You are looking at preliminary documentation for a future release. Not what you want? See the current release documentation.
- Forbid granting the all permission in production #27548
Breaking Java Changesedit
- Machine Learning
- Added the ability to identify scheduled events and prevent anomaly detection during these periods. For more information, see Calendars and Scheduled Events.
- Plugin Ingest GeoIp
- Plugin Lang Painless
- Rank Evaluation
- Reindex API
- X-Pack security now supports user authentication using SAML Single Sign on. For more information, see SAML authentication.
- Added a transform input for chained input. For more information, see Transforming Chained Input Data.
- Circuit Breakers
- Introduce elasticsearch-core jar #28191 (issue: #27933)
- Rename core module to server #28190 (issue: #27933)
- Rename core module to server #28180 (issue: #27933)
- Introduce elasticsearch-core jar #28178 (issue: #27933)
- Add Writeable.Reader support to TransportResponseHandler #28010 (issue: #26315)
- Simplify rejected execution exception #27664 (issue: #27663)
- Add node name to thread pool executor name #27663 (issues: #26007, #26835)
- Java API
- Java High Level REST Client
- add toString implementation for UpdateRequest. #27997 (issue: #27986)
- Add Close Index API to the high level REST client #27734 (issue: #27205)
- Add Open Index API to the high level REST client #27574 (issue: #27205)
- Added Create Index support to high-level REST client #27351 (issue: #27205)
- Add multi get api to the high level rest client #27337 (issue: #27205)
- Add msearch api to high level client #27274
- Machine Learning
- Increased tokenization flexibility for categorization. Now all Elasticsearch analyzer functionality is available, which opens up the possibility of sensibly categorizing non-English log messages. For more information, see Customizing the Categorization Analyzer.
- Improved the sensitivity of the analysis to high variance data with lots of values near zero.
- Improved the decay rate of the model memory by using a weighted moving average.
Machine learning indices created after upgrading to 6.2 have the
auto_expand_replicas: 0-1setting rather than a fixed setting of 1 replica. As a result, machine learning indices created after upgrading to 6.2 can have a green status on single node clusters. There is no impact in multi-node clusters.
- Changed the credentials that are used by datafeeds. When X-Pack security is enabled, a datafeed stores the roles of the user who created or updated the datafeed at that time. This means that if those roles are updated, the datafeed subsequently runs with the new permissions that are associated with the roles. However, if the user’s roles are adjusted after creating or updating the datafeed then the datafeed continues to run with the permissions that are associated with the original roles. For more information, see Datafeeds.
Added a new
scheduledforecast status, which indicates that the forecast has not started yet.
X-Pack monitoring indices (
.monitoring) created after upgrading to 6.2 have the
auto_expand_replicas: 0-1setting rather than a fixed setting of 1 replica. As a result, monitoring indices created after upgrading to 6.2 can have a green status on single node clusters. There is no impact in multi-node clusters.
- Added a cluster alert that triggers whenever a node is added, removed, or restarted.
- X-Pack monitoring indices (
- Add NioGroup for use in different transports #27737 (issue: #27260)
- Add read timeouts to http module #27713
Implement byte array reusage in
NioTransport#27696 (issue: #27563)
- Introduce resizable inbound byte buffer #27551 (issue: #27563)
- Decouple nio constructs from the tcp transport #27484 (issue: #27260)
- also extract match_all queries when indexing percolator queries #27585
- Plugin Lang Painless
- Rank Evaluation
- Simplify RankEvalResponse output #28266
- Truncate tlog cli should assign global checkpoint #28192 (issue: #28181)
- Replica starts peer recovery with safe commit #28181 (issue: #10708)
- Primary send safe commit in file-based recovery #28038 (issue: #10708)
- Fail resync-failed shards in subsequent writes #28005
- Introduce promoting index shard state #28004 (issue: #24841)
- Non-peer recovery should set the global checkpoint #27965
- Persist global checkpoint when finalizing a peer recovery #27947 (issue: #27861)
- Rollback a primary before recovering from translog #27804 (issue: #10708)
- Added the ability to refresh tokens that were created by the token API. The API provides information about a refresh token, which you can use within 24 hours of its creation to extend the life of a token. For more information, see Token Management APIs.
Added principal and role information to
run_as_deniedaudit events. For more information about these events, see Auditing Security Events.
- Added audit event ignore policies, which are a way to tune the verbosity of an audit trail. These policies define rules for ignoring audit events that match specific attribute values. For more information, see Logfile Audit Events Ignore Policies.
- Added a certificates API, which enables you to retrieve information about the X.509 certificates that are used to encrypt communications in your Elasticsearch cluster. For more information, see SSL Certificate API.
- Sequence IDs
- Do not keep 5.x commits when having 6.x commits #28188 (issues: #27606, #28038)
- Use lastSyncedGlobalCheckpoint in deletion policy #27826 (issue: #27606)
- Use CountedBitSet in LocalCheckpointTracker #27793
- Only fsync global checkpoint if needed #27652
- Keep commits and translog up to the global checkpoint #27606
- Adjust CombinedDeletionPolicy for multiple commits #27456 (issues: #10708, #27367)
- Keeps index commits up to the current global checkpoint #27367 (issue: #10708)
- Dedup translog operations by reading in reverse #27268 (issue: #10708)
- Add validation of keystore setting names #27626
- Task Manager
Added the ability to set the
doc_typedynamically in an index action. For more information, see Index Action.
refreshindex action attribute, which enables you to set the refresh policy of the write request. For more information, see Index Action.
