SDK Enterprise

Tarantool 2.11.9 -> 2.11.10

This is a bugfix release. It resolves 15 issues since 2.11.9 (r708) and additionally introduces one optional feature. It is disabled by default and has no effect unless explicitly enabled.

• The 2.x series is the previous stable branch; upgrading to 3.x is recommended.
• To upgrade from Tarantool 2.x to 3.x, see the upgrade procedure.

crud 1.7.5 -> 1.7.6

Fixed:

• A consistency violation during rebalancing: a request could be retried on a storage that did not own the bucket yet, so a read returned no data and a write went to the wrong replica set (#509).
• Requests are no longer retried blindly: a request is now retried only if the router has recovered from the error, and a retry never runs past the original timeout — previously, a request could take up to twice as long as requested.

tt-ee 2.13.0 -> 2.14.0

Added:

• tt connect: the \history command to display the last executed commands in the interactive console.

Changed:

• tt stop: preliminary interrupts the processes to enable parallel termination.

Fixed:

• tt log -f: possible line loss/duplication on rename, hanging after a watched log directory is removed, and lines written just as the file was read to the end not showing up until the next write.

cartridge 2.17.1 -> 2.18.1

This release spans versions 2.17.2, 2.17.3, 2.18.0, and 2.18.1.

Added:

• New option --vshard-bootstrap-timeout (env TARANTOOL_VSHARD_BOOTSTRAP_TIMEOUT) of the vshard-router role to configure the vshard bootstrap timeout. The default value remains 10 seconds, which may not be enough to distribute buckets across the storages on slow CI runners.
• Optional timeout argument to cartridge.test-helpers.Cluster:start, cartridge.test-helpers.Cluster:bootstrap, and cartridge.test-helpers.Cluster:bootstrap_vshard to extend the bootstrap HTTP request timeout.

Updated:

• vshard dependency from version 0.1.41 to 0.1.42.
• checks dependency from version 3.4.0 to 3.4.1.
• http dependency from version 1.9.0 to 1.9.1. This release includes infrastructure fixes that do not alter the module’s behavior.
• membership dependency from version 2.5.3 to 2.5.4.

Fixed:

• A race condition during instance shutdown where membership.leave() could execute before roles were stopped, causing errors.
• When box.ctl.promote() returns ER_INTERFERING_PROMOTE during failover, the retry now happens 3 times with a 1-second delay so the new master doesn’t get stuck in read-only mode.

kafka 1.6.14 -> 1.6.15

Added:

• The WITH_ZSTD build flag to enable zstd compression support. In the Enterprise SDK build, zstd support is enabled by default.

Changed:

• Bumped librdkafka to 2.15.1.

metrics 1.8.1 -> 1.8.3

This release spans versions 1.8.2 and 1.8.3: it introduces EmmyLua annotations and fixes the CI pipeline to run on external pull requests.

Added:

• EmmyLua analyzer support and annotations (#551).

Changed:

• Set config.checks to 'off' in default_config to avoid environment-dependent alerts (THP, readahead) in tests (#549).
• CI is now run on pull requests in addition to branch pushes; runs are skipped for pull requests from the base repository to avoid duplicates (#555).

Fixed:

• config_metrics_test: disable system alerts only if the config.checks option exists (instead of a Tarantool-version-based check), so the test passes on any Tarantool 3.x version (#553).

membership 2.5.3 -> 2.5.4

Fixed:

• A member’s payload could be lost permanently after leave() and a subsequent rejoin.

migrations 1.1.0 -> 1.2.1

This release spans versions 1.2.0 and 1.2.1.

Added:

• drop_sharding_key() helper to clean up _ddl_sharding_key after dropping a space.
• register_sharding_func() and drop_sharding_func() helpers (#59).

Fixed:

• When a replica set had no healthy leader candidate (its leader was disabled, down, or not yet elected during stateful failover), migrator.up() used to silently skip it and report success. Now every replica set is required to have a healthy leader candidate before migrating: if a replica set has no active leader yet or its leader is not healthy, the call fails so the caller can wait for the leader and retry.

checks 3.4.0 -> 3.4.1

Added:

• EmmyLua annotations (#68).

vshard 0.1.41 -> 0.1.42

VShard 0.1.42 is fully compatible with previous vshard versions.

Added:

• A backend for Tarantool’s recovery point manager: the vshard-router backend. It lets the manager periodically create cluster-consistent recovery points that can later be used for backup and point-in-time recovery of the whole VShard cluster. A new public router method, vshard.router.create_cluster_recovery_point(), creates a recovery point on every replica set master (#648). Only with tarantool 3.8.0+. No user-visible changes for Tarantool 2.11 users.

graphqlapi 0.0.11 -> 0.0.15

This release spans versions 0.0.12, 0.0.13, 0.0.14, and 0.0.15.

Changed:

• Support for ddl-ee was added in 0.0.12 and reverted in 0.0.13; the change did not persist to 0.0.15 (#55).

