Too many retries to fetch parts from {}", best_replica_path); Strings parts = zookeeper->getChildren(fs::path(best_replica_path) / "parts"); ActiveDataPartSet active_parts_set(format_version, parts); Strings parts_to_fetch; if (missing_parts.empty()) { parts_to_fetch = active_parts_set.getParts(); /// Leaving only the parts of the desired partition. Strings parts_to_fetch_partition; for (const String & part : parts_to_fetch) { if (MergeTreePartInfo::fromPartName(part, format_version).partition_id == partition_id) parts_to_fetch_partition.push_back(part); } parts_to_fetch = std::move(parts_to_fetch_partition); if (parts_to_fetch.empty()) throw Exception(ErrorCodes::PARTITION_DOESNT_EXIST, "Partition {} on {} doesn't exist", partition_id, best_replica_path); } else { for (const String & missing_part : missing_parts) { String containing_part = active_parts_set.getContainingPart(missing_part); if (!containing_part.empty()) parts_to_fetch.push_back(containing_part); else LOG_WARNING(log, "Part {} on replica {} has been vanished.", missing_part, best_replica_path); } } LOG_INFO(log, "Parts to fetch: {}", parts_to_fetch.size()); missing_parts.clear(); for (const String & part : parts_to_fetch) { bool fetched = false; try { fetched = fetchPart(part, metadata_snapshot, best_replica_path, true, 0, zookeeper, /* try_fetch_shared = */ false); } catch (const DB::Exception & e) { if (e.code() != ErrorCodes::RECEIVED_ERROR_FROM_REMOTE_IO_SERVER && e.code() != ErrorCodes::RECEIVED_ERROR_TOO_MANY_REQUESTS && e.code() != ErrorCodes::CANNOT_READ_ALL_DATA) throw; LOG_INFO(log, getExceptionMessageAndPattern(e, /* with_stacktrace */ false)); } if (!fetched) missing_parts.push_back(part); } ++try_no; } while (!missing_parts.empty()); } void StorageReplicatedMergeTree::mutate(const MutationCommands & commands, ContextPtr query_context) { /// Overview of the mutation algorithm. /// /// When the client executes a mutation, this method is called. It acquires block numbers in all /// partitions, saves them in the mutation entry and writes the mutation entry to a new ZK node in /// the /mutations folder. This block numbers are needed to determine which parts should be mutated and /// which shouldn't (parts inserted after the mutation will have the block number higher than the /// block number acquired by the mutation in that partition and so will not be mutated). /// This block number is called "mutation version" in that partition. /// /// Mutation versions are acquired atomically in all partitions, so the case when an insert in some /// partition has the block number higher than the mutation version but the following insert into another /// partition acquires the block number lower than the mutation version in that partition is impossible. /// Another important invariant: mutation entries appear in /mutations in the order of their mutation /// versions (in any partition). This means that mutations form a sequence and we can execute them in /// the order of their mutation versions and not worry that some mutation with the smaller version /// will suddenly appear. /// /// During mutations individual parts are immutable - when we want to change the contents of a part /// we prepare the new part and add it to MergeTreeData (the original part gets replaced). The fact that /// we have mutated the part is recorded in the part->info.mutation field of MergeTreePartInfo. /// The relation with the original part is preserved because the new part covers the same block range /// as the original one. /// /// We then can for each part determine its "mutation version": the version of the last mutation in /// the mutation sequence that we regard as already applied to that part. All mutations with the greater /// version number will still need to be applied to that part. /// /// Execution of mutations is done asynchronously. All replicas watch the /mutations directory and /// load new mutation entries as they appear (see mutationsUpdatingTask()). Next we need to determine /// how to mutate individual parts consistently with part merges. This is done by the leader replica /// (see mergeSelectingTask() and class ReplicatedMergeTreeMergePredicate for details). Important /// invariants here are that a) all source parts for a single merge must have the same mutation version /// and b) any part can be mutated only once or merged only once (e.g. once we have decided to mutate /// a part then we need to execute that mutation and can assign merges only to the new part and not to the /// original part). Multiple consecutive mutations can be executed at once (without writing the /// intermediate result to a part). /// /// Leader replica records its decisions to the replication log (/log directory in ZK) in the form of /// MUTATE_PART entries and all replicas then execute them in the background pool /// (see MutateTask class). When a replica encounters a MUTATE_PART command, it is /// guaranteed that the corresponding mutation entry is already loaded (when we pull entries from /// replication log into the replica queue, we also load mutation entries). Note that just as with merges /// the replica can decide not to do the mutation locally and fetch the mutated part from another replica /// instead. /// /// Mutations of individual parts are in fact pretty similar to merges, e.g. their assignment and execution /// is governed by the same storage_settings. TODO: support a single "merge-mutation" operation when the data /// read from the the source parts is first mutated on the fly to some uniform mutation version and then /// merged to a resulting part. /// /// After all needed parts are mutated (i.e. all active parts have the mutation version greater than /// the version of this mutation), the mutation is considered done and can be deleted. ReplicatedMergeTreeMutationEntry mutation_entry; mutation_entry.source_replica = replica_name; mutation_entry.commands = commands; const String mutations_path = fs::path(zookeeper_path) / "mutations"; const auto zookeeper = getZooKeeper(); /// Update the mutations_path node when creating the mutation and check its version to ensure that /// nodes for mutations are created in the same order as the corresponding block numbers. /// Should work well if the number of concurrent mutation requests is small. while (true) { Coordination::Stat mutations_stat; zookeeper->get(mutations_path, &mutations_stat); PartitionBlockNumbersHolder partition_block_numbers_holder = allocateBlockNumbersInAffectedPartitions(mutation_entry.commands, query_context, zookeeper); mutation_entry.block_numbers = partition_block_numbers_holder.getBlockNumbers(); mutation_entry.create_time = time(nullptr); /// The following version check guarantees the linearizability property for any pair of mutations: /// mutation with higher sequence number is guaranteed to have higher block numbers in every partition /// (and thus will be applied strictly according to sequence numbers of mutations) Coordination::Requests requests; requests.emplace_back(zkutil::makeSetRequest(mutations_path, String(), mutations_stat.version)); requests.emplace_back(zkutil::makeCreateRequest( fs::path(mutations_path) / "", mutation_entry.toString(), zkutil::CreateMode::PersistentSequential)); if (auto txn = query_context->getZooKeeperMetadataTransaction()) txn->moveOpsTo(requests); Coordination::Responses responses; Coordination::Error rc = zookeeper->tryMulti(requests, responses); partition_block_numbers_holder.reset(); if (rc == Coordination::Error::ZOK) { const String & path_created = dynamic_cast(responses[1].get())->path_created; mutation_entry.znode_name = path_created.substr(path_created.find_last_of('/') + 1); LOG_TRACE(log, "Created mutation with ID {}", mutation_entry.znode_name); break; } else if (rc == Coordination::Error::ZBADVERSION) { /// Cannot retry automatically, because some zookeeper ops were lost on the first attempt. Will retry on DDLWorker-level. if (query_context->getZooKeeperMetadataTransaction()) throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot execute alter, because mutations version was suddenly changed due " "to concurrent alter"); LOG_TRACE(log, "Version conflict when trying to create a mutation node, retrying..."); continue; } else throw Coordination::Exception("Unable to create a mutation znode", rc); } waitMutation(mutation_entry.znode_name, query_context->getSettingsRef().mutations_sync); } void StorageReplicatedMergeTree::waitMutation(const String & znode_name, size_t mutations_sync) const { if (!mutations_sync) return; /// we have to wait auto zookeeper = getZooKeeper(); Strings replicas; if (mutations_sync == 2) /// wait for all replicas { replicas = zookeeper->getChildren(fs::path(zookeeper_path) / "replicas"); /// This replica should be first, to ensure that the mutation will be loaded into memory for (auto it = replicas.begin(); it != replicas.end(); ++it) { if (*it == replica_name) { std::iter_swap(it, replicas.begin()); break; } } } else if (mutations_sync == 1) /// just wait for ourself replicas.push_back(replica_name); waitMutationToFinishOnReplicas(replicas, znode_name); } std::vector StorageReplicatedMergeTree::getMutationsStatus() const { return queue.getMutationsStatus(); } CancellationCode StorageReplicatedMergeTree::killMutation(const String & mutation_id) { assertNotReadonly(); zkutil::ZooKeeperPtr zookeeper = getZooKeeperAndAssertNotReadonly(); LOG_INFO(log, "Killing mutation {}", mutation_id); auto mutation_entry = queue.removeMutation(zookeeper, mutation_id); if (!mutation_entry) return CancellationCode::NotFound; /// After this point no new part mutations will start and part mutations that still exist /// in the queue will be skipped. /// Cancel already running part mutations. for (const auto & pair : mutation_entry->block_numbers) { const String & partition_id = pair.first; Int64 block_number = pair.second; getContext()->getMergeList().cancelPartMutations(getStorageID(), partition_id, block_number); } return CancellationCode::CancelSent; } bool StorageReplicatedMergeTree::hasLightweightDeletedMask() const { return has_lightweight_delete_parts.load(std::memory_order_relaxed); } void StorageReplicatedMergeTree::clearOldPartsAndRemoveFromZK() { auto table_lock = lockForShare( RWLockImpl::NO_QUERY, getSettings()->lock_acquire_timeout_for_background_operations); auto zookeeper = getZooKeeper(); /// Now these parts are in Deleting state. If we fail to remove some of them we must roll them back to Outdated state. /// Otherwise they will not be deleted. DataPartsVector parts = grabOldParts(); if (parts.empty()) return; DataPartsVector parts_to_delete_only_from_filesystem; // Only duplicates DataPartsVector parts_to_delete_completely; // All parts except duplicates DataPartsVector parts_to_retry_deletion; // Parts that should be retried due to network problems DataPartsVector parts_to_remove_from_filesystem; // Parts removed from ZK for (const auto & part : parts) { if (!part->is_duplicate) parts_to_delete_completely.emplace_back(part); else parts_to_delete_only_from_filesystem.emplace_back(part); } parts.clear(); auto delete_parts_from_fs_and_rollback_in_case_of_error = [this] (const DataPartsVector & parts_to_delete, const String & parts_type) { NameSet parts_failed_to_delete; clearPartsFromFilesystem(parts_to_delete, false, &parts_failed_to_delete); DataPartsVector finally_remove_parts; if (!parts_failed_to_delete.empty()) { DataPartsVector rollback_parts; for (const auto & part : parts_to_delete) { if (!parts_failed_to_delete.contains(part->name)) finally_remove_parts.push_back(part); else rollback_parts.push_back(part); } if (!rollback_parts.empty()) rollbackDeletingParts(rollback_parts); } else /// all parts was successfully removed { finally_remove_parts = parts_to_delete; } try { removePartsFinally(finally_remove_parts); LOG_DEBUG(log, "Removed {} {} parts", finally_remove_parts.size(), parts_type); } catch (...) { tryLogCurrentException(log, "Failed to remove some parts from memory, or write info about them into part log"); } }; /// Delete duplicate parts from filesystem if (!parts_to_delete_only_from_filesystem.empty()) { /// It can happen that some error appear during part removal from FS. /// In case of such exception we have to change state of failed parts from Deleting to Outdated. /// Otherwise nobody will try to remove them again (see grabOldParts). delete_parts_from_fs_and_rollback_in_case_of_error(parts_to_delete_only_from_filesystem, "old duplicate"); } /// Delete normal parts from ZooKeeper NameSet part_names_to_retry_deletion; try { Strings part_names_to_delete_completely; for (const auto & part : parts_to_delete_completely) part_names_to_delete_completely.emplace_back(part->name); LOG_DEBUG(log, "Removing {} old parts from ZooKeeper", parts_to_delete_completely.size()); removePartsFromZooKeeper(zookeeper, part_names_to_delete_completely, &part_names_to_retry_deletion); } catch (...) { LOG_ERROR(log, "There is a problem with deleting parts from ZooKeeper: {}", getCurrentExceptionMessage(true)); } /// Part names that were reliably deleted from ZooKeeper should be deleted from filesystem auto num_reliably_deleted_parts = parts_to_delete_completely.size() - part_names_to_retry_deletion.size(); LOG_DEBUG(log, "Removed {} old parts from ZooKeeper. Removing them from filesystem.", num_reliably_deleted_parts); /// Delete normal parts on two sets for (auto & part : parts_to_delete_completely) { if (!part_names_to_retry_deletion.contains(part->name)) parts_to_remove_from_filesystem.emplace_back(part); else parts_to_retry_deletion.emplace_back(part); } /// Will retry deletion if (!parts_to_retry_deletion.empty()) { rollbackDeletingParts(parts_to_retry_deletion); LOG_DEBUG(log, "Will retry deletion of {} parts in the next time", parts_to_retry_deletion.size()); } /// Remove parts from filesystem and finally from data_parts if (!parts_to_remove_from_filesystem.empty()) { /// It can happen that some error appear during part removal from FS. /// In case of such exception we have to change state of failed parts from Deleting to Outdated. /// Otherwise nobody will try to remove them again (see grabOldParts). delete_parts_from_fs_and_rollback_in_case_of_error(parts_to_remove_from_filesystem, "old"); } } void StorageReplicatedMergeTree::removePartsFromZooKeeperWithRetries(PartsToRemoveFromZooKeeper & parts, size_t max_retries) { Strings part_names_to_remove; for (const auto & part : parts) part_names_to_remove.emplace_back(part.getPartName()); return removePartsFromZooKeeperWithRetries(part_names_to_remove, max_retries); } void StorageReplicatedMergeTree::removePartsFromZooKeeperWithRetries(const Strings & part_names, size_t max_retries) { size_t num_tries = 0; bool success = false; while (!success && (max_retries == 0 || num_tries < max_retries)) { try { ++num_tries; success = true; auto zookeeper = getZooKeeper(); Strings exists_paths; exists_paths.reserve(part_names.size()); for (const String & part_name : part_names) { exists_paths.emplace_back(fs::path(replica_path) / "parts" / part_name); } auto exists_results = zookeeper->exists(exists_paths); std::vector> remove_futures; remove_futures.reserve(part_names.size()); for (size_t i = 0; i < part_names.size(); ++i) { Coordination::ExistsResponse exists_resp = exists_results[i]; if (exists_resp.error == Coordination::Error::ZOK) { Coordination::Requests ops; getRemovePartFromZooKeeperOps(part_names[i], ops, exists_resp.stat.numChildren > 0); remove_futures.emplace_back(zookeeper->asyncTryMultiNoThrow(ops)); } } for (auto & future : remove_futures) { auto response = future.get(); if (response.error == Coordination::Error::ZOK || response.error == Coordination::Error::ZNONODE) continue; if (Coordination::isHardwareError(response.error)) { success = false; continue; } throw Coordination::Exception(response.error); } } catch (Coordination::Exception & e) { success = false; if (Coordination::isHardwareError(e.code)) tryLogCurrentException(log, __PRETTY_FUNCTION__); else throw; } if (!success && num_tries < max_retries) std::this_thread::sleep_for(std::chrono::milliseconds(1000)); } if (!success) throw Exception(ErrorCodes::UNFINISHED, "Failed to remove parts from ZooKeeper after {} retries", num_tries); } void StorageReplicatedMergeTree::removePartsFromZooKeeper( zkutil::ZooKeeperPtr & zookeeper, const Strings & part_names, NameSet * parts_should_be_retried) { Strings exists_paths; std::vector> remove_futures; exists_paths.reserve(part_names.size()); remove_futures.reserve(part_names.size()); try { /// Exception can be thrown from loop /// if zk session will be dropped for (const String & part_name : part_names) { exists_paths.emplace_back(fs::path(replica_path) / "parts" / part_name); } auto exists_results = zookeeper->exists(exists_paths); for (size_t i = 0; i < part_names.size(); ++i) { auto exists_resp = exists_results[i]; if (exists_resp.error == Coordination::Error::ZOK) { Coordination::Requests ops; getRemovePartFromZooKeeperOps(part_names[i], ops, exists_resp.stat.numChildren > 0); remove_futures.emplace_back(zookeeper->asyncTryMultiNoThrow(ops)); } else { LOG_DEBUG(log, "There is no part {} in ZooKeeper, it was only in filesystem", part_names[i]); // emplace invalid future so that the total number of futures is the same as part_names.size(); remove_futures.emplace_back(); } } } catch (const Coordination::Exception & e) { if (parts_should_be_retried && Coordination::isHardwareError(e.code)) parts_should_be_retried->insert(part_names.begin(), part_names.end()); throw; } for (size_t i = 0; i < remove_futures.size(); ++i) { auto & future = remove_futures[i]; if (!future.valid()) continue; auto response = future.get(); if (response.error == Coordination::Error::ZOK) continue; else if (response.error == Coordination::Error::ZNONODE) { LOG_DEBUG(log, "There is no part {} in ZooKeeper, it was only in filesystem", part_names[i]); continue; } else if (Coordination::isHardwareError(response.error)) { if (parts_should_be_retried) parts_should_be_retried->insert(part_names[i]); continue; } else LOG_WARNING(log, "Cannot remove part {} from ZooKeeper: {}", part_names[i], Coordination::errorMessage(response.error)); } } void StorageReplicatedMergeTree::clearLockedBlockNumbersInPartition( zkutil::ZooKeeper & zookeeper, const String & partition_id, Int64 min_block_num, Int64 max_block_num) { /// Imagine that some INSERT query has allocated block number 42, but it's still in progress. /// Some DROP PARTITION query gets block number 43 and commits DROP_RANGE all_0_42_999_999. /// And after that INSERT commits GET_PART all_42_42_0. Oops, intersecting parts. /// So we have to either wait for unfinished INSERTs or cancel them. /// It's totally fine to cancel since we are going to remove data anyway. /// We can safely cancel INSERT query by removing its ephemeral block number. /// Usually it's bad idea to remove ephemeral nodes owned by someone else, /// but INSERTs remove such nodes atomically with part commit, so INSERT will fail if node does not exist. fs::path partition_path = fs::path(zookeeper_path) / "block_numbers" / partition_id; Strings queries_in_progress = zookeeper.getChildren(partition_path); if (queries_in_progress.empty()) return; Strings paths_to_get; for (const auto & block : queries_in_progress) { if (!startsWith(block, "block-")) continue; Int64 block_number = parse(block.substr(strlen("block-"))); if (min_block_num <= block_number && block_number <= max_block_num) paths_to_get.push_back(partition_path / block); } auto results = zookeeper.tryGet(paths_to_get); for (size_t i = 0; i < paths_to_get.size(); ++i) { auto & result = results[i]; /// The query already finished if (result.error == Coordination::Error::ZNONODE) continue; /// The query is not an insert (it does not have block_id) if (result.data.ends_with(EphemeralLockInZooKeeper::LEGACY_LOCK_OTHER)) continue; if (result.data.ends_with(EphemeralLockInZooKeeper::LEGACY_LOCK_INSERT)) { /// Remove block number, so insert will fail to commit (it will try to remove this node too) LOG_WARNING(log, "Some query is trying to concurrently insert block {}, will cancel it", paths_to_get[i]); zookeeper.tryRemove(paths_to_get[i]); } else { constexpr const char * old_version_warning = "Ephemeral lock {} (referencing {}) is created by a replica " "that running old version of ClickHouse (< 22.11). Cannot remove it, will wait for this lock to disappear. " "Upgrade remaining hosts in the cluster to address this warning."; constexpr const char * new_version_warning = "Ephemeral lock {} has unexpected content ({}), " "probably it is created by a replica that running newer version of ClickHouse. " "Cannot remove it, will wait for this lock to disappear. Upgrade remaining hosts in the cluster to address this warning."; if (result.data.starts_with(zookeeper_path + EphemeralLockInZooKeeper::LEGACY_LOCK_PREFIX)) LOG_WARNING(log, old_version_warning, paths_to_get[i], result.data); else LOG_WARNING(log, new_version_warning, paths_to_get[i], result.data); Stopwatch time_waiting; const auto & stop_waiting = [this, &time_waiting]() { auto timeout = getContext()->getSettingsRef().lock_acquire_timeout.value.seconds(); return partial_shutdown_called || (timeout < time_waiting.elapsedSeconds()); }; zookeeper.waitForDisappear(paths_to_get[i], stop_waiting); } } } void StorageReplicatedMergeTree::getClearBlocksInPartitionOps( Coordination::Requests & ops, zkutil::ZooKeeper & zookeeper, const String & partition_id, Int64 min_block_num, Int64 max_block_num) { getClearBlocksInPartitionOpsImpl(ops, zookeeper, partition_id, min_block_num, max_block_num, "blocks"); getClearBlocksInPartitionOpsImpl(ops, zookeeper, partition_id, min_block_num, max_block_num, "async_blocks"); } void StorageReplicatedMergeTree::getClearBlocksInPartitionOpsImpl( Coordination::Requests & ops, zkutil::ZooKeeper & zookeeper, const String & partition_id, Int64 min_block_num, Int64 max_block_num, const String & blocks_dir_name) { Strings blocks; if (Coordination::Error::ZOK != zookeeper.tryGetChildren(fs::path(zookeeper_path) / blocks_dir_name, blocks)) throw Exception(ErrorCodes::NOT_FOUND_NODE, "Node {}/{} doesn't exist", zookeeper_path, blocks_dir_name); String partition_prefix = partition_id + "_"; Strings paths_to_get; for (const String & block_id : blocks) if (startsWith(block_id, partition_prefix)) paths_to_get.push_back(fs::path(zookeeper_path) / blocks_dir_name / block_id); auto results = zookeeper.tryGet(paths_to_get); for (size_t i = 0; i < paths_to_get.size(); ++i) { const String & path = paths_to_get[i]; auto & result = results[i]; if (result.error == Coordination::Error::ZNONODE) continue; ReadBufferFromString buf(result.data); const auto part_info = MergeTreePartInfo::tryParsePartName(result.data, format_version); if (!part_info || (min_block_num <= part_info->min_block && part_info->max_block <= max_block_num)) ops.emplace_back(zkutil::makeRemoveRequest(path, -1)); } } void StorageReplicatedMergeTree::clearBlocksInPartition( zkutil::ZooKeeper & zookeeper, const String & partition_id, Int64 min_block_num, Int64 max_block_num) { Coordination::Requests delete_requests; getClearBlocksInPartitionOps(delete_requests, zookeeper, partition_id, min_block_num, max_block_num); Coordination::Responses delete_responses; auto code = zookeeper.tryMulti(delete_requests, delete_responses); if (code != Coordination::Error::ZOK) { for (size_t i = 0; i < delete_requests.size(); ++i) if (delete_responses[i]->error != Coordination::Error::ZOK) LOG_WARNING(log, "Error while deleting ZooKeeper path `{}`: {}, ignoring.", delete_requests[i]->getPath(), Coordination::errorMessage(delete_responses[i]->error)); } LOG_TRACE(log, "Deleted {} deduplication block IDs in partition ID {}", delete_requests.size(), partition_id); } void StorageReplicatedMergeTree::replacePartitionFrom( const StoragePtr & source_table, const ASTPtr & partition, bool replace, ContextPtr query_context) { /// First argument is true, because we possibly will add new data to current table. auto lock1 = lockForShare(query_context->getCurrentQueryId(), query_context->getSettingsRef().lock_acquire_timeout); auto lock2 = source_table->lockForShare(query_context->getCurrentQueryId(), query_context->getSettingsRef().lock_acquire_timeout); auto storage_settings_ptr = getSettings(); auto source_metadata_snapshot = source_table->getInMemoryMetadataPtr(); auto metadata_snapshot = getInMemoryMetadataPtr(); Stopwatch watch; ProfileEventsScope profile_events_scope; MergeTreeData & src_data = checkStructureAndGetMergeTreeData(source_table, source_metadata_snapshot, metadata_snapshot); String partition_id = getPartitionIDFromQuery(partition, query_context); /// NOTE: Some covered parts may be missing in src_all_parts if corresponding log entries are not executed yet. DataPartsVector src_all_parts = src_data.getVisibleDataPartsVectorInPartition(query_context, partition_id); LOG_DEBUG(log, "Cloning {} parts", src_all_parts.size()); static const String TMP_PREFIX = "tmp_replace_from_"; auto zookeeper = getZooKeeper(); /// Retry if alter_partition_version changes for (size_t retry = 0; retry < 1000; ++retry) { DataPartsVector src_parts; MutableDataPartsVector dst_parts; std::vector dst_parts_locks; Strings block_id_paths; Strings part_checksums; std::vector ephemeral_locks; String alter_partition_version_path = zookeeper_path + "/alter_partition_version"; Coordination::Stat alter_partition_version_stat; zookeeper->get(alter_partition_version_path, &alter_partition_version_stat); /// Firstly, generate last block number and compute drop_range /// NOTE: Even if we make ATTACH PARTITION instead of REPLACE PARTITION drop_range will not be empty, it will contain a block. /// So, such case has special meaning, if drop_range contains only one block it means that nothing to drop. /// TODO why not to add normal DROP_RANGE entry to replication queue if `replace` is true? MergeTreePartInfo drop_range; std::optional delimiting_block_lock; bool partition_was_empty = !getFakePartCoveringAllPartsInPartition(partition_id, drop_range, delimiting_block_lock, true); if (replace && partition_was_empty) { /// Nothing to drop, will just attach new parts LOG_INFO(log, "Partition {} was empty, REPLACE PARTITION will work as ATTACH PARTITION FROM", drop_range.partition_id); replace = false; } if (!replace) { /// It's ATTACH PARTITION FROM, not REPLACE PARTITION. We have to reset drop range drop_range = makeDummyDropRangeForMovePartitionOrAttachPartitionFrom(partition_id); } assert(replace == !LogEntry::ReplaceRangeEntry::isMovePartitionOrAttachFrom(drop_range)); String drop_range_fake_part_name = getPartNamePossiblyFake(format_version, drop_range); std::vector hardlinked_files_for_parts; for (const auto & src_part : src_all_parts) { /// We