Elasticsearch version 8.0.0-rc2edit

Also see Breaking changes in 8.0.

Known issuesedit

  • Do not upgrade production clusters to Elasticsearch 8.0.0-rc2. Elasticsearch 8.0.0-rc2 is a pre-release of Elasticsearch 8.0 and is intended for testing purposes only.

    Upgrades from pre-release builds are not supported and could result in errors or data loss. If you upgrade from a released version, such as 7.16, to a pre-release version for testing, discard the contents of the cluster when you are done. Do not attempt to upgrade to the final 8.0 release.

Deprecationsedit

Engine
  • Deprecate setting max_merge_at_once_explicit #80574
Search
  • Configure IndexSearcher.maxClauseCount() based on node characteristics #81525 (issue: #46433)

New featuresedit

Snapshot/Restore
  • Support IAM roles for Kubernetes service accounts #81255 (issue: #52625)
Watcher
  • Use startsWith rather than exact matches for Watcher history template names #82396

Enhancementsedit

Cluster Coordination
ILM+SLM
  • Avoid unnecessary LifecycleExecutionState recalculation #81558 (issues: #77466, #79692)
  • Make unchanged ILM policy updates into no-op #82240 (issue: #82065)
Infra/Core
  • Prevent upgrades to 8.0 without first upgrading to the last 7.x release #82321 (issue: #81865)
Machine Learning
  • Add deployment_stats to trained model stats #80531
  • The setting use_auto_machine_memory_percent now defaults to max_model_memory_limit #80532 (issue: #80415)
Network
  • Improve slow inbound handling to include response type #80425
Packaging
  • Convert repository plugins to modules #81870 (issue: #81652)
Search
  • Check nested fields earlier in kNN search #80516 (issue: #78473)

Bug fixesedit

Autoscaling
  • Use adjusted total memory instead of total memory #80528 (issue: #78750)
Infra/Scripting
  • Fix duplicated allow lists upon script engine creation #82820 (issue: #82778)
Ingest
  • Adjust default geoip logging to be less verbose #81404 (issue: #81356)
Machine Learning
  • Check that total_definition_length is consistent before starting a deployment #80553
  • Fail inference processor more consistently on certain error types #81475
  • Optimize the job stats call to do fewer searches #82362 (issue: #82255)
Recovery
  • Make shard started response handling only return after the cluster state update completes #82790 (issue: #81628)
Search
  • Reject zero-length vectors when using cosine similarity #82241 (issue: #81167)
Security
  • Auto-generated TLS files under fixed config path #81547 (issue: #81057)
  • Bind to non-localhost for transport in some cases #82973
  • Correct file ownership on node reconfiguration #82789 (issue: #80990)
  • Display security auto-configuration with fancy unicode #82740 (issue: #82364)
Snapshot/Restore

Upgradesedit

Infra/Logging
  • Upgrade ECS logging layout to latest version #80500
Search
  • Upgrade to released lucene 9.0.0 #81426