- Added support for actions in slack attachments, which enables you to add buttons that can be clicked in slack messages. For more information, see Slack Action.
Watcher indices (
triggered_watches) created after upgrading to 6.2 have the
auto_expand_replicas: 0-1setting rather than a fixed setting of 1 replica. As a result, Watcher indices created after upgrading to 6.2 can have a green status on single node clusters. There is no impact in multi-node clusters.
- Added the ability to set the
- Adds metadata to rewritten aggregations #28185 (issue: #28170)
- Fix NPE on composite aggregation with sub-aggregations that need scores #28129
- StringTerms.Bucket.getKeyAsNumber detection type #28118 (issue: #28012)
- Fix incorrect results for aggregations nested under a nested aggregation #27946 (issue: #27912)
- Fix global aggregation that requires breadth first and scores #27942 (issues: #22321, #27928)
- Fix composite aggregation when after term is missing in the shard #27936
- Fix preserving FiltersAggregationBuilder#keyed field on rewrite #27900 (issue: #27841)
- Using DocValueFormat::parseBytesRef for parsing missing value parameter #27855 (issue: #27788)
Fix illegal cast of the "low cardinality" optimization of the
- Always include the _index and _id for nested search hits. #27201 (issue: #27053)
- Do not open indices with broken settings #26995
- Fix lock accounting in releasable lock #28202
- Fixes ByteSizeValue to serialise correctly #27702 (issue: #27568)
- Do not set data paths on no local storage required #27587 (issue: #27572)
- Ensure threadcontext is preserved when refresh listeners are invoked #27565
- Ensure logging is configured for CLI commands #27523 (issue: #27521)
- Fix highlighting on a keyword field that defines a normalizer #27604
- Inner Hits
- Java REST Client
- Machine Learning
- Improved error reporting for crashes and resource problems on Linux.
- Improved the detection of seasonal trends in bucket spans longer than 1 hour.
- Updated the forecast API to wait for validation and return an error if the validation fails.
- Set the actual bucket value to 0 in model plots for empty buckets for count and sum functions. The count and sum functions treat empty buckets as 0 rather than unknown for anomaly detection, so it was inconsistent not to do the same for model plots. This inconsistency resulted in problems plotting these buckets in Kibana.
- Only bind loopback addresses when binding to local #28029 (issue: #1877)
- Remove potential nio selector leak #27825
- Fix issue where the incorrect buffers are written #27695 (issue: #27551)
- Throw UOE from compressible bytes stream reset #27564 (issue: #24927)
- Bubble exceptions when closing compressible streams #27542 (issue: #27540)
- Plugin Analysis ICU
- Catch InvalidPathException in IcuCollationTokenFilterFactory #27202
- Plugin Analysis Phonetic
- Plugin Lang Painless
- Plugin Repository HDFS
- Fix SecurityException when HDFS Repository used against HA Namenodes #27196
- Open engine should keep only starting commit #28228 (issues: #27804, #28181)
- Allow shrinking of indices from a previous major #28076 (issue: #28061)
- Set global checkpoint before open engine from store #27972 (issues: #27965, #27970)
- Check and repair index under the store metadata lock #27768 (issues: #24481, #24787, #27731)
- Flush old indices on primary promotion and relocation #27580 (issue: #27536)
- Ensure we protect Collections obtained from scripts from self-referencing #28335
- Fix simple_query_string on invalid input #28219 (issue: #28204)
- Use the underlying connection version for CCS connections #28093
- Fix synonym phrase query expansion for cross_fields parsing #28045
- Carry forward weights, etc on rescore rewrite #27981 (issue: #27979)
- Fix routing with leading or trailing whitespace #27712 (issue: #27708)
setup-passwordscommand to generate passwords with characters
0-9, so that they are safe to use in shell scripts. For more information about this command, see elasticsearch-setup-passwords.
Improved the error messages that occur if the
x-packdirectory is missing when you run
- Fixed the ordering of realms in a realm chain, which determines the order in which the realms are consulted. For more information, see Realms.
- Updated the
- Sequence IDs
- Settings: Introduce settings updater for a list of settings #28338 (issue: #28047)
- Fix setting notification for complex setting (affixMap settings) that could cause transient settings to be ignored #28317 (issue: #28316)
- Fix environment variable substitutions in list setting #28106 (issue: #27926)
- Allow index settings to be reset by wildcards #27671 (issue: #27537)
- Consistent updates of IndexShardSnapshotStatus #28130 (issue: #26480)
- Avoid concurrent snapshot finalizations when deleting an INIT snapshot #28078 (issues: #27214, #27931, #27974)
- Do not start snapshots that are deleted during initialization #27931
- Do not swallow exception in ChecksumBlobStoreFormat.writeAtomic() #27597
- Consistent update of stage and failure message in IndexShardSnapshotStatus #27557 (issue: #26480)
- Fail restore when the shard allocations max retries count is reached #27493 (issue: #26865)
- Delete shard store files before restoring a snapshot #27476 (issues: #20220, #26865)
- Term Vectors
- Replaced group settings with affix key settings where filters are needed. For more information, see https://github.com/elastic/elasticsearch/pull/28338.