also make some kind of deduplication to avoid duplicated parts in case of ATTACH PARTITION /// Assume that merges in the partition are quite rare /// Save deduplication block ids with special prefix replace_partition if (!canReplacePartition(src_part)) throw Exception(ErrorCodes::LOGICAL_ERROR, "Cannot replace partition '{}' because part '{}" "' has inconsistent granularity with table", partition_id, src_part->name); String hash_hex = src_part->checksums.getTotalChecksumHex(); if (replace) LOG_INFO(log, "Trying to replace {} with hash_hex {}", src_part->name, hash_hex); else LOG_INFO(log, "Trying to attach {} with hash_hex {}", src_part->name, hash_hex); String block_id_path = replace ? "" : (fs::path(zookeeper_path) / "blocks" / (partition_id + "_replace_from_" + hash_hex)); auto lock = allocateBlockNumber(partition_id, zookeeper, block_id_path); if (!lock) { LOG_INFO(log, "Part {} (hash {}) has been already attached", src_part->name, hash_hex); continue; } UInt64 index = lock->getNumber(); MergeTreePartInfo dst_part_info(partition_id, index, index, src_part->info.level); MergeTreeData::HardlinkedFiles hardlinked_files; bool copy_instead_of_hardlink = storage_settings_ptr->allow_remote_fs_zero_copy_replication && src_part->isStoredOnRemoteDiskWithZeroCopySupport(); auto [dst_part, part_lock] = cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, copy_instead_of_hardlink, {}); src_parts.emplace_back(src_part); dst_parts.emplace_back(dst_part); dst_parts_locks.emplace_back(std::move(part_lock)); ephemeral_locks.emplace_back(std::move(*lock)); block_id_paths.emplace_back(block_id_path); part_checksums.emplace_back(hash_hex); hardlinked_files_for_parts.emplace_back(hardlinked_files); } ReplicatedMergeTreeLogEntryData entry; { auto src_table_id = src_data.getStorageID(); entry.type = ReplicatedMergeTreeLogEntryData::REPLACE_RANGE; entry.source_replica = replica_name; entry.create_time = time(nullptr); entry.replace_range_entry = std::make_shared(); auto & entry_replace = *entry.replace_range_entry; entry_replace.drop_range_part_name = drop_range_fake_part_name; entry_replace.from_database = src_table_id.database_name; entry_replace.from_table = src_table_id.table_name; for (const auto & part : src_parts) entry_replace.src_part_names.emplace_back(part->name); for (const auto & part : dst_parts) entry_replace.new_part_names.emplace_back(part->name); for (const String & checksum : part_checksums) entry_replace.part_names_checksums.emplace_back(checksum); entry_replace.columns_version = -1; } if (replace) { /// Cancel concurrent inserts in range clearLockedBlockNumbersInPartition(*zookeeper, drop_range.partition_id, drop_range.max_block, drop_range.max_block); /// Remove deduplication block_ids of replacing parts clearBlocksInPartition(*zookeeper, drop_range.partition_id, drop_range.max_block, drop_range.max_block); } PartsToRemoveFromZooKeeper parts_to_remove; Coordination::Responses op_results; try { Coordination::Requests ops; for (size_t i = 0; i < dst_parts.size(); ++i) { getCommitPartOps(ops, dst_parts[i], block_id_paths[i]); ephemeral_locks[i].getUnlockOp(ops); } if (auto txn = query_context->getZooKeeperMetadataTransaction()) txn->moveOpsTo(ops); delimiting_block_lock->getUnlockOp(ops); /// Check and update version to avoid race with DROP_RANGE ops.emplace_back(zkutil::makeSetRequest(alter_partition_version_path, "", alter_partition_version_stat.version)); /// Just update version, because merges assignment relies on it ops.emplace_back(zkutil::makeSetRequest(fs::path(zookeeper_path) / "log", "", -1)); ops.emplace_back(zkutil::makeCreateRequest(fs::path(zookeeper_path) / "log/log-", entry.toString(), zkutil::CreateMode::PersistentSequential)); Transaction transaction(*this, NO_TRANSACTION_RAW); { auto data_parts_lock = lockParts(); for (auto & part : dst_parts) renameTempPartAndReplaceUnlocked(part, transaction, data_parts_lock); } for (size_t i = 0; i < dst_parts.size(); ++i) lockSharedData(*dst_parts[i], false, hardlinked_files_for_parts[i]); Coordination::Error code = zookeeper->tryMulti(ops, op_results); if (code == Coordination::Error::ZOK) delimiting_block_lock->assumeUnlocked(); else if (code == Coordination::Error::ZBADVERSION) { /// Cannot retry automatically, because some zookeeper ops were lost on the first attempt. Will retry on DDLWorker-level. if (query_context->getZooKeeperMetadataTransaction()) throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot execute alter, because alter partition version was suddenly changed due " "to concurrent alter"); continue; } else zkutil::KeeperMultiException::check(code, ops, op_results); { auto data_parts_lock = lockParts(); transaction.commit(&data_parts_lock); if (replace) parts_to_remove = removePartsInRangeFromWorkingSetAndGetPartsToRemoveFromZooKeeper(NO_TRANSACTION_RAW, drop_range, data_parts_lock); } PartLog::addNewParts(getContext(), PartLog::createPartLogEntries(dst_parts, watch.elapsed(), profile_events_scope.getSnapshot())); } catch (...) { PartLog::addNewParts(getContext(), PartLog::createPartLogEntries(dst_parts, watch.elapsed()), ExecutionStatus::fromCurrentException("", true)); for (const auto & dst_part : dst_parts) unlockSharedData(*dst_part); throw; } String log_znode_path = dynamic_cast(*op_results.back()).path_created; entry.znode_name = log_znode_path.substr(log_znode_path.find_last_of('/') + 1); for (auto & lock : ephemeral_locks) lock.assumeUnlocked(); /// Forcibly remove replaced parts from ZooKeeper removePartsFromZooKeeperWithRetries(parts_to_remove); /// Speedup removing of replaced parts from filesystem parts_to_remove.clear(); cleanup_thread.wakeup(); lock2.reset(); lock1.reset(); waitForLogEntryToBeProcessedIfNecessary(entry, query_context); return; } throw Exception( ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot assign ALTER PARTITION, because another ALTER PARTITION query was concurrently executed"); } void StorageReplicatedMergeTree::movePartitionToTable(const StoragePtr & dest_table, const ASTPtr & partition, ContextPtr query_context) { auto lock1 = lockForShare(query_context->getCurrentQueryId(), query_context->getSettingsRef().lock_acquire_timeout); auto lock2 = dest_table->lockForShare(query_context->getCurrentQueryId(), query_context->getSettingsRef().lock_acquire_timeout); auto storage_settings_ptr = getSettings(); auto dest_table_storage = std::dynamic_pointer_cast(dest_table); if (!dest_table_storage) throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Table {} supports movePartitionToTable only for ReplicatedMergeTree family of table engines. " "Got {}", getStorageID().getNameForLogs(), dest_table->getName()); if (dest_table_storage->getStoragePolicy() != this->getStoragePolicy()) throw Exception(ErrorCodes::UNKNOWN_POLICY, "Destination table {} should have the same storage policy of source table {}. {}: {}, {}: {}", dest_table_storage->getStorageID().getNameForLogs(), getStorageID().getNameForLogs(), getStorageID().getNameForLogs(), this->getStoragePolicy()->getName(), getStorageID().getNameForLogs(), dest_table_storage->getStoragePolicy()->getName()); auto dest_metadata_snapshot = dest_table->getInMemoryMetadataPtr(); auto metadata_snapshot = getInMemoryMetadataPtr(); Stopwatch watch; ProfileEventsScope profile_events_scope; MergeTreeData & src_data = dest_table_storage->checkStructureAndGetMergeTreeData(*this, metadata_snapshot, dest_metadata_snapshot); auto src_data_id = src_data.getStorageID(); String partition_id = getPartitionIDFromQuery(partition, query_context); /// A range for log entry to remove parts from the source table (myself). auto zookeeper = getZooKeeper(); /// Retry if alter_partition_version changes for (size_t retry = 0; retry < 1000; ++retry) { String alter_partition_version_path = zookeeper_path + "/alter_partition_version"; Coordination::Stat alter_partition_version_stat; zookeeper->get(alter_partition_version_path, &alter_partition_version_stat); MergeTreePartInfo drop_range; std::optional delimiting_block_lock; getFakePartCoveringAllPartsInPartition(partition_id, drop_range, delimiting_block_lock, true); String drop_range_fake_part_name = getPartNamePossiblyFake(format_version, drop_range); DataPartPtr covering_part; DataPartsVector src_all_parts; { /// NOTE: Some covered parts may be missing in src_all_parts if corresponding log entries are not executed yet. auto parts_lock = src_data.lockParts(); src_all_parts = src_data.getActivePartsToReplace(drop_range, drop_range_fake_part_name, covering_part, parts_lock); } if (covering_part) throw Exception(ErrorCodes::LOGICAL_ERROR, "Got part {} covering drop range {}, it's a bug", covering_part->name, drop_range_fake_part_name); /// After allocating block number for drop_range we must ensure that it does not intersect block numbers /// allocated by concurrent REPLACE query. /// We could check it in multi-request atomically with creation of DROP_RANGE entry in source table log, /// but it's better to check it here and fail as early as possible (before we have done something to destination table). Coordination::Error version_check_code = zookeeper->trySet(alter_partition_version_path, "", alter_partition_version_stat.version); if (version_check_code != Coordination::Error::ZOK) throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot DROP PARTITION in {} after copying partition to {}, " "because another ALTER PARTITION query was concurrently executed", getStorageID().getFullTableName(), dest_table_storage->getStorageID().getFullTableName()); DataPartsVector src_parts; MutableDataPartsVector dst_parts; Strings block_id_paths; Strings part_checksums; std::vector ephemeral_locks; LOG_DEBUG(log, "Cloning {} parts", src_all_parts.size()); static const String TMP_PREFIX = "tmp_move_from_"; /// Clone parts into destination table. String dest_alter_partition_version_path = dest_table_storage->zookeeper_path + "/alter_partition_version"; Coordination::Stat dest_alter_partition_version_stat; zookeeper->get(dest_alter_partition_version_path, &dest_alter_partition_version_stat); std::vector hardlinked_files_for_parts; std::vector temporary_parts_locks; for (const auto & src_part : src_all_parts) { if (!dest_table_storage->canReplacePartition(src_part)) throw Exception(ErrorCodes::LOGICAL_ERROR, "Cannot move partition '{}' because part '{}" "' has inconsistent granularity with table", partition_id, src_part->name); String hash_hex = src_part->checksums.getTotalChecksumHex(); String block_id_path; auto lock = dest_table_storage->allocateBlockNumber(partition_id, zookeeper, block_id_path); if (!lock) { LOG_INFO(log, "Part {} (hash {}) has been already attached", src_part->name, hash_hex); continue; } UInt64 index = lock->getNumber(); MergeTreePartInfo dst_part_info(partition_id, index, index, src_part->info.level); MergeTreeData::HardlinkedFiles hardlinked_files; bool copy_instead_of_hardlink = storage_settings_ptr->allow_remote_fs_zero_copy_replication && src_part->isStoredOnRemoteDiskWithZeroCopySupport(); auto [dst_part, dst_part_lock] = dest_table_storage->cloneAndLoadDataPartOnSameDisk(src_part, TMP_PREFIX, dst_part_info, dest_metadata_snapshot, NO_TRANSACTION_PTR, &hardlinked_files, copy_instead_of_hardlink, {}); src_parts.emplace_back(src_part); dst_parts.emplace_back(dst_part); temporary_parts_locks.emplace_back(std::move(dst_part_lock)); ephemeral_locks.emplace_back(std::move(*lock)); block_id_paths.emplace_back(block_id_path); part_checksums.emplace_back(hash_hex); hardlinked_files_for_parts.emplace_back(hardlinked_files); } ReplicatedMergeTreeLogEntryData entry_delete; { entry_delete.type = LogEntry::DROP_RANGE; entry_delete.source_replica = replica_name; entry_delete.new_part_name = drop_range_fake_part_name; entry_delete.detach = false; entry_delete.create_time = time(nullptr); } ReplicatedMergeTreeLogEntryData entry; { MergeTreePartInfo drop_range_dest = makeDummyDropRangeForMovePartitionOrAttachPartitionFrom(partition_id); entry.type = ReplicatedMergeTreeLogEntryData::REPLACE_RANGE; entry.source_replica = dest_table_storage->replica_name; entry.create_time = time(nullptr); entry.replace_range_entry = std::make_shared(); auto & entry_replace = *entry.replace_range_entry; entry_replace.drop_range_part_name = getPartNamePossiblyFake(format_version, drop_range_dest); entry_replace.from_database = src_data_id.database_name; entry_replace.from_table = src_data_id.table_name; for (const auto & part : src_parts) entry_replace.src_part_names.emplace_back(part->name); for (const auto & part : dst_parts) entry_replace.new_part_names.emplace_back(part->name); for (const String & checksum : part_checksums) entry_replace.part_names_checksums.emplace_back(checksum); entry_replace.columns_version = -1; } /// Cancel concurrent inserts in range clearLockedBlockNumbersInPartition(*zookeeper, drop_range.partition_id, drop_range.max_block, drop_range.max_block); clearBlocksInPartition(*zookeeper, drop_range.partition_id, drop_range.max_block, drop_range.max_block); PartsToRemoveFromZooKeeper parts_to_remove; Coordination::Responses op_results; try { Coordination::Requests ops; for (size_t i = 0; i < dst_parts.size(); ++i) { dest_table_storage->getCommitPartOps(ops, dst_parts[i], block_id_paths[i]); ephemeral_locks[i].getUnlockOp(ops); } /// Check and update version to avoid race with DROP_RANGE ops.emplace_back(zkutil::makeSetRequest(dest_alter_partition_version_path, "", dest_alter_partition_version_stat.version)); /// Just