Release notes

Fixed a bug in AQP with nested joins when a child join was pruned by the planner.

Fixed a bug in multi-conditional index join swapping when the parent data was empty.

Fixed a regression in Searchguard FLX when calculating the index fingerprint.

Fixed failures in join queries with match queries on semantic_text fields.

Fixed a stack overflow in the graph API when a path pattern led to an unreachable entity.

Fixed a stack overflow when processing graph queries with two patterns.

Fixed inner-join strategy when the root var is not the first node pattern.

Fixed an issue on join strategy selection for graph requests.

Fixed an ArrayIndexOutOfBoundsException in INDEX_JOIN strategy when joining on numeric fields.

Fixed an issue where the msearch request ignored the preference parameter when at least one sub-request contained a join.

Fixed an issue where the preference parameter contained a non-existent shard id, causing the request to fail instead of returning an empty response.

Fixed rolling-upgrade serialization in broadcast requests.

Fixed an issue with query name and boost handling in KNN queries.

Fixed a race condition in search locks for graph requests.

Fixed a race condition in the data store.

Fixed root selection in graph requests when the Jordan Center strategy could not find a center.

Fixed label-combination exploration while backtracking in graph pattern matching.

Fixed a race condition between task resource release and job allocator cleanup.

Fixed scoring inconsistencies in dis_max queries when min_score was used with tie_breaker.

Fixed a bug in unary-expression hashing that could cause incorrect cached plan reuse for equivalent-looking queries.

Added support for LIMIT clause in graph API queries.

Added ANY direction in edge patterns in graph API queries.

Added support for disjunction of labels in graph API queries.

Added support for disjunctions in path patterns.

Added support for early termination in AQP so tasks can complete as soon as downstream evaluation proves there are no results.

Added support for task cancellation in graph requests.

Added node labels to the default vertex output.

Added support for element property specifications in graph requests.

Added a new graph execution pipeline that improved scalability and provided a stronger foundation for future graph features.

Upgraded to Elasticsearch 9.4.5

Improved the performance of the graph API during label expansion.

Improved performance on multi-condition index join when using EQ operator on keyword fields.

Improved graph result-fetch performance by removing the use of named queries.

Improved memory usage in the graph API when rendering final results.

Improved the memory settings documentation section.

Improved data lifecycle by borrowing vector container per tasks.

Improved the search-lock lifecycle for graph requests.

Added cancel-aware graph expansion handling so resources are released as soon as they are no longer required.

Added graph-pattern pruning to skip zero-result branches earlier, reducing unnecessary processing.

Added a fail-fast approach when the data model was not available.

Added job throttling for graph API requests to limit resource overuse.

Improved the cost optimizer for KNN requests by using KNN queries during cardinality estimation.

Improved creation of table rows with a streaming Cartesian Product computation.

Updated Apache Arrow to version 18.3.0.

Updated Apache Calcite to version 1.42.

Improved code stability by removing reflection on Elasticsearch internal API.

Added a documentation example of runtime fields accessing nested fields.

Clarified the license requirements for the graph API in the documentation.