Release notes

8.10.4-32.1

Upgraded to Elasticsearch 8.10.3.
Added a circuit-breaker mechanism that makes multi-conditional joins fail fast if the INDEX_JOIN strategy needs to perform many index lookups that require a long processing time.
- The maximum number of allowed lookups is controlled by a new dynamic setting (siren.planner.index_join.max_lookups). The default value is set to 20,000.

Added support for spatial joins using Elasticsearch range fields and their related operators within, includes, and contains.

Upgraded to Elasticsearch 8.8.2.
Upgraded to Apache Arrow 11.0.0.
Improved the error message sent to the client when trying to create a point in time (PIT) in virtual indices.
Adapted inner join to work with Adaptive Query Planner (AQP).
Implemented a normalization of Boolean expressions for AQP.
Improved the projections of fields by using Elasticsearch runtime mapping capabilities.
Increased the default queue_size of the worker thread pool to 10_000 to support large number of workers created by join tasks.
Improved the performance of spatial joins when using a single field of the parent set with lt, lte, gt, or gte operators.
Improved the memory configuration documentation.

Fixed the preservation of the thread context for queued data uploads.
Fixed an issue of wrongly cached PIT queries.
Fixed an issue where there was a need to explicitly specify the INDEX_JOIN strategy when performing a multi-conditional join on a single node cluster.
Fixed the calculation of debug information for the throughput value, which was previously based on the output data size but is now calculated using the input data size.
Fixed the distribution of keys during the hash partitioning.