update version, because merges assignment relies on it ops.emplace_back(zkutil::makeSetRequest(fs::path(dest_table_storage->zookeeper_path) / "log", "", -1)); ops.emplace_back(zkutil::makeCreateRequest(fs::path(dest_table_storage->zookeeper_path) / "log/log-", entry.toString(), zkutil::CreateMode::PersistentSequential)); { Transaction transaction(*dest_table_storage, NO_TRANSACTION_RAW); auto src_data_parts_lock = lockParts(); auto dest_data_parts_lock = dest_table_storage->lockParts(); for (auto & part : dst_parts) dest_table_storage->renameTempPartAndReplaceUnlocked(part, transaction, dest_data_parts_lock); for (size_t i = 0; i < dst_parts.size(); ++i) dest_table_storage->lockSharedData(*dst_parts[i], false, hardlinked_files_for_parts[i]); Coordination::Error code = zookeeper->tryMulti(ops, op_results); if (code == Coordination::Error::ZBADVERSION) continue; else zkutil::KeeperMultiException::check(code, ops, op_results); parts_to_remove = removePartsInRangeFromWorkingSetAndGetPartsToRemoveFromZooKeeper(NO_TRANSACTION_RAW, drop_range, src_data_parts_lock); transaction.commit(&src_data_parts_lock); } PartLog::addNewParts(getContext(), PartLog::createPartLogEntries(dst_parts, watch.elapsed(), profile_events_scope.getSnapshot())); } catch (...) { PartLog::addNewParts(getContext(), PartLog::createPartLogEntries(dst_parts, watch.elapsed()), ExecutionStatus::fromCurrentException("", true)); for (const auto & dst_part : dst_parts) dest_table_storage->unlockSharedData(*dst_part); throw; } String log_znode_path = dynamic_cast(*op_results.back()).path_created; entry.znode_name = log_znode_path.substr(log_znode_path.find_last_of('/') + 1); for (auto & lock : ephemeral_locks) lock.assumeUnlocked(); removePartsFromZooKeeperWithRetries(parts_to_remove); parts_to_remove.clear(); cleanup_thread.wakeup(); lock2.reset(); dest_table_storage->waitForLogEntryToBeProcessedIfNecessary(entry, query_context); /// Create DROP_RANGE for the source table Coordination::Requests ops_src; ops_src.emplace_back(zkutil::makeCreateRequest( fs::path(zookeeper_path) / "log/log-", entry_delete.toString(), zkutil::CreateMode::PersistentSequential)); /// Just update version, because merges assignment relies on it ops_src.emplace_back(zkutil::makeSetRequest(fs::path(zookeeper_path) / "log", "", -1)); delimiting_block_lock->getUnlockOp(ops_src); op_results = zookeeper->multi(ops_src); log_znode_path = dynamic_cast(*op_results.front()).path_created; entry_delete.znode_name = log_znode_path.substr(log_znode_path.find_last_of('/') + 1); lock1.reset(); waitForLogEntryToBeProcessedIfNecessary(entry_delete, query_context); /// Cleaning possibly stored information about parts from /quorum/last_part node in ZooKeeper. cleanLastPartNode(partition_id); return; } throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot assign ALTER PARTITION, because another ALTER PARTITION query was concurrently executed"); } void StorageReplicatedMergeTree::movePartitionToShard( const ASTPtr & partition, bool move_part, const String & to, ContextPtr /*query_context*/) { /// This is a lightweight operation that only optimistically checks if it could succeed and queues tasks. if (!move_part) throw Exception(ErrorCodes::NOT_IMPLEMENTED, "MOVE PARTITION TO SHARD is not supported, use MOVE PART instead"); if (zkutil::normalizeZooKeeperPath(zookeeper_path, /* check_starts_with_slash */ true) == zkutil::normalizeZooKeeperPath(to, /* check_starts_with_slash */ true)) throw Exception(ErrorCodes::BAD_ARGUMENTS, "Source and destination are the same"); auto zookeeper = getZooKeeperAndAssertNotReadonly(); String part_name = partition->as().value.safeGet(); auto part_info = MergeTreePartInfo::fromPartName(part_name, format_version); auto part = getPartIfExists(part_info, {MergeTreeDataPartState::Active}); if (!part) throw Exception(ErrorCodes::NO_SUCH_DATA_PART, "Part {} not found locally", part_name); if (part->uuid == UUIDHelpers::Nil) throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Part {} does not have an uuid assigned and it can't be moved between shards", part_name); ReplicatedMergeTreeMergePredicate merge_pred = queue.getMergePredicate(zookeeper, PartitionIdsHint{part_info.partition_id}); /// The following block is pretty much copy & paste from StorageReplicatedMergeTree::dropPart to avoid conflicts while this is WIP. /// Extract it to a common method and re-use it before merging. { if (partIsLastQuorumPart(part->info)) { throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Part {} is last inserted part with quorum in partition. Would not be able to drop", part_name); } /// canMergeSinglePart is overlapping with dropPart, let's try to use the same code. String out_reason; if (!merge_pred.canMergeSinglePart(part, &out_reason)) throw Exception(ErrorCodes::PART_IS_TEMPORARILY_LOCKED, "Part is busy, reason: {}", out_reason); } { /// Optimistic check that for compatible destination table structure. checkTableStructure(to, getInMemoryMetadataPtr()); } PinnedPartUUIDs src_pins; PinnedPartUUIDs dst_pins; { String s = zookeeper->get(zookeeper_path + "/pinned_part_uuids", &src_pins.stat); src_pins.fromString(s); } { String s = zookeeper->get(to + "/pinned_part_uuids", &dst_pins.stat); dst_pins.fromString(s); } if (src_pins.part_uuids.contains(part->uuid) || dst_pins.part_uuids.contains(part->uuid)) throw Exception(ErrorCodes::PART_IS_TEMPORARILY_LOCKED, "Part {} has it's uuid ({}) already pinned.", part_name, part->uuid); src_pins.part_uuids.insert(part->uuid); dst_pins.part_uuids.insert(part->uuid); PartMovesBetweenShardsOrchestrator::Entry part_move_entry; part_move_entry.state = PartMovesBetweenShardsOrchestrator::EntryState::SYNC_SOURCE; part_move_entry.create_time = std::time(nullptr); part_move_entry.update_time = part_move_entry.create_time; part_move_entry.task_uuid = UUIDHelpers::generateV4(); part_move_entry.part_name = part->name; part_move_entry.part_uuid = part->uuid; part_move_entry.to_shard = to; Coordination::Requests ops; ops.emplace_back(zkutil::makeCheckRequest(zookeeper_path + "/log", merge_pred.getVersion())); /// Make sure no new events were added to the log. ops.emplace_back(zkutil::makeSetRequest(zookeeper_path + "/pinned_part_uuids", src_pins.toString(), src_pins.stat.version)); ops.emplace_back(zkutil::makeSetRequest(to + "/pinned_part_uuids", dst_pins.toString(), dst_pins.stat.version)); ops.emplace_back(zkutil::makeCreateRequest( part_moves_between_shards_orchestrator.entries_znode_path + "/task-", part_move_entry.toString(), zkutil::CreateMode::PersistentSequential)); Coordination::Responses responses; Coordination::Error rc = zookeeper->tryMulti(ops, responses); zkutil::KeeperMultiException::check(rc, ops, responses); String task_znode_path = dynamic_cast(*responses.back()).path_created; LOG_DEBUG(log, "Created task for part movement between shards at {}", task_znode_path); /// TODO(nv): Nice to have support for `alter_sync`. /// For now use the system.part_moves_between_shards table for status. } CancellationCode StorageReplicatedMergeTree::killPartMoveToShard(const UUID & task_uuid) { return part_moves_between_shards_orchestrator.killPartMoveToShard(task_uuid); } void StorageReplicatedMergeTree::getCommitPartOps( Coordination::Requests & ops, const DataPartPtr & part, const String & block_id_path) const { if (block_id_path.empty()) return getCommitPartOps(ops, part, std::vector()); else return getCommitPartOps(ops, part, std::vector({block_id_path})); } void StorageReplicatedMergeTree::getCommitPartOps( Coordination::Requests & ops, const DataPartPtr & part, const std::vector & block_id_paths) const { const String & part_name = part->name; const auto storage_settings_ptr = getSettings(); for (const String & block_id_path : block_id_paths) { /// Make final duplicate check and commit block_id ops.emplace_back( zkutil::makeCreateRequest( block_id_path, part_name, /// We will be able to know original part number for duplicate blocks, if we want. zkutil::CreateMode::Persistent)); } /// Information about the part, in the replica if (storage_settings_ptr->use_minimalistic_part_header_in_zookeeper) { ops.emplace_back(zkutil::makeCreateRequest( fs::path(replica_path) / "parts" / part->name, ReplicatedMergeTreePartHeader::fromColumnsAndChecksums(part->getColumns(), part->checksums).toString(), zkutil::CreateMode::Persistent)); } else { ops.emplace_back(zkutil::makeCreateRequest( fs::path(replica_path) / "parts" / part->name, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest( fs::path(replica_path) / "parts" / part->name / "columns", part->getColumns().toString(), zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeCreateRequest( fs::path(replica_path) / "parts" / part->name / "checksums", getChecksumsForZooKeeper(part->checksums), zkutil::CreateMode::Persistent)); } } ReplicatedMergeTreeAddress StorageReplicatedMergeTree::getReplicatedMergeTreeAddress() const { auto host_port = getContext()->getInterserverIOAddress(); auto table_id = getStorageID(); ReplicatedMergeTreeAddress res; res.host = host_port.first; res.replication_port = host_port.second; res.queries_port = getContext()->getTCPPort(); res.database = table_id.database_name; res.table = table_id.table_name; res.scheme = getContext()->getInterserverScheme(); return res; } ActionLock StorageReplicatedMergeTree::getActionLock(StorageActionBlockType action_type) { if (action_type == ActionLocks::PartsMerge) return merger_mutator.merges_blocker.cancel(); if (action_type == ActionLocks::PartsTTLMerge) return merger_mutator.ttl_merges_blocker.cancel(); if (action_type == ActionLocks::PartsFetch) return fetcher.blocker.cancel(); if (action_type == ActionLocks::PartsSend) { auto data_parts_exchange_ptr = std::atomic_load(&data_parts_exchange_endpoint); return data_parts_exchange_ptr ? data_parts_exchange_ptr->blocker.cancel() : ActionLock(); } if (action_type == ActionLocks::ReplicationQueue) return queue.actions_blocker.cancel(); if (action_type == ActionLocks::PartsMove) return parts_mover.moves_blocker.cancel(); return {}; } void StorageReplicatedMergeTree::onActionLockRemove(StorageActionBlockType action_type) { if (action_type == ActionLocks::PartsMerge || action_type == ActionLocks::PartsTTLMerge || action_type == ActionLocks::PartsFetch || action_type == ActionLocks::PartsSend || action_type == ActionLocks::ReplicationQueue) background_operations_assignee.trigger(); else if (action_type == ActionLocks::PartsMove) background_moves_assignee.trigger(); } bool StorageReplicatedMergeTree::waitForProcessingQueue(UInt64 max_wait_milliseconds) { Stopwatch watch; /// Let's fetch new log entries firstly queue.pullLogsToQueue(getZooKeeperAndAssertNotReadonly(), {}, ReplicatedMergeTreeQueue::SYNC); /// This is significant, because the execution of this task could be delayed at BackgroundPool. /// And we force it to be executed. background_operations_assignee.trigger(); std::unordered_set wait_for_ids; bool set_ids_to_wait = true; Poco::Event target_entry_event; auto callback = [&target_entry_event, &wait_for_ids, &set_ids_to_wait](size_t new_queue_size, std::unordered_set log_entry_ids, std::optional removed_log_entry_id) { if (set_ids_to_wait) { wait_for_ids = log_entry_ids; set_ids_to_wait = false; } if (removed_log_entry_id.has_value()) wait_for_ids.erase(removed_log_entry_id.value()); if (wait_for_ids.empty() || new_queue_size == 0) target_entry_event.set(); }; const auto handler = queue.addSubscriber(std::move(callback)); while (!target_entry_event.tryWait(50)) { if (max_wait_milliseconds && watch.elapsedMilliseconds() > max_wait_milliseconds) return false; if (partial_shutdown_called) throw Exception(ErrorCodes::ABORTED, "Shutdown is called for table"); } return true; } bool StorageReplicatedMergeTree::dropPartImpl( zkutil::ZooKeeperPtr & zookeeper, String part_name, LogEntry & entry, bool detach, bool throw_if_noop) { LOG_TRACE(log, "Will try to insert a log entry to DROP_PART for part {}", part_name); auto part_info = MergeTreePartInfo::fromPartName(part_name, format_version); while (true) { ReplicatedMergeTreeMergePredicate merge_pred = queue.getMergePredicate(zookeeper, PartitionIdsHint{part_info.partition_id}); auto part = getPartIfExists(part_info, {MergeTreeDataPartState::Active}); if (!part) { if (throw_if_noop) throw Exception(ErrorCodes::NO_SUCH_DATA_PART, "Part {} not found locally, won't try to drop it.", part_name); return false; } if (merge_pred.isGoingToBeDropped(part->info)) { if (throw_if_noop) throw Exception(ErrorCodes::PART_IS_TEMPORARILY_LOCKED, "Already has DROP RANGE for part {} in queue.", part_name); return false; } /// There isn't a lot we can do otherwise. Can't cancel merges because it is possible that a replica already /// finished the merge. String out_reason; if (!merge_pred.canMergeSinglePart(part, &out_reason)) { if (throw_if_noop) throw Exception::createDeprecated(out_reason, ErrorCodes::PART_IS_TEMPORARILY_LOCKED); return false; } if (merge_pred.partParticipatesInReplaceRange(part, &out_reason)) { if (throw_if_noop) throw Exception::createDeprecated(out_reason, ErrorCodes::PART_IS_TEMPORARILY_LOCKED); return false; } if (partIsLastQuorumPart(part->info)) { if (throw_if_noop) throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Part {} is last inserted part with quorum in partition. Cannot drop", part_name); return