Fixed:

• Tarantool Enterprise detection in utils.get_tnt_version() by checking tarantool.package (#58).
• Flaky cluster.test_get_instances and cluster.test_get_replicaset_instances tests by waiting until all instances report healthy status before running assertions.
• utils.get_tnt_version() aborting with variable 'tarantool' is not declared in Cartridge applications, which broke Tarantool Enterprise detection (#61).

graphqlapi-helpers 0.0.11 -> 0.0.12

This release bumps dependencies and the test environment.

Added:

• New box.cfg options for Tarantool 2.11: auth_delay, auth_type, bootstrap_strategy, disable_guest, election_fencing_mode, metrics, password_enforce_digits, password_enforce_lowercase, password_enforce_specialchars, password_enforce_uppercase, password_history_length, password_lifetime_days, password_min_length.

Fixed:

• Account for the crud stats driver default changing to metrics.

metrics 1.7.0 -> 1.8.1

This release adds selector-based filtering for custom metrics. Custom metrics can now be associated with hierarchical selectors and controlled via metrics.set_filter() or metrics.cfg() include/exclude options. Unknown include/exclude entries are treated as custom selectors, while built-in metric group names keep the existing behavior.

Added:

• metrics.namespace() and metrics.set_filter() to mark custom collectors/callbacks with selectors and filter them at collection time.

Fixed:

• metrics.cfg{exclude = {'all'}} so it excludes custom metric selectors in addition to built-in metric groups.

cartridge 2.16.7 -> 2.17.1

Added:

• Support for Tarantool election_mode = 'manual' in Cartridge stateful failover, including single-instance replicasets.

• Added runtime Lua API helpers:

  • require('cartridge.lua-api.failover').switch_to_manual_election_mode()
  • require('cartridge.lua-api.failover').switch_to_off_election_mode()

  to migrate a stateful replicaset between Tarantool election_mode = 'off' and election_mode = 'manual'.

• Documentation for migrating a stateful replicaset to Tarantool manual election mode, including restart-based migration, runtime helpers, rollback, and fencing recommendations.

Changed:

• Updated vshard dependency to 0.1.41.

Fixed:

• A race condition during instance shutdown where membership.leave() could execute before roles were stopped, causing errors.
• When box.ctl.promote() returns ER_INTERFERING_PROMOTE during failover, retry promotion 3 times with a 1 second delay so the new master will not be stuck in read-only mode.
• A stateful failover race when adding a brand new replicaset: if the state provider has no appointment yet for this replicaset, failover.cfg() now falls back to topology.get_leaders_order(...)[1] (failover_priority[1]) for the initial appointment. This prevents the future leader from being switched to read_only=true and avoids deadlock during topology apply.

tt-ee 2.12.0 -> 2.13.0

This release adds cluster worker configuration management and fixes incorrect error reporting in tt start when directory permissions are insufficient.

Added:

• tt cluster worker publish: publish (upload) a worker configuration to etcd or Tarantool-based configuration storage.
• tt cluster worker show: display a worker configuration stored in etcd or Tarantool-based configuration storage.
• tt cluster worker delete: delete a worker configuration from etcd or Tarantool-based configuration storage.

Fixed:

• tt start: fixed a bug where an error did not appear when access rights to the var directive were insufficient.

vshard 0.1.40 -> 0.1.41

VShard 0.1.41 is fully compatible with the previous VShard versions.

Added:

• A new storage configuration option: rebalancer_bucket_send_timeout. It specifies the timeout in seconds for sending a single bucket during rebalancing. This can be used to limit how long a bucket may be unavailable for writes while it is being sent.

Changed:

• Improved rebalancer bucket selection: the rebalancer now prefers buckets that have no active RW requests on top of them.

Fixed:

• An issue where duplicated active buckets could appear due to delayed stray TCP #214.
• An issue where RW requests could break replication after a master switch during rebalancing #573.
• An issue where master switches during rebalancing could lead to duplicated active buckets #576.

Tarantool 2.11.8 -> 2.11.9

This is a bugfix release: 34 issues have been fixed since 2.11.8 (r702).

• The 2.x series is the previous stable branch; upgrading to 3.x is recommended.
• To upgrade from Tarantool 2.x to 3.x, see the upgrade procedure.

When upgrading the SDK, it is recommended to:

• before starting the upgrade, ensure that all replica sets are healthy and in a consistent state;
• perform the upgrade sequentially, verifying the cluster state after upgrading each component;
• do not run data rebalancing while passing through CRUD 1.7.0 until the upgrade is fully completed.

crud 1.6.1 -> 1.7.5

Added:

• crud.locate() to determine where a tuple is stored (memtx or vinyl). Works for spaces managed by the enterprise module cooler.
• crud.len now supports options: mode, balance, prefer_replica, request_timeout.
• Safe mode to prevent writing data to the wrong replica set during vshard rebalancing.
• Metric tnt_crud_router_cache_clear_ts to help properly disable safe mode in a cluster.
• Automatic switch to safe mode when rebalancing starts.
• Ability to manually switch back to fast mode (fast mode).
• Metric tnt_crud_storage_nil_bucket_id_compat_total to track operations performed without bucket_ref (compatibility mode with older routers).

