# This suite is intended to exercise transaction behavior with failovers by running existing # concurrency workloads with state functions wrapped in transactions while shard and config # primaries are being killed. State functions that use operations not supported in a transaction # are not wrapped in one. test_kind: fsm_workload_test selector: roots: - jstests/concurrency/fsm_workloads/**/*.js exclude_files: ## # Blacklists from concurrency_sharded_replication ## # SERVER-13116 distinct isn't sharding aware - jstests/concurrency/fsm_workloads/distinct.js - jstests/concurrency/fsm_workloads/distinct_noindex.js - jstests/concurrency/fsm_workloads/distinct_projection.js # SERVER-17397 Drops of sharded namespaces may not fully succeed - jstests/concurrency/fsm_workloads/create_database.js - jstests/concurrency/fsm_workloads/drop_database.js # Disabled due to SERVER-33753, '.count() without a predicate can be wrong on sharded # collections'. This bug is problematic for these workloads because they assert on count() # values: - jstests/concurrency/fsm_workloads/agg_match.js # $lookup and $graphLookup are not supported on sharded collections. - jstests/concurrency/fsm_workloads/agg_graph_lookup.js - jstests/concurrency/fsm_workloads/view_catalog_cycle_lookup.js # Disabled due to SERVER-20057, 'Concurrent, sharded mapReduces can fail when temporary # namespaces collide across mongos processes' - jstests/concurrency/fsm_workloads/map_reduce_drop.js - jstests/concurrency/fsm_workloads/map_reduce_inline.js - jstests/concurrency/fsm_workloads/map_reduce_interrupt.js - jstests/concurrency/fsm_workloads/map_reduce_merge.js - jstests/concurrency/fsm_workloads/map_reduce_merge_nonatomic.js - jstests/concurrency/fsm_workloads/map_reduce_reduce.js - jstests/concurrency/fsm_workloads/map_reduce_reduce_nonatomic.js - jstests/concurrency/fsm_workloads/map_reduce_replace.js - jstests/concurrency/fsm_workloads/map_reduce_replace_nonexistent.js - jstests/concurrency/fsm_workloads/map_reduce_replace_remove.js # Disabled due to MongoDB restrictions and/or workload restrictions # These workloads sometimes trigger 'Could not lock auth data update lock' # errors because the AuthorizationManager currently waits for only five # seconds to acquire the lock for authorization documents - jstests/concurrency/fsm_workloads/auth_create_role.js - jstests/concurrency/fsm_workloads/auth_create_user.js - jstests/concurrency/fsm_workloads/auth_drop_role.js - jstests/concurrency/fsm_workloads/auth_drop_user.js # uses >100MB of data, which can overwhelm test hosts - jstests/concurrency/fsm_workloads/agg_group_external.js - jstests/concurrency/fsm_workloads/agg_sort_external.js # compact can only be run against a standalone mongod - jstests/concurrency/fsm_workloads/compact.js - jstests/concurrency/fsm_workloads/compact_while_creating_indexes.js # convertToCapped can't be run on mongos processes - jstests/concurrency/fsm_workloads/convert_to_capped_collection.js - jstests/concurrency/fsm_workloads/convert_to_capped_collection_index.js # findAndModify requires a shard key - jstests/concurrency/fsm_workloads/findAndModify_mixed_queue_unindexed.js - jstests/concurrency/fsm_workloads/findAndModify_remove_queue_unindexed.js - jstests/concurrency/fsm_workloads/findAndModify_update_collscan.js - jstests/concurrency/fsm_workloads/findAndModify_update_queue.js - jstests/concurrency/fsm_workloads/findAndModify_update_queue_unindexed.js # remove cannot be {} for findAndModify - jstests/concurrency/fsm_workloads/findAndModify_remove_queue.js # can cause OOM kills on test hosts - jstests/concurrency/fsm_workloads/findAndModify_update_grow.js # cannot ensureIndex after dropDatabase without sharding first - jstests/concurrency/fsm_workloads/plan_cache_drop_database.js # reIndex is not supported in mongos. - jstests/concurrency/fsm_workloads/reindex.js - jstests/concurrency/fsm_workloads/reindex_background.js # The WTWriteConflictException failpoint is not supported on mongos. - jstests/concurrency/fsm_workloads/collmod_writeconflict.js - jstests/concurrency/fsm_workloads/reindex_writeconflict.js # our .remove(query, {justOne: true}) calls lack shard keys - jstests/concurrency/fsm_workloads/remove_single_document.js # The rename_* workloads are disabled since renameCollection doesn't work with sharded # collections - jstests/concurrency/fsm_workloads/rename_capped_collection_chain.js - jstests/concurrency/fsm_workloads/rename_capped_collection_dbname_chain.js - jstests/concurrency/fsm_workloads/rename_capped_collection_dbname_droptarget.js - jstests/concurrency/fsm_workloads/rename_capped_collection_droptarget.js - jstests/concurrency/fsm_workloads/rename_collection_chain.js - jstests/concurrency/fsm_workloads/rename_collection_dbname_chain.js - jstests/concurrency/fsm_workloads/rename_collection_dbname_droptarget.js - jstests/concurrency/fsm_workloads/rename_collection_droptarget.js # cannot use upsert command with $where with sharded collections - jstests/concurrency/fsm_workloads/upsert_where.js # stagedebug can only be run against a standalone mongod - jstests/concurrency/fsm_workloads/yield_and_hashed.js - jstests/concurrency/fsm_workloads/yield_and_sorted.js # TODO Unblacklist (SERVER-35538). - jstests/concurrency/fsm_workloads/database_versioning.js # TODO Unblacklist (SERVER-38852). - jstests/concurrency/fsm_workloads/agg_out_interrupt_cleanup.js # serverStatus does not include transaction metrics on mongos. - jstests/concurrency/fsm_workloads/multi_statement_transaction_atomicity_isolation_metrics_test.js # Uses the same transaction id across different routers, which is not allowed because when either # router tries to commit, it may not know the full participant list. - jstests/concurrency/fsm_workloads/multi_statement_transaction_all_commands_same_session.js ## # Blacklists from concurrency_replication_multi_stmt_txn ## # Relies on having one thread observe writes from the other threads, which won't become visible # once a transaction in the thread is started because it'll keep reading from the same snapshot. - jstests/concurrency/fsm_workloads/create_index_background.js - jstests/concurrency/fsm_workloads/create_index_background_partial_filter.js - jstests/concurrency/fsm_workloads/create_index_background_wildcard.js # Expects reads to die with a particular error, but other errors are possible if the read is part # of a transaction (e.g. ErrorCodes.LockTimeout). - jstests/concurrency/fsm_workloads/drop_index_during_replan.js # Performs direct writes to system.views - jstests/concurrency/fsm_workloads/view_catalog_direct_system_writes.js ## # Blacklists from concurrency_sharded_multi_stmt_txn ## # Use updates that do not contain the shard key, so they are rejected before any commands are # sent to participant shards, but these workloads do not fail, so the auto retry transaction logic # attempts to commit, which fails because no participants have been contacted. - jstests/concurrency/fsm_workloads/update_rename.js - jstests/concurrency/fsm_workloads/update_rename_noindex.js # Expects DocumentValidationFailure which causes transaction to abort and retry indefinitely. - jstests/concurrency/fsm_workloads/schema_validator_with_expr_variables.js ## # Blacklists from concurrency_sharded_with_stepdowns ## # ChunkHelper directly talks to the config servers and doesn't support retries for network errors - jstests/concurrency/fsm_workloads/cleanupOrphanedWhileMigrating.js - jstests/concurrency/fsm_workloads/sharded_base_partitioned.js - jstests/concurrency/fsm_workloads/sharded_mergeChunks_partitioned.js - jstests/concurrency/fsm_workloads/sharded_moveChunk_drop_shard_key_index.js - jstests/concurrency/fsm_workloads/sharded_moveChunk_partitioned.js - jstests/concurrency/fsm_workloads/sharded_splitChunk_partitioned.js # These workloads frequently time out waiting for the distributed lock to drop a sharded # collection. - jstests/concurrency/fsm_workloads/kill_aggregation.js - jstests/concurrency/fsm_workloads/kill_rooted_or.js # Uses non retryable commands. - jstests/concurrency/fsm_workloads/agg_out.js - jstests/concurrency/fsm_workloads/agg_out_interrupt_cleanup.js - jstests/concurrency/fsm_workloads/agg_sort.js - jstests/concurrency/fsm_workloads/collmod.js - jstests/concurrency/fsm_workloads/collmod_separate_collections.js - jstests/concurrency/fsm_workloads/invalidated_cursors.js - jstests/concurrency/fsm_workloads/kill_multicollection_aggregation.js - jstests/concurrency/fsm_workloads/view_catalog.js - jstests/concurrency/fsm_workloads/view_catalog_cycle_with_drop.js - jstests/concurrency/fsm_workloads/view_catalog_direct_system_writes.js ## # Blacklists specific to this suite ## # Uses getMore in the same state function as a command not supported in a transaction. - jstests/concurrency/fsm_workloads/list_indexes.js - jstests/concurrency/fsm_workloads/create_capped_collection.js - jstests/concurrency/fsm_workloads/create_capped_collection_maxdocs.js # Uses non-retryable commands in the same state function as a command not supported in a # transaction. - jstests/concurrency/fsm_workloads/agg_merge_when_not_matched_insert.js - jstests/concurrency/fsm_workloads/agg_merge_when_matched_replace_with_new.js # TODO SERVER-40713 moveChunk is not considered retryable by the network retry override. - jstests/concurrency/fsm_workloads/agg_with_chunk_migrations.js - jstests/concurrency/fsm_workloads/random_moveChunk_broadcast_delete_transaction.js - jstests/concurrency/fsm_workloads/random_moveChunk_broadcast_update_transaction.js # TODO SERVER-40493: JS engine interruptions on mongos return ErrorCodes::Interrupted, which isn't # considered a retryable or transient transaction error. - jstests/concurrency/fsm_workloads/indexed_insert_where.js - jstests/concurrency/fsm_workloads/remove_where.js - jstests/concurrency/fsm_workloads/update_where.js # TODO SERVER-40705: Stepdown with prepared transaction that performed text search triggers # invariant - jstests/concurrency/fsm_workloads/indexed_insert_text.js - jstests/concurrency/fsm_workloads/indexed_insert_text_multikey.js exclude_with_any_tags: - requires_replication # The touch command does not exist on mongos, which is enforced before session information is # parsed, causing state functions in these workloads to fail before starting a transaction despite # sending a command with startTransaction=true, leading to an infinite loop in the auto # transaction retry override. - requires_touch # Sharing cursors between state functions will fail in this suite because it will attempt to use # the same cursor in multiple transactions. - state_functions_share_cursor # These start a transaction in one state function and use it in other state functions. This suite # would instead execute each state function as its own transaction. - state_functions_share_transaction # Note that "requires_non_retryable_writes" does not need to be blacklisted because some writes # that are not individually retryable can be retried if they are part of a transaction. executor: archive: hooks: - CheckReplDBHash - ValidateCollections tests: true config: shell_options: readMode: commands global_vars: TestData: runInsideTransaction: true runningWithAutoSplit: false runningWithBalancer: false runningWithConfigStepdowns: true runningWithSessions: true runningWithShardStepdowns: true traceExceptions: false useStepdownPermittedFile: true usingReplicaSetShards: true hooks: # We use a stepdown interval of 15 seconds because we will retry all commands in a transaction # so we need to allow time for at most 10 operations to be re-run and then re-committed. If # too many network errors occur when re-running a transaction we will run out of retries. - class: ContinuousStepdown config_stepdown: true kill: true shard_stepdown: true stepdown_interval_ms: 15000 use_stepdown_permitted_file: true wait_for_mongos_retarget: true - class: CheckReplDBHash - class: ValidateCollections - class: CleanupConcurrencyWorkloads fixture: class: ShardedClusterFixture enable_balancer: false enable_autosplit: false mongos_options: set_parameters: enableTestCommands: 1 configsvr_options: num_nodes: 3 all_nodes_electable: true replset_config_options: settings: catchUpTimeoutMillis: 0 voting_secondaries: true shard_options: all_nodes_electable: true mongod_options: oplogSize: 1024 replset_config_options: settings: catchUpTimeoutMillis: 0 voting_secondaries: true mongod_options: set_parameters: enableTestCommands: 1 enableElectionHandoff: 0 num_rs_nodes_per_shard: 3 num_shards: 2 num_mongos: 2