false; } if (partIsInsertingWithParallelQuorum(part->info)) { if (throw_if_noop) throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Part {} is inserting with parallel quorum. Cannot drop", part_name); return false; } Coordination::Requests ops; /// NOTE Don't need to remove block numbers too, because no in-progress inserts in the range are possible getClearBlocksInPartitionOps(ops, *zookeeper, part_info.partition_id, part_info.min_block, part_info.max_block); size_t clear_block_ops_size = ops.size(); /// If `part_name` is result of a recent merge and source parts are still available then /// DROP_PART with detach will move this part together with source parts to `detached/` dir. entry.type = LogEntry::DROP_PART; entry.source_replica = replica_name; /// We don't set fake drop level (999999999) for the single part drop range. /// First of all we don't guarantee anything other than the part will not be /// active after DROP_PART, but covering part (without data of dropped part) can exist. /// If we add part with 9999999 level than we can break invariant in virtual_parts of /// the queue. entry.new_part_name = getPartNamePossiblyFake(format_version, part->info); entry.detach = detach; entry.create_time = time(nullptr); ops.emplace_back(zkutil::makeCheckRequest(fs::path(zookeeper_path) / "log", merge_pred.getVersion())); /// Make sure no new events were added to the log. ops.emplace_back(zkutil::makeCreateRequest(fs::path(zookeeper_path) / "log/log-", entry.toString(), zkutil::CreateMode::PersistentSequential)); /// Just update version, because merges assignment relies on it ops.emplace_back(zkutil::makeSetRequest(fs::path(zookeeper_path) / "log", "", -1)); Coordination::Responses responses; Coordination::Error rc = zookeeper->tryMulti(ops, responses); if (rc == Coordination::Error::ZBADVERSION) { LOG_TRACE(log, "A new log entry appeared while trying to commit DROP RANGE. Retry."); continue; } else if (rc == Coordination::Error::ZNONODE) { LOG_TRACE(log, "Other replica already removing same part {} or part deduplication node was removed by background thread. Retry.", part_name); continue; } else zkutil::KeeperMultiException::check(rc, ops, responses); String log_znode_path = dynamic_cast(*responses[clear_block_ops_size + 1]).path_created; entry.znode_name = log_znode_path.substr(log_znode_path.find_last_of('/') + 1); LOG_TRACE(log, "DROP RANGE for part {} inserted with znode name {}", part_name, entry.znode_name); return true; } } bool StorageReplicatedMergeTree::addOpsToDropAllPartsInPartition( zkutil::ZooKeeper & zookeeper, const String & partition_id, bool detach, Coordination::Requests & ops, std::vector & entries, std::vector & delimiting_block_locks, std::vector & log_entry_ops_idx) { MergeTreePartInfo drop_range_info; /// It would prevent other replicas from assigning merges which intersect locked block number. std::optional delimiting_block_lock; if (!getFakePartCoveringAllPartsInPartition(partition_id, drop_range_info, delimiting_block_lock)) { LOG_INFO(log, "Will not drop partition {}, it is empty.", partition_id); return false; } /// Cancel concurrent inserts in range clearLockedBlockNumbersInPartition(zookeeper, partition_id, drop_range_info.min_block, drop_range_info.max_block); clearBlocksInPartition(zookeeper, partition_id, drop_range_info.min_block, drop_range_info.max_block); String drop_range_fake_part_name = getPartNamePossiblyFake(format_version, drop_range_info); LOG_DEBUG(log, "Disabled merges covered by range {}", drop_range_fake_part_name); /// Finally, having achieved the necessary invariants, you can put an entry in the log. auto entry = std::make_shared(); entry->type = LogEntry::DROP_RANGE; entry->source_replica = replica_name; entry->new_part_name = drop_range_fake_part_name; entry->detach = detach; entry->create_time = time(nullptr); log_entry_ops_idx.push_back(ops.size()); ops.emplace_back(zkutil::makeCreateRequest(fs::path(zookeeper_path) / "log/log-", entry->toString(), zkutil::CreateMode::PersistentSequential)); delimiting_block_lock->getUnlockOp(ops); delimiting_block_locks.push_back(std::move(*delimiting_block_lock)); entries.push_back(std::move(entry)); return true; } void StorageReplicatedMergeTree::dropAllPartsInPartitions( zkutil::ZooKeeper & zookeeper, const Strings & partition_ids, std::vector & entries, ContextPtr query_context, bool detach) { entries.reserve(partition_ids.size()); /// Retry if alter_partition_version changes for (size_t retry = 0; retry < 1000; ++retry) { entries.clear(); String alter_partition_version_path = zookeeper_path + "/alter_partition_version"; Coordination::Stat alter_partition_version_stat; zookeeper.get(alter_partition_version_path, &alter_partition_version_stat); Coordination::Requests ops; std::vector delimiting_block_locks; std::vector log_entry_ops_idx; ops.reserve(partition_ids.size() * 2); delimiting_block_locks.reserve(partition_ids.size()); log_entry_ops_idx.reserve(partition_ids.size()); for (const auto & partition_id : partition_ids) addOpsToDropAllPartsInPartition(zookeeper, partition_id, detach, ops, entries, delimiting_block_locks, log_entry_ops_idx); /// Check and update version to avoid race with REPLACE_RANGE. /// Otherwise new parts covered by drop_range_info may appear after execution of current DROP_RANGE entry /// as a result of execution of concurrently created REPLACE_RANGE entry. ops.emplace_back(zkutil::makeSetRequest(alter_partition_version_path, "", alter_partition_version_stat.version)); /// Just update version, because merges assignment relies on it ops.emplace_back(zkutil::makeSetRequest(fs::path(zookeeper_path) / "log", "", -1)); if (auto txn = query_context->getZooKeeperMetadataTransaction()) txn->moveOpsTo(ops); Coordination::Responses responses; Coordination::Error code = zookeeper.tryMulti(ops, responses); if (code == Coordination::Error::ZOK) { for (auto & lock : delimiting_block_locks) lock.assumeUnlocked(); } else if (code == Coordination::Error::ZBADVERSION) { /// Cannot retry automatically, because some zookeeper ops were lost on the first attempt. Will retry on DDLWorker-level. if (query_context->getZooKeeperMetadataTransaction()) throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot execute alter, because alter partition version was suddenly changed due " "to concurrent alter"); continue; } else zkutil::KeeperMultiException::check(code, ops, responses); assert(entries.size() == log_entry_ops_idx.size()); for (size_t i = 0; i < entries.size(); ++i) { String log_znode_path = dynamic_cast(*responses[log_entry_ops_idx[i]]).path_created; entries[i]->znode_name = log_znode_path.substr(log_znode_path.find_last_of('/') + 1); auto drop_range_info = MergeTreePartInfo::fromPartName(entries[i]->new_part_name, format_version); getContext()->getMergeList().cancelInPartition(getStorageID(), drop_range_info.partition_id, drop_range_info.max_block); } return; } throw Exception(ErrorCodes::CANNOT_ASSIGN_ALTER, "Cannot assign ALTER PARTITION because another ALTER PARTITION query was concurrently executed"); } StorageReplicatedMergeTree::LogEntryPtr StorageReplicatedMergeTree::dropAllPartsInPartition( zkutil::ZooKeeper & zookeeper, const String & partition_id, ContextPtr query_context, bool detach) { Strings partition_ids = {partition_id}; std::vector entries; dropAllPartsInPartitions(zookeeper, partition_ids, entries, query_context, detach); if (entries.empty()) return {}; return entries[0]; } void StorageReplicatedMergeTree::enqueuePartForCheck(const String & part_name, time_t delay_to_check_seconds) { MergeTreePartInfo covering_drop_range; /// NOTE This check is just an optimization, it's not reliable because drop entry could be removed concurrently. /// See also ReplicatedMergeTreePartCheckThread::cancelRemovedPartsCheck if (queue.isGoingToBeDropped(MergeTreePartInfo::fromPartName(part_name, format_version), &covering_drop_range)) { LOG_WARNING(log, "Do not enqueue part {} for check because it's covered by drop range {} and going to be removed", part_name, covering_drop_range.getPartNameForLogs()); return; } part_check_thread.enqueuePart(part_name, delay_to_check_seconds); } CheckResults StorageReplicatedMergeTree::checkData(const ASTPtr & query, ContextPtr local_context) { CheckResults results; DataPartsVector data_parts; if (const auto & check_query = query->as(); check_query.partition) { String partition_id = getPartitionIDFromQuery(check_query.partition, local_context); data_parts = getVisibleDataPartsVectorInPartition(local_context, partition_id); } else data_parts = getVisibleDataPartsVector(local_context); { auto part_check_lock = part_check_thread.pausePartsCheck(); for (auto & part : data_parts) { try { results.push_back(part_check_thread.checkPart(part->name)); } catch (const Exception & ex) { tryLogCurrentException(log, __PRETTY_FUNCTION__); results.emplace_back(part->name, false, "Check of part finished with error: '" + ex.message() + "'"); } } } return results; } bool StorageReplicatedMergeTree::canUseZeroCopyReplication() const { auto settings_ptr = getSettings(); if (!settings_ptr->allow_remote_fs_zero_copy_replication) return false; auto disks = getStoragePolicy()->getDisks(); for (const auto & disk : disks) { if (disk->supportZeroCopyReplication()) return true; } return false; } void StorageReplicatedMergeTree::checkBrokenDisks() { auto disks = getStoragePolicy()->getDisks(); std::unique_ptr parts; for (auto disk_it = disks.rbegin(); disk_it != disks.rend(); ++disk_it) { auto disk_ptr = *disk_it; if (disk_ptr->isBroken()) { { std::lock_guard lock(last_broken_disks_mutex); if (!last_broken_disks.insert(disk_ptr->getName()).second) continue; } LOG_INFO(log, "Scanning parts to recover on broken disk {} with path {}", disk_ptr->getName(), disk_ptr->getPath()); if (!parts) parts = std::make_unique(getDataPartsVectorForInternalUsage()); for (auto & part : *parts) { if (part->getDataPartStorage().getDiskName() == disk_ptr->getName()) broken_part_callback(part->name); } continue; } else { { std::lock_guard lock(last_broken_disks_mutex); if (last_broken_disks.erase(disk_ptr->getName()) > 0) LOG_INFO( log, "Disk {} with path {} is recovered. Exclude it from last_broken_disks", disk_ptr->getName(), disk_ptr->getPath()); } } } } bool StorageReplicatedMergeTree::canUseAdaptiveGranularity() const { const auto storage_settings_ptr = getSettings(); return storage_settings_ptr->index_granularity_bytes != 0 && (storage_settings_ptr->enable_mixed_granularity_parts || (!has_non_adaptive_index_granularity_parts && !other_replicas_fixed_granularity)); } MutationCommands StorageReplicatedMergeTree::getFirstAlterMutationCommandsForPart(const DataPartPtr & part) const { return queue.getFirstAlterMutationCommandsForPart(part); } void StorageReplicatedMergeTree::startBackgroundMovesIfNeeded() { if (areBackgroundMovesNeeded()) background_moves_assignee.start(); } std::unique_ptr StorageReplicatedMergeTree::getDefaultSettings() const { return std::make_unique(getContext()->getReplicatedMergeTreeSettings()); } String StorageReplicatedMergeTree::getTableSharedID() const { std::lock_guard lock(table_shared_id_mutex); /// If we has metadata or, we don't know about metadata -- try to create shared ID /// Otherwise table is already dropped, doesn't make sense to do anything with shared ID if (has_metadata_in_zookeeper.value_or(true)) { /// Can happen if table was partially initialized before drop by DatabaseCatalog if (table_shared_id == UUIDHelpers::Nil) createTableSharedID(); } else { return toString(UUIDHelpers::Nil); } return toString(table_shared_id); } void StorageReplicatedMergeTree::createTableSharedID() const { LOG_DEBUG(log, "Creating shared ID for table {}", getStorageID().getNameForLogs()); // can be set by the call to getTableSharedID if (table_shared_id != UUIDHelpers::Nil) { LOG_INFO(log, "Shared ID already set to {}", table_shared_id); return; } /// We may call getTableSharedID when table is shut down. If exception happen, restarting thread will be already turned /// off and nobody will reconnect our zookeeper connection. In this case we use zookeeper connection from /// context. ZooKeeperPtr zookeeper; if (shutdown_called.load()) zookeeper = getZooKeeperIfTableShutDown(); else zookeeper = getZooKeeper(); String zookeeper_table_id_path = fs::path(zookeeper_path) / "table_shared_id"; String id; if (!zookeeper->tryGet(zookeeper_table_id_path, id)) { LOG_DEBUG(log, "Shared ID for table {} doesn't exist in ZooKeeper on path {}", getStorageID().getNameForLogs(), zookeeper_table_id_path); UUID table_id_candidate; auto local_storage_id = getStorageID(); if (local_storage_id.uuid != UUIDHelpers::Nil) table_id_candidate = local_storage_id.uuid; else table_id_candidate = UUIDHelpers::generateV4(); id = toString(table_id_candidate); LOG_DEBUG(log, "Got candidate ID {}, will try to create it in ZooKeeper on path {}", id, zookeeper_table_id_path); auto code = zookeeper->tryCreate(zookeeper_table_id_path, id, zkutil::CreateMode::Persistent); if (code == Coordination::Error::ZNODEEXISTS) { /// Other replica create node early id = zookeeper->get(zookeeper_table_id_path); LOG_DEBUG(log, "Shared ID on path {} concurrently created, will set ID {}", zookeeper_table_id_path, id); } else if (code == Coordination::Error::ZNONODE) /// table completely dropped, we can choose any id we want { id = toString(UUIDHelpers::Nil); LOG_DEBUG(log, "Table was completely