Fixed:

• Read-only operations (get, select, pairs, count, min, max) are now executed via healthy replicas even if all master nodes in the cluster are unavailable.
• Storage compatibility with routers < 1.7.0: bucket_id = nil is now handled correctly in get, update, and delete. In this case, storage skips bucket referencing and logs a rate-limited warning about reduced rebalancing safety during rolling upgrades.
• bucket_ref errors in crud.*_many methods are now returned as an array.
• bucket_unref was moved out of the transaction.
• Prevented duplicate metrics from being created on repeated init calls.
• Prevented duplicate triggers on the _crud_settings_local space on repeated init calls.
• A deadlock in crud.schema() after a schema reload error.
• Removed metric tnt_crud_storage_safe_mode_enabled from the router.
• Removed wrap_box_space_func_result wrapper to reduce allocations and speed up storage calls.
• Optimize crud.select() and crud.pairs() pagination with after cursor by using native after option on Tarantool 2.10+ for O(1) cursor positioning.

Changed:

• When switching to safe mode, the practice of marking/stopping iproto fibers in fast mode was discontinued; operation correctness on storage is validated via yield_checks in tests.
• Switching to safe mode was moved from the on_commit trigger to on_replace.
• Vinyl spaces always operate in safe mode.

vshard 0.1.37 -> 0.1.40

Version 0.1.40 is fully compatible with previous vshard versions.

Added:

• Ability to disable the log rate limiter via the consts module.
• Calling vshard.router.info() and vshard.storage.info() is now allowed even when the router or storage is disabled. The functions now also return a new boolean field: is_enabled.
• Improved logging for rebalancer and recovery related activities.

Fixed:

• An issue where the old master node could not discover the new master instance within a replica set.
• Connection leak: connections were not released by the garbage collector after reconfiguration or reload.
• Transaction limitation when working with _bucket: previously, the on_commit trigger on _bucket blocked writes to other spaces within the same transaction (for example, from on_replace triggers). Such scenarios are now allowed: on_commit skips changes related to “foreign” spaces.
• An issue where a user error was masked by the Transaction is active... error when calling a persistent function that throws an error and does not close a transaction on Tarantool versions earlier than 3.0.0-beta1-18. Now, vshard automatically closes such transactions and returns the original user error.
• An issue where a connection to a replica was not automatically restored when it was closed either by vshard or by the user.

metrics 1.6.2 -> 1.7.0

• graphite: added support for sending metrics to multiple servers.
• Removing a replica via box.space._cluster:delete() does not remove that replica’s information from metrics; it disappears only after a cluster restart.
• Backward compatibility with the previous plugin version is preserved.
• Behavior changes:
  • init now assigns a unique name to the created fiber based on the input graphite server options (if provided).
  • Added stop() to stop all fibers started by the plugin.

tt-ee v2.11.0 -> v2.12.0

Added:

• tt pack: added support for nested .packignore files in the root of a tt environment.
• tt status: added the --format option to output status in JSON and YAML formats (machine-readable output).

Changed:

• tt export: changed the default behavior for compound fields (arrays and maps): they are now exported in JSON format by default. To restore the previous behavior, use --compound-value-format=ignore.

Fixed:

• Integrity checking for an application using the Cartridge directory layout (a single application whose root directory is the environment root).
• An issue with Tarantool 3.5+: the instance did not stop when the periodic integrity check failed.
• Minor fixes identified by the Svacer static analyzer and CVE scanners.

cartridge 2.16.4 -> 2.16.7

Changed:

• Do not expand the file tree by default on the code page.
• Update vshard dependency to 0.1.40.
• Update membership dependency to 2.5.3.
• Update cartridge-metrics-role dependency to 0.1.3.
• Update graphql dependency to 0.3.1.
• Update http dependency to 1.9.0.

Fixed:

• Refactor synchronous spaces monitoring to consider the actual failover mode. Sync spaces warning is now logged when failover is configured in a mode that doesn’t support them (eventual, stateful without synchro_mode), instead of unconditionally at instance startup.
• Added is_sync_spaces_supported() function to cartridge.failover module.
• Sync spaces are now detected dynamically, allowing detection of spaces added at runtime.

http 1.8.0 -> 1.9.0

The release introduces a new ssl_verify_client option and changes default behavior with provided ca_file parameter. Also a few bugs were fixed.

Added:

• New ssl_verify_client option.

Fixed:

• Do not recreate server if it’s address and port were not changed.
• Server doesn’t change after updating parameters on config reload.

Breaking change:

Mutual TLS with ca_file option enabled by default.

Module http documentation.

Breaking changes

• Default audit log format was changed to CSV.

Functionality added or changed

Bugs fixed