drop, we can use anything as ID (will use {})", id); } else if (code != Coordination::Error::ZOK) { throw zkutil::KeeperException(code, zookeeper_table_id_path); } } LOG_DEBUG(log, "Initializing table shared ID with {}", id); table_shared_id = parseFromString(id); } std::optional StorageReplicatedMergeTree::tryGetTableSharedIDFromCreateQuery(const IAST & create_query, const ContextPtr & global_context) { auto zk_path = tryExtractZkPathFromCreateQuery(create_query, global_context); if (!zk_path) return {}; String zk_name = zkutil::extractZooKeeperName(*zk_path); zk_path = zkutil::extractZooKeeperPath(*zk_path, false, nullptr); zkutil::ZooKeeperPtr zookeeper = (zk_name == getDefaultZooKeeperName()) ? global_context->getZooKeeper() : global_context->getAuxiliaryZooKeeper(zk_name); String id; if (!zookeeper->tryGet(fs::path(*zk_path) / "table_shared_id", id)) return {}; return id; } void StorageReplicatedMergeTree::lockSharedDataTemporary(const String & part_name, const String & part_id, const DiskPtr & disk) const { auto settings = getSettings(); if (!disk || !disk->supportZeroCopyReplication() || !settings->allow_remote_fs_zero_copy_replication) return; zkutil::ZooKeeperPtr zookeeper = tryGetZooKeeper(); if (!zookeeper) return; String id = part_id; boost::replace_all(id, "/", "_"); Strings zc_zookeeper_paths = getZeroCopyPartPath(*getSettings(), toString(disk->getDataSourceDescription().type), getTableSharedID(), part_name, zookeeper_path); for (const auto & zc_zookeeper_path : zc_zookeeper_paths) { String zookeeper_node = fs::path(zc_zookeeper_path) / id / replica_name; LOG_TRACE(log, "Set zookeeper temporary ephemeral lock {}", zookeeper_node); createZeroCopyLockNode( std::make_shared(zookeeper), zookeeper_node, zkutil::CreateMode::Ephemeral, false); } } void StorageReplicatedMergeTree::lockSharedData( const IMergeTreeDataPart & part, bool replace_existing_lock, std::optional hardlinked_files) const { auto zookeeper = tryGetZooKeeper(); if (zookeeper) return lockSharedData(part, std::make_shared(zookeeper), replace_existing_lock, hardlinked_files); else return lockSharedData(part, std::make_shared(nullptr), replace_existing_lock, hardlinked_files); } void StorageReplicatedMergeTree::lockSharedData( const IMergeTreeDataPart & part, const ZooKeeperWithFaultInjectionPtr & zookeeper, bool replace_existing_lock, std::optional hardlinked_files) const { auto settings = getSettings(); if (!part.isStoredOnDisk() || !settings->allow_remote_fs_zero_copy_replication) return; if (!part.getDataPartStorage().supportZeroCopyReplication()) return; if (zookeeper->isNull()) return; String id = part.getUniqueId(); boost::replace_all(id, "/", "_"); Strings zc_zookeeper_paths = getZeroCopyPartPath( *getSettings(), part.getDataPartStorage().getDiskType(), getTableSharedID(), part.name, zookeeper_path); String path_to_set_hardlinked_files; NameSet hardlinks; if (hardlinked_files.has_value() && !hardlinked_files->hardlinks_from_source_part.empty()) { path_to_set_hardlinked_files = getZeroCopyPartPath( *getSettings(), part.getDataPartStorage().getDiskType(), hardlinked_files->source_table_shared_id, hardlinked_files->source_part_name, zookeeper_path)[0]; hardlinks = hardlinked_files->hardlinks_from_source_part; } for (const auto & zc_zookeeper_path : zc_zookeeper_paths) { String zookeeper_node = fs::path(zc_zookeeper_path) / id / replica_name; LOG_TRACE(log, "Set zookeeper persistent lock {}", zookeeper_node); createZeroCopyLockNode( zookeeper, zookeeper_node, zkutil::CreateMode::Persistent, replace_existing_lock, path_to_set_hardlinked_files, hardlinks); } } std::pair StorageReplicatedMergeTree::unlockSharedData(const IMergeTreeDataPart & part) const { return unlockSharedData(part, std::make_shared(nullptr)); } std::pair StorageReplicatedMergeTree::unlockSharedData(const IMergeTreeDataPart & part, const ZooKeeperWithFaultInjectionPtr & zookeeper) const { auto settings = getSettings(); if (!settings->allow_remote_fs_zero_copy_replication) return std::make_pair(true, NameSet{}); if (!part.isStoredOnDisk()) { LOG_TRACE(log, "Part {} is not stored on disk, blobs can be removed", part.name); return std::make_pair(true, NameSet{}); } if (!part.getDataPartStorage().supportZeroCopyReplication()) { LOG_TRACE(log, "Part {} is not stored on zero-copy replicated disk, blobs can be removed", part.name); return std::make_pair(true, NameSet{}); } auto shared_id = getTableSharedID(); if (shared_id == toString(UUIDHelpers::Nil)) { if (zookeeper->exists(zookeeper_path)) { LOG_WARNING(log, "Not removing shared data for part {} because replica does not have metadata in ZooKeeper, " "but table path exist and other replicas may exist. It may leave some garbage on S3", part.name); return std::make_pair(false, NameSet{}); } LOG_TRACE(log, "Part {} blobs can be removed, because table {} completely dropped", part.name, getStorageID().getNameForLogs()); return std::make_pair(true, NameSet{}); } /// If part is temporary refcount file may be absent if (part.getDataPartStorage().exists(IMergeTreeDataPart::FILE_FOR_REFERENCES_CHECK)) { auto ref_count = part.getDataPartStorage().getRefCount(IMergeTreeDataPart::FILE_FOR_REFERENCES_CHECK); if (ref_count > 0) /// Keep part shard info for frozen backups { LOG_TRACE(log, "Part {} has more than zero local references ({}), blobs cannot be removed", part.name, ref_count); return std::make_pair(false, NameSet{}); } else { LOG_TRACE(log, "Part {} local references is zero, will check blobs can be removed in zookeeper", part.name); } } else { LOG_TRACE(log, "Part {} looks temporary, because {} file doesn't exists, blobs can be removed", part.name, IMergeTreeDataPart::FILE_FOR_REFERENCES_CHECK); /// Temporary part with some absent file cannot be locked in shared mode return std::make_pair(true, NameSet{}); } if (has_metadata_in_zookeeper.has_value() && !has_metadata_in_zookeeper) { if (zookeeper->exists(zookeeper_path)) { LOG_WARNING(log, "Not removing shared data for part {} because replica does not have metadata in ZooKeeper, " "but table path exist and other replicas may exist. It may leave some garbage on S3", part.name); return std::make_pair(false, NameSet{}); } /// If table was completely dropped (no meta in zookeeper) we can safely remove parts return std::make_pair(true, NameSet{}); } /// We remove parts during table shutdown. If exception happen, restarting thread will be already turned /// off and nobody will reconnect our zookeeper connection. In this case we use zookeeper connection from /// context. if (shutdown_called.load()) zookeeper->setKeeper(getZooKeeperIfTableShutDown()); else zookeeper->setKeeper(getZooKeeper()); /// It can happen that we didn't had the connection to zookeeper during table creation, but actually /// table is completely dropped, so we can drop it without any additional checks. if (!has_metadata_in_zookeeper.has_value() && !zookeeper->exists(zookeeper_path)) return std::make_pair(true, NameSet{}); return unlockSharedDataByID( part.getUniqueId(), shared_id, part.name, replica_name, part.getDataPartStorage().getDiskType(), zookeeper, *getSettings(), log, zookeeper_path, format_version); } namespace { /// What is going on here? /// Actually we need this code because of flaws in hardlinks tracking. When we create child part during mutation we can hardlink some files from parent part, like /// all_0_0_0: /// a.bin a.mrk2 columns.txt ... /// all_0_0_0_1: ^ ^ /// a.bin a.mrk2 columns.txt /// So when we deleting all_0_0_0 it doesn't remove blobs for a.bin and a.mrk2 because all_0_0_0_1 use them. /// But sometimes we need an opposite. When we deleting all_0_0_0_1 it can be non replicated to other replicas, so we are the only owner of this part. /// In this case when we will drop all_0_0_0_1 we will drop blobs for all_0_0_0. But it will lead to dataloss. For such case we need to check that other replicas /// still need parent part. std::pair getParentLockedBlobs(const ZooKeeperWithFaultInjectionPtr & zookeeper_ptr, const std::string & zero_copy_part_path_prefix, const std::string & part_info_str, MergeTreeDataFormatVersion format_version, Poco::Logger * log) { NameSet files_not_to_remove; MergeTreePartInfo part_info = MergeTreePartInfo::fromPartName(part_info_str, format_version); /// No mutations -- no hardlinks -- no issues if (part_info.mutation == 0) return {false, files_not_to_remove}; /// Getting all zero copy parts Strings parts_str; zookeeper_ptr->tryGetChildren(zero_copy_part_path_prefix, parts_str); /// Parsing infos. It's hard to convert info -> string for old-format merge tree /// so storing string as is. std::vector> parts_infos; for (const auto & part_str : parts_str) { MergeTreePartInfo parent_candidate_info = MergeTreePartInfo::fromPartName(part_str, format_version); parts_infos.emplace_back(parent_candidate_info, part_str); } /// Sort is important. We need to find our closest parent, like: /// for part all_0_0_0_64 we can have parents /// all_0_0_0_6 < we need the closest parent, not others /// all_0_0_0_1 /// all_0_0_0 std::sort(parts_infos.begin(), parts_infos.end()); /// In reverse order to process from bigger to smaller for (const auto & [parent_candidate_info, part_candidate_info_str] : parts_infos | std::views::reverse) { if (parent_candidate_info == part_info) continue; /// We are mutation child of this parent if (part_info.isMutationChildOf(parent_candidate_info)) { LOG_TRACE(log, "Found mutation parent {} for part {}", part_candidate_info_str, part_info_str); /// Get hardlinked files String files_not_to_remove_str; Coordination::Error code; zookeeper_ptr->tryGet(fs::path(zero_copy_part_path_prefix) / part_candidate_info_str, files_not_to_remove_str, nullptr, nullptr, &code); if (code != Coordination::Error::ZOK) LOG_TRACE(log, "Cannot get parent files from ZooKeeper on path ({}), error {}", (fs::path(zero_copy_part_path_prefix) / part_candidate_info_str).string(), errorMessage(code)); if (!files_not_to_remove_str.empty()) { boost::split(files_not_to_remove, files_not_to_remove_str, boost::is_any_of("\n ")); LOG_TRACE(log, "Found files not to remove from parent part {}: [{}]", part_candidate_info_str, fmt::join(files_not_to_remove, ", ")); } return {true, files_not_to_remove}; } } return {false, files_not_to_remove}; } } std::pair StorageReplicatedMergeTree::unlockSharedDataByID( String part_id, const String & table_uuid, const String & part_name, const String & replica_name_, const std::string & disk_type, const ZooKeeperWithFaultInjectionPtr & zookeeper_ptr, const MergeTreeSettings & settings, Poco::Logger * logger, const String & zookeeper_path_old, MergeTreeDataFormatVersion data_format_version) { boost::replace_all(part_id, "/", "_"); Strings zc_zookeeper_paths = getZeroCopyPartPath(settings, disk_type, table_uuid, part_name, zookeeper_path_old); bool part_has_no_more_locks = true; NameSet files_not_to_remove; for (const auto & zc_zookeeper_path : zc_zookeeper_paths) { String files_not_to_remove_str; zookeeper_ptr->tryGet(zc_zookeeper_path, files_not_to_remove_str); files_not_to_remove.clear(); if (!files_not_to_remove_str.empty()) boost::split(files_not_to_remove, files_not_to_remove_str, boost::is_any_of("\n ")); auto [has_parent, parent_not_to_remove] = getParentLockedBlobs( zookeeper_ptr, fs::path(zc_zookeeper_path).parent_path(), part_name, data_format_version, logger); files_not_to_remove.insert(parent_not_to_remove.begin(), parent_not_to_remove.end()); String zookeeper_part_uniq_node = fs::path(zc_zookeeper_path) / part_id; /// Delete our replica node for part from zookeeper (we are not interested in it anymore) String zookeeper_part_replica_node = fs::path(zookeeper_part_uniq_node) / replica_name_; LOG_TRACE(logger, "Remove zookeeper lock {} for part {}", zookeeper_part_replica_node, part_name); if (auto ec = zookeeper_ptr->tryRemove(zookeeper_part_replica_node); ec != Coordination::Error::ZOK) { /// Very complex case. It means that lock already doesn't exist when we tried to remove it. /// So we don't know are we owner of this part or not. Maybe we just mutated it, renamed on disk and failed to lock in ZK. /// But during mutation we can have hardlinks to another part. So it's not Ok to remove blobs of this part if it was mutated. if (ec == Coordination::Error::ZNONODE) { if (has_parent) { LOG_INFO(logger, "Lock on path {} for part {} doesn't exist, refuse to remove blobs", zookeeper_part_replica_node, part_name); return {false, {}}; } } else { throw zkutil::KeeperException(ec, zookeeper_part_replica_node); } } /// Check, maybe we were the last replica and can remove part forever Strings children; zookeeper_ptr->tryGetChildren(zookeeper_part_uniq_node, children); if (!children.empty()) { LOG_TRACE(logger, "Found {} ({}) zookeeper locks for {}", children.size(), fmt::join(children, ", "), zookeeper_part_uniq_node); part_has_no_more_locks = false; continue; } else { LOG_TRACE(logger, "No more children left for for {}, will try to remove the whole node", zookeeper_part_uniq_node); } auto error_code = zookeeper_ptr->tryRemove(zookeeper_part_uniq_node); if (error_code == Coordination::Error::ZOK) { LOG_TRACE(logger, "Removed last parent zookeeper lock {} for part {} with id {}", zookeeper_part_uniq_node, part_name, part_id); } else if (error_code == Coordination::Error::ZNOTEMPTY) { LOG_TRACE(logger, "Cannot remove last parent zookeeper lock {} for part {} with id {}, another replica locked part concurrently", zookeeper_part_uniq_node, part_name, part_id); } else if (error_code == Coordination::Error::ZNONODE) { LOG_TRACE(logger, "Node with parent zookeeper lock {} for part {} with id {} doesn't exist", zookeeper_part_uniq_node, part_name, part_id); } else { throw zkutil::KeeperException(error_code, zookeeper_part_uniq_node); } /// Even when we have lock with same part name, but with different uniq, we can remove files on S3 children.clear(); String zookeeper_part_node = fs::path(zookeeper_part_uniq_node).parent_path(); zookeeper_ptr->tryGetChildren(zookeeper_part_node, children); if (children.empty()) { /// Cleanup after last uniq removing error_code = zookeeper_ptr->tryRemove(zookeeper_part_node); if (error_code == Coordination::Error::ZOK) { LOG_TRACE(logger, "Removed last parent zookeeper lock {} for part {} (part is finally unlocked)", zookeeper_part_uniq_node, part_name); } else if (error_code == Coordination::Error::ZNOTEMPTY) { LOG_TRACE(logger, "Cannot remove last parent zookeeper lock {} for part {}, another replica locked part concurrently", zookeeper_part_uniq_node, part_name); } else if (error_code == Coordination::Error::ZNONODE) { LOG_TRACE(logger, "Node with parent zookeeper lock {} for part {} doesn't exist (part was unlocked before)", zookeeper_part_uniq_node, part_name); } else { throw zkutil::KeeperException(error_code, zookeeper_part_uniq_node); } } else { LOG_TRACE(logger, "Can't remove parent zookeeper lock {} for part {}, because children {} ({}) exists", zookeeper_part_node, part_name, children.size(), fmt::join(children, ", ")); } } return std::make_pair(part_has_no_more_locks, files_not_to_remove); } MutableDataPartStoragePtr StorageReplicatedMergeTree::tryToFetchIfShared( const IMergeTreeDataPart & part, const DiskPtr & disk, const String & path) { const auto settings = getSettings(); auto data_source_description = disk->getDataSourceDescription(); if (!(disk->supportZeroCopyReplication() && settings->allow_remote_fs_zero_copy_replication)) return nullptr; String replica = getSharedDataReplica(part, data_source_description.type); /// We can't fetch part when none replicas have this part on a same type remote disk if (replica.empty()) return nullptr; return executeFetchShared(replica, part.name, disk, path); } String StorageReplicatedMergeTree::getSharedDataReplica( const IMergeTreeDataPart & part, DataSourceType data_source_type) const { String best_replica; zkutil::ZooKeeperPtr zookeeper = tryGetZooKeeper(); if (!zookeeper) return ""; Strings zc_zookeeper_paths = getZeroCopyPartPath(*getSettings(), toString(data_source_type), getTableSharedID(), part.name, zookeeper_path); std::set replicas; for (const auto & zc_zookeeper_path : zc_zookeeper_paths) { Strings ids; zookeeper->tryGetChildren(zc_zookeeper_path, ids); for (const auto & id : ids) { String zookeeper_part_uniq_node = fs::path(zc_zookeeper_path) / id; Strings id_replicas; zookeeper->tryGetChildren(zookeeper_part_uniq_node, id_replicas); LOG_TRACE(log, "Found zookeeper replicas for {}: {}", zookeeper_part_uniq_node, id_replicas.size()); replicas.insert(id_replicas.begin(), id_replicas.end()); } } LOG_TRACE(log, "Found zookeeper replicas for part {}: {}", part.name, replicas.size()); Strings active_replicas; /// TODO: Move best replica choose in common method (here is the same code as in StorageReplicatedMergeTree::fetchPartition) /// Leave only active replicas. active_replicas.reserve(replicas.size()); for (const String & replica : replicas) if ((replica != replica_name) && (zookeeper->exists(fs::path(zookeeper_path) / "replicas" / replica / "is_active"))) active_replicas.push_back(replica); LOG_TRACE(log, "Found zookeeper active replicas for part {}: {}", part.name, active_replicas.size()); if (active_replicas.empty()) return ""; /** You must select the best (most relevant) replica. * This is a replica with the maximum `log_pointer`, then with the minimum `queue` size. * NOTE This is not exactly the best criteria. It does not make sense to download old partitions, * and it would be nice to be able to choose the replica closest by network. * NOTE Of course, there are data races here. You can solve it by retrying. */ Int64 max_log_pointer = -1; UInt64 min_queue_size = std::numeric_limits::max(); for (const String & replica : active_replicas) { String current_replica_path = fs::path(zookeeper_path) / "replicas" / replica; String log_pointer_str = zookeeper->get(fs::path(current_replica_path) / "log_pointer"); Int64 log_pointer = log_pointer_str.empty() ? 0 : parse(log_pointer_str); Coordination::Stat stat; zookeeper->get(fs::path(current_replica_path) / "queue", &stat); size_t queue_size = stat.numChildren; if (log_pointer > max_log_pointer || (log_pointer == max_log_pointer && queue_size < min_queue_size)) { max_log_pointer = log_pointer; min_queue_size = queue_size; best_replica = replica; } } return best_replica; } Strings StorageReplicatedMergeTree::getZeroCopyPartPath( const MergeTreeSettings & settings, const std::string & disk_type, const String & table_uuid, const String & part_name, const String & zookeeper_path_old) { Strings res; String zero_copy = fmt::format("zero_copy_{}", disk_type); String new_path = fs::path(settings.remote_fs_zero_copy_zookeeper_path.toString()) / zero_copy / table_uuid / part_name; res.push_back(std::move(new_path)); if (settings.remote_fs_zero_copy_path_compatible_mode && !zookeeper_path_old.empty()) { /// Compatibility mode for cluster with old and new versions String old_path = fs::path(zookeeper_path_old) / zero_copy / "shared" / part_name; res.push_back(std::move(old_path)); } return res; } bool StorageReplicatedMergeTree::checkZeroCopyLockExists(const String & part_name, const DiskPtr & disk, String & lock_replica) { auto path = getZeroCopyPartPath(part_name, disk); if (path) { /// FIXME auto lock_path = fs::path(*path) / "part_exclusive_lock"; if (getZooKeeper()->tryGet(lock_path, lock_replica)) { return true; } } return false; } std::optional StorageReplicatedMergeTree::getZeroCopyPartPath(const String & part_name, const DiskPtr & disk) { if (!disk || !disk->supportZeroCopyReplication()) return std::nullopt; return getZeroCopyPartPath(*getSettings(), toString(disk->getDataSourceDescription().type), getTableSharedID(), part_name, zookeeper_path)[0]; } std::optional StorageReplicatedMergeTree::tryCreateZeroCopyExclusiveLock(const String & part_name, const DiskPtr & disk) { if (!disk || !disk->supportZeroCopyReplication()) return std::nullopt; zkutil::ZooKeeperPtr zookeeper = tryGetZooKeeper(); if (!zookeeper) return std::nullopt; String zc_zookeeper_path = *getZeroCopyPartPath(part_name, disk); /// Just recursively create ancestors for lock zookeeper->createAncestors(zc_zookeeper_path); zookeeper->createIfNotExists(zc_zookeeper_path, ""); /// Create actual lock ZeroCopyLock lock(zookeeper, zc_zookeeper_path, replica_name); if (lock.lock->tryLock()) return lock; else return std::nullopt; } String StorageReplicatedMergeTree::findReplicaHavingPart( const String & part_name, const String & zookeeper_path_, zkutil::ZooKeeper::Ptr zookeeper_ptr) { Strings replicas = zookeeper_ptr->getChildren(fs::path(zookeeper_path_) / "replicas"); /// Select replicas in uniformly random order. std::shuffle(replicas.begin(), replicas.end(), thread_local_rng); for (const String & replica : replicas) { if (zookeeper_ptr->exists(fs::path(zookeeper_path_) / "replicas" / replica / "parts" / part_name) && zookeeper_ptr->exists(fs::path(zookeeper_path_) / "replicas" / replica / "is_active")) return fs::path(zookeeper_path_) / "replicas" / replica; } return {}; } bool StorageReplicatedMergeTree::checkIfDetachedPartExists(const String & part_name) { fs::directory_iterator dir_end; for (const std::string & path : getDataPaths()) for (fs::directory_iterator dir_it{fs::path(path) / "detached/"}; dir_it != dir_end; ++dir_it) if (dir_it->path().filename().string() == part_name) return true; return false; } bool StorageReplicatedMergeTree::checkIfDetachedPartitionExists(const String & partition_name) { fs::directory_iterator dir_end; for (const std::string & path : getDataPaths()) { for (fs::directory_iterator dir_it{fs::path(path) / "detached/"}; dir_it != dir_end; ++dir_it) { const String file_name = dir_it->path().filename().string(); auto part_info = MergeTreePartInfo::tryParsePartName(file_name, format_version); if (part_info && part_info->partition_id == partition_name) return true; } } return false; } bool StorageReplicatedMergeTree::createEmptyPartInsteadOfLost(zkutil::ZooKeeperPtr zookeeper, const String & lost_part_name) { LOG_INFO(log, "Going to replace lost part {} with empty part", lost_part_name); auto new_part_info = MergeTreePartInfo::fromPartName(lost_part_name, format_version); auto metadata_snapshot = getInMemoryMetadataPtr(); MergeTreePartition partition; { DataPartsLock lock = lockParts(); auto parts_in_partition = getDataPartsPartitionRange(new_part_info.partition_id); if (!parts_in_partition.empty()) { partition = (*parts_in_partition.begin())->partition; } else if (auto parsed_partition = MergeTreePartition::tryParseValueFromID( new_part_info.partition_id, metadata_snapshot->getPartitionKey().sample_block)) { partition = MergeTreePartition(*parsed_partition); } else { LOG_WARNING(log, "Empty part {} is not created instead of lost part because there are no parts in partition {} (it's empty), " "resolve this manually using DROP/DETACH PARTITION.", lost_part_name, new_part_info.partition_id); return false; } } MergeTreeData::MutableDataPartPtr new_data_part = createEmptyPart(new_part_info, partition, lost_part_name, NO_TRANSACTION_PTR); new_data_part->name = lost_part_name; try { MergeTreeData::Transaction transaction(*this, NO_TRANSACTION_RAW); auto replaced_parts = renameTempPartAndReplace(new_data_part, transaction); if (!replaced_parts.empty()) { Strings part_names; for (const auto & part : replaced_parts) part_names.emplace_back(part->name); /// Why this exception is not a LOGICAL_ERROR? Because it's possible /// to have some source parts for the lost part if replica currently /// cloning from another replica, but source replica lost covering /// part and finished MERGE_PARTS before clone. It's an extremely /// rare case and it's unclear how to resolve it better. Eventually /// source replica will replace lost part with empty part and we /// will fetch this empty part instead of our source parts. This /// will make replicas consistent, but some data will be lost. throw Exception(ErrorCodes::INCORRECT_DATA, "Tried to create empty part {}, but it replaces existing parts {}.", lost_part_name, fmt::join(part_names, ", ")); } lockSharedData(*new_data_part, false, {}); while (true) { /// We should be careful when creating an empty part, because we are not sure that this part is still needed. /// For example, it's possible that part (or partition) was dropped (or replaced) concurrently. /// We can enqueue part for check from DataPartExchange or SelectProcessor /// and it's hard to synchronize it with ReplicatedMergeTreeQueue and PartCheckThread... /// But at least we can ignore parts that are definitely not needed according to virtual parts and drop ranges. auto pred = queue.getMergePredicate(zookeeper, PartitionIdsHint{new_part_info.partition_id}); String covering_virtual = pred.getCoveringVirtualPart(lost_part_name); if (covering_virtual.empty()) { LOG_WARNING(log, "Will not create empty part instead of lost {}, because there's no covering part in replication queue", lost_part_name); return false; } MergeTreePartInfo drop_info; if (pred.isGoingToBeDropped(MergeTreePartInfo::fromPartName(lost_part_name, format_version), &drop_info)) { LOG_WARNING(log, "Will not create empty part instead of lost {}, " "because it's going to be removed (by range {})", lost_part_name, drop_info.getPartNameForLogs()); return false; } Coordination::Requests ops; Coordination::Stat replicas_stat; auto replicas_path = fs::path(zookeeper_path) / "replicas"; Strings replicas = zookeeper->getChildren(replicas_path, &replicas_stat); ops.emplace_back(zkutil::makeCheckRequest(zookeeper_path + "/log", pred.getVersion())); /// In rare cases new replica can appear during check ops.emplace_back(zkutil::makeCheckRequest(replicas_path, replicas_stat.version)); for (const String & replica : replicas) { String current_part_path = fs::path(zookeeper_path) / "replicas" / replica / "parts" / lost_part_name; /// We must be sure that this part doesn't exist on other replicas if (!zookeeper->exists(current_part_path)) { ops.emplace_back(zkutil::makeCreateRequest(current_part_path, "", zkutil::CreateMode::Persistent)); ops.emplace_back(zkutil::makeRemoveRequest(current_part_path, -1)); } else { throw Exception(ErrorCodes::DUPLICATE_DATA_PART, "Part {} already exists on replica {} on path {}", lost_part_name, replica, current_part_path); } } getCommitPartOps(ops, new_data_part); Coordination::Responses responses; if (auto code = zookeeper->tryMulti(ops, responses); code == Coordination::Error::ZOK) { transaction.commit(); break; } else if (code == Coordination::Error::ZBADVERSION) { LOG_INFO(log, "Looks like log was updated or new replica appeared while creating new empty part, will retry"); } else { zkutil::KeeperMultiException::check(code, ops, responses); } } } catch (const Exception & ex) { LOG_WARNING(log, "Cannot commit empty part {} with error {}", lost_part_name, ex.displayText()); return false; } LOG_INFO(log, "Created empty part {} instead of lost part", lost_part_name); return true; } void StorageReplicatedMergeTree::createZeroCopyLockNode( const ZooKeeperWithFaultInjectionPtr & zookeeper, const String & zookeeper_node, int32_t mode, bool replace_existing_lock, const String & path_to_set_hardlinked_files, const NameSet & hardlinked_files) { /// In rare case other replica can remove path between createAncestors and createIfNotExists /// So we make up to 5 attempts bool created = false; for (int attempts = 5; attempts > 0; --attempts) { try { /// Ephemeral locks can be created only when we fetch shared data. /// So it never require to create ancestors. If we create them /// race condition with source replica drop is possible. if (mode == zkutil::CreateMode::Persistent) zookeeper->createAncestors(zookeeper_node); if (replace_existing_lock && zookeeper->exists(zookeeper_node)) { Coordination::Requests ops; ops.emplace_back(zkutil::makeRemoveRequest(zookeeper_node, -1)); ops.emplace_back(zkutil::makeCreateRequest(zookeeper_node, "", mode)); if (!path_to_set_hardlinked_files.empty() && !hardlinked_files.empty()) { std::string data = boost::algorithm::join(hardlinked_files, "\n"); /// List of files used to detect hardlinks. path_to_set_hardlinked_files -- /// is a path to source part zero copy node. During part removal hardlinked /// files will be left for source part. ops.emplace_back(zkutil::makeSetRequest(path_to_set_hardlinked_files, data, -1)); } Coordination::Responses responses; auto error = zookeeper->tryMulti(ops, responses); if (error == Coordination::Error::ZOK) { created = true; break; } else if (error == Coordination::Error::ZNONODE && mode != zkutil::CreateMode::Persistent) { throw Exception(ErrorCodes::NOT_FOUND_NODE, "Cannot create ephemeral zero copy lock {} because part was unlocked from zookeeper", zookeeper_node); } } else { Coordination::Requests ops; if (!path_to_set_hardlinked_files.empty() && !hardlinked_files.empty()) { std::string data = boost::algorithm::join(hardlinked_files, "\n"); /// List of files used to detect hardlinks. path_to_set_hardlinked_files -- /// is a path to source part zero copy node. During part removal hardlinked /// files will be left for source part. ops.emplace_back(zkutil::makeSetRequest(path_to_set_hardlinked_files, data, -1)); } ops.emplace_back(zkutil::makeCreateRequest(zookeeper_node, "", mode)); Coordination::Responses responses; auto error = zookeeper->tryMulti(ops, responses); if (error == Coordination::Error::ZOK || error == Coordination::Error::ZNODEEXISTS) { created = true; break; } else if (error == Coordination::Error::ZNONODE && mode != zkutil::CreateMode::Persistent) { /// Ephemeral locks used during fetches so if parent node was removed we cannot do anything throw Exception(ErrorCodes::NOT_FOUND_NODE, "Cannot create ephemeral zero copy lock {} because part was unlocked from zookeeper", zookeeper_node); } } } catch (const zkutil::KeeperException & e) { if (e.code == Coordination::Error::ZNONODE) continue; throw; } } if (!created) { String mode_str = mode == zkutil::CreateMode::Persistent ? "persistent" : "ephemeral"; throw Exception(ErrorCodes::NOT_FOUND_NODE, "Cannot create {} zero copy lock {} because part was unlocked from zookeeper", mode_str, zookeeper_node); } } bool StorageReplicatedMergeTree::removeDetachedPart(DiskPtr disk, const String & path, const String & part_name) { if (disk->supportZeroCopyReplication()) { String table_id = getTableSharedID(); return removeSharedDetachedPart(disk, path, part_name, table_id, replica_name, zookeeper_path, getContext(), current_zookeeper); } disk->removeRecursive(path); return false; } bool StorageReplicatedMergeTree::removeSharedDetachedPart(DiskPtr disk, const String & path, const String & part_name, const String & table_uuid, const String & detached_replica_name, const String & detached_zookeeper_path, const ContextPtr & local_context, const zkutil::ZooKeeperPtr & zookeeper) { bool keep_shared = false; NameSet files_not_to_remove; fs::path checksums = fs::path(path) / IMergeTreeDataPart::FILE_FOR_REFERENCES_CHECK; if (disk->exists(checksums)) { if (disk->getRefCount(checksums) == 0) { String id = disk->getUniqueId(checksums); bool can_remove = false; std::tie(can_remove, files_not_to_remove) = StorageReplicatedMergeTree::unlockSharedDataByID( id, table_uuid, part_name, detached_replica_name, toString(disk->getDataSourceDescription().type), std::make_shared(zookeeper), local_context->getReplicatedMergeTreeSettings(), &Poco::Logger::get("StorageReplicatedMergeTree"), detached_zookeeper_path, MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING); keep_shared = !can_remove; } else keep_shared = true; } disk->removeSharedRecursive(path, keep_shared, files_not_to_remove); return keep_shared; } void StorageReplicatedMergeTree::createAndStoreFreezeMetadata(DiskPtr disk, DataPartPtr, String backup_part_path) const { if (disk->supportZeroCopyReplication()) { FreezeMetaData meta; meta.fill(*this); meta.save(disk, backup_part_path); } } void StorageReplicatedMergeTree::adjustCreateQueryForBackup(ASTPtr & create_query) const { /// Adjust the create query using values from ZooKeeper. auto zookeeper = getZooKeeper(); auto columns_from_entry = ColumnsDescription::parse(zookeeper->get(fs::path(zookeeper_path) / "columns")); auto metadata_from_entry = ReplicatedMergeTreeTableMetadata::parse(zookeeper->get(fs::path(zookeeper_path) / "metadata")); auto current_metadata = getInMemoryMetadataPtr(); auto metadata_diff = ReplicatedMergeTreeTableMetadata(*this, current_metadata).checkAndFindDiff(metadata_from_entry, current_metadata->getColumns(), getContext()); auto adjusted_metadata = metadata_diff.getNewMetadata(columns_from_entry, getContext(), *current_metadata); applyMetadataChangesToCreateQuery(create_query, adjusted_metadata); /// Check that tryGetTableSharedIDFromCreateQuery() works for this storage. if (tryGetTableSharedIDFromCreateQuery(*create_query, getContext()) != getTableSharedID()) throw Exception(ErrorCodes::LOGICAL_ERROR, "Table {} has its shared ID to be different from one from the create query"); } void StorageReplicatedMergeTree::backupData( BackupEntriesCollector & backup_entries_collector, const String & data_path_in_backup, const std::optional & partitions) { /// First we generate backup entries in the same way as an ordinary MergeTree does. /// But then we don't add them to the BackupEntriesCollector right away, /// because we need to coordinate them with other replicas (other replicas can have better parts). auto local_context = backup_entries_collector.getContext(); DataPartsVector data_parts; if (partitions) data_parts = getVisibleDataPartsVectorInPartitions(local_context, getPartitionIDsFromQuery(*partitions, local_context)); else data_parts = getVisibleDataPartsVector(local_context); auto backup_entries = backupParts(data_parts, /* data_path_in_backup */ "", local_context); auto coordination = backup_entries_collector.getBackupCoordination(); String shared_id = getTableSharedID(); coordination->addReplicatedDataPath(shared_id, data_path_in_backup); std::unordered_map part_names_with_hashes_calculating; for (auto & [relative_path, backup_entry] : backup_entries) { size_t slash_pos = relative_path.find('/'); if (slash_pos != String::npos) { String part_name = relative_path.substr(0, slash_pos); if (MergeTreePartInfo::tryParsePartName(part_name, MERGE_TREE_DATA_MIN_FORMAT_VERSION_WITH_CUSTOM_PARTITIONING)) { auto & hash = part_names_with_hashes_calculating[part_name]; if (relative_path.ends_with(".bin")) { auto checksum = backup_entry->getChecksum(); hash.update(relative_path); hash.update(backup_entry->getSize()); hash.update(*checksum); } continue; } } /// Not a part name, probably error. throw Exception(ErrorCodes::LOGICAL_ERROR, "{} doesn't follow the format /", quoteString(relative_path)); } std::vector part_names_with_hashes; part_names_with_hashes.reserve(part_names_with_hashes_calculating.size()); for (auto & [part_name, hash] : part_names_with_hashes_calculating) { UInt128 checksum; hash.get128(checksum); auto & part_name_with_hash = part_names_with_hashes.emplace_back(); part_name_with_hash.part_name = part_name; part_name_with_hash.checksum = checksum; } /// Send our list of part names to the coordination (to compare with other replicas). coordination->addReplicatedPartNames(shared_id, getStorageID().getFullTableName(), getReplicaName(), part_names_with_hashes); /// Send a list of mutations to the coordination too (we need to find the mutations which are not finished for added part names). { auto zookeeper = getZooKeeper(); Strings mutation_ids; if (zookeeper->tryGetChildren(fs::path(zookeeper_path) / "mutations", mutation_ids) == Coordination::Error::ZOK) { std::vector mutation_infos; mutation_infos.reserve(mutation_ids.size()); for (const auto & mutation_id : mutation_ids) { String mutation; if (zookeeper->tryGet(fs::path(zookeeper_path) / "mutations" / mutation_id, mutation)) mutation_infos.emplace_back(IBackupCoordination::MutationInfo{mutation_id, mutation}); } coordination->addReplicatedMutations(shared_id, getStorageID().getFullTableName(), getReplicaName(), mutation_infos); } } /// This task will be executed after all replicas have collected their parts and the coordination is ready to /// give us the final list of parts to add to the BackupEntriesCollector. auto post_collecting_task = [shared_id, replica_name = getReplicaName(), coordination, backup_entries = std::move(backup_entries), &backup_entries_collector]() { Strings data_paths = coordination->getReplicatedDataPaths(shared_id); std::vector data_paths_fs; data_paths_fs.reserve(data_paths.size()); for (const auto & data_path : data_paths) data_paths_fs.push_back(data_path); Strings part_names = coordination->getReplicatedPartNames(shared_id, replica_name); std::unordered_set part_names_set{part_names.begin(), part_names.end()}; for (const auto & [relative_path, backup_entry] : backup_entries) { size_t slash_pos = relative_path.find('/'); String part_name = relative_path.substr(0, slash_pos); if (!part_names_set.contains(part_name)) continue; for (const auto & data_path : data_paths_fs) backup_entries_collector.addBackupEntry(data_path / relative_path, backup_entry); } auto mutation_infos = coordination->getReplicatedMutations(shared_id, replica_name); for (const auto & mutation_info : mutation_infos) { auto backup_entry = ReplicatedMergeTreeMutationEntry::parse(mutation_info.entry, mutation_info.id).backup(); for (const auto & data_path : data_paths_fs) backup_entries_collector.addBackupEntry(data_path / "mutations" / (mutation_info.id + ".txt"), backup_entry); } }; backup_entries_collector.addPostTask(post_collecting_task); } void StorageReplicatedMergeTree::restoreDataFromBackup(RestorerFromBackup & restorer, const String & data_path_in_backup, const std::optional & partitions) { String full_zk_path = getZooKeeperName() + getZooKeeperPath(); if (!restorer.getRestoreCoordination()->acquireInsertingDataIntoReplicatedTable(full_zk_path)) { /// Other replica is already restoring the data of this table. /// We'll get them later due to replication, it's not necessary to read it from the backup. return; } if (!restorer.isNonEmptyTableAllowed()) { bool empty = !getTotalActiveSizeInBytes(); if (empty) { /// New parts could be in the replication queue but not fetched yet. /// In that case we consider the table as not empty. ReplicatedTableStatus status; getStatus(status, /* with_zk_fields = */ false); if (status.queue.inserts_in_queue) empty = false; } auto backup = restorer.getBackup(); if (!empty && backup->hasFiles(data_path_in_backup)) restorer.throwTableIsNotEmpty(getStorageID()); } restorePartsFromBackup(restorer, data_path_in_backup, partitions); } void StorageReplicatedMergeTree::attachRestoredParts(MutableDataPartsVector && parts) { auto metadata_snapshot = getInMemoryMetadataPtr(); auto sink = std::make_shared(*this, metadata_snapshot, 0, 0, 0, false, false, false, getContext(), /*is_attach*/true); for (auto part : parts) sink->writeExistingPart(part); } template std::optional StorageReplicatedMergeTree::allocateBlockNumber( const String & partition_id, const ZooKeeperWithFaultInjectionPtr & zookeeper, const String & zookeeper_block_id_path, const String & zookeeper_path_prefix) const; template std::optional StorageReplicatedMergeTree::allocateBlockNumber>( const String & partition_id, const ZooKeeperWithFaultInjectionPtr & zookeeper, const std::vector & zookeeper_block_id_path, const String & zookeeper_path_prefix) const; #if 0 PartsTemporaryRename renamed_parts(*this, "detached/"); MutableDataPartsVector loaded_parts = tryLoadPartsToAttach(partition, attach_part, query_context, renamed_parts); /// TODO Allow to use quorum here. ReplicatedMergeTreeSink output(*this, metadata_snapshot, 0, 0, 0, false, false, query_context, /*is_attach*/true); for (size_t i = 0; i < loaded_parts.size(); ++i) { const String old_name = loaded_parts[i]->name; output.writeExistingPart(loaded_parts[i]); renamed_parts.old_and_new_names[i].old_name.clear(); LOG_DEBUG(log, "Attached part {} as {}", old_name, loaded_parts[i]->name); results.push_back(PartitionCommandResultInfo{ .partition_id = loaded_parts[i]->info.partition_id, .part_name = loaded_parts[i]->name, .old_part_name = old_name, }); } #endif }