Version 3.10.0 Release Notes
Compatible with OpenSearch and OpenSearch Dashboards version 3.10.0
Breaking Changes
Features
- Add
RestoreListenerPlugininterface, called after a snapshot restore finishes (#23295) - Add
limitparameter to_catAPIs for top-N row retrieval (#23439) - Add
GET /_cluster/settings/_describeAPI to report whether cluster settings are dynamic, with defaults, validation bounds, and descriptions (#23360), (#23400), (#23412), (#23456), (#23474) - Add cluster setting to enable/disable request and query cache cluster-wide (#23383)
- Add per-node request throttling to workload management (#22869)
- Add cluster-wide shared limit for workload management request throttling (#23264)
- Add administrative REST route metadata for authorization plugins (#22979)
- Expose registered action names to plugins via
DynamicActionRegistry(#23226) - Add
RoaringOrdinalsCollectorfor memory-efficient cardinality aggregation under multi-bucket parents (#21970) - Add object store source for pull-based ingestion (#23339)
- Add pluggable format seams for nested and flat_object field types (#23121)
- Compose additional index reader wrappers beneath the
setReaderWrapperslot (#23312) - Expose caller TLS session on Arrow Flight server channels (#23465)
- Extend
getFieldFilter()contract and make it available viaFieldFilterProviderandPluginComponentRegistry(#23292) - Enable full-copy snapshots for pluggable data format shards (#23155)
- Decouple shard snapshots from Lucene
IndexCommitusingCatalogSnapshot(#22930) - Multi-part parallel download of large segment files from remote store (#23016)
- Support pre-materialized get results in
_explain(#23110)
Enhancements
- Allow limiting total S3 API call time via
s3.client.<name>.api_call_timeout(#23185) - Make
telemetry-otelhistogram aggregation configurable (#23178) - Make
telemetry-otelservice.nameconfigurable viatelemetry.otel.service.name(#23114) - Skip
IndicesStatsfan-out in_cat/shardswhen only routing columns are requested (#23435) - Answer cluster settings describe on the receiving node instead of forwarding to cluster manager (#23426)
- Skip the token stream for the default keyword normalizer to reduce indexing CPU overhead (#23306)
- Use thread-local JSON recycler pool to fix indexing throughput collapse under concurrency (#23150)
- Restore search performance with cancellation enabled by preserving Lucene bulk postings operations (#23117)
- Forward bulk visit methods in
ExitableIntersectVisitorfor faster point queries with cancellation (#23201) - Avoid blocking keyed cache invalidation when a load is in flight (#22761)
- Remove redundant
synchronizedfromShardCoreKeyMap#getShardIdto reduce search contention (#22803) - Block
ptraceand process memory access syscalls in the system call filter (#23491) - Bound
min_hashfilter parameters to prevent OOM (#23362) - Mask secret values in reported JVM arguments (#23356)
- Ignore excluded nodes in zone awareness count so shards can drain off fully excluded zones (#22967)
- Push down ordered list limits in Azure blob listings (#23237)
- Add conditional blob writes for Azure Blob Storage (#23265)
- Configure Azure SDK read and write transfer options (#23239)
- Propagate Azure blob existence failures instead of silently returning false (#23266)
- Propagate metadata on synchronous remote translog blob uploads (#23274)
- Skip re-downloading remote translog generations already present locally (#23146)
- Batch primary translog appends to amortize per-operation overhead for remote store indices (#23224)
- Hydrate from the remote store outside the engine open lock to avoid blocking shard close (#23203)
- Remove deleted index data left behind by a held shard lock (#23393)
- Resolve primary operation policy on promotion for document replication (#23175)
- Add
_msearch/templatesub-request count limit and circuit-breaking logic to prevent OOM (#23320) - Let a second
./gradlew runadd nodes to the cluster the first one started (#23221) - Reject invalid values for
_catlimitparameter (#23453) - Count bulk visits in approximate range early termination (#23470)
- Describe cluster settings by prefix with wildcard support (#23474)
Bug Fixes
- Fix
RoutingMissingExceptionNPE when indexing with generated ID on routing-required index (#23096) - Fix
UnsupportedOperationExceptionon nested sort during early termination (#22534) - Fix
addEmptyBucketsfrom creating too many buckets when given large extended bounds (#17718) - Fix bulk indexing request corruption on shard backpressure rejection (#20727)
- Fix duplicate elements left by
CollectionUtils.sortAndDedup(#22445) - Accept zero term statistics in term vectors responses for terms not in the index (#23467)
- Fix stuck
primaryReplicaResyncInProgressflag blocking primary relocation (#23075) - Fix refreshed checkpoint stuck behind visible ops (#23195)
- Fix segment replication lag reported as Unix epoch when checkpoint lacks created timestamp (#23311)
- Fix shard indexing pressure limit never growing after idle (#23471)
- Exclude recovering replicas from segment replication lag metrics (#23123)
- Fix tracer heap leak:
TraceableTcpTransportChannelretains a close listener per request (#23119) - Fix translog UUID mismatch on pull-based ingestion index restart (#23090)
- Fix monitor mode workload group rejections (#22654)
- Fix phantom WLM rejections and stats for coordinator and shard search tasks (#22839)
- Fix workload group cancel interval being ignored on the coordinating node (#23449)
- Fix
WarmDiskThresholdDecidercanMovechecks to prevent unnecessary calculations (#23176) - Reject joins of nodes the cluster manager is no longer connected to (#23296)
- Reject out-of-range cluster setting updates at validation time instead of at apply time (#22655)
- Fix data loss and stalls in
segments_onlyremote store mode (#23288) - Fail shard when a snapshot hits a corrupt commit (#23492)
- Fail the shard copy in the local cluster state in
LocalShardStateActionfor clusterless mode (#23220) - Decode policy codebase URLs before canonicalizing to fix symlinked paths with spaces (#23546)
- Prevent stuck snapshot deletes and queue starvation with DELETE-based recovery (#23194)
- Fix auto force merge continuing after warm node removal (#23079)
- Store
match_only_textfields when derived source is enabled (#23357) - Handle partial routing tables in shard pagination in ingestion states API (#23498)
- Support segment replication in
segments_onlyremote store mode (#23098) - Fix WLM auto-tagging failure on rule-matching nested searches (#22932)
Maintenance
- Upgrade Lucene to 10.6.0 (#23475)
- Upgrade Lucene to 10.5.2 (#23310)
- Update Netty to 4.2.20 (#23489)
- Update Netty to 4.2.19.Final (#23245)
- Update OpenTelemetry to 1.68.0 and OpenTelemetry SemConv to 1.45.0 (#23590)
- Update OpenTelemetry to 1.67.0 (#23309)
- Update Apache HttpClient5 to 5.6.5 and HttpCore5 to 5.4.4 (#23335)
- Update Reactor Netty to 1.3.8 and Project Reactor to 3.8.8 (#23417)
- Update Bouncy Castle FIPS support libraries (#23233)
- Update bundled JDK to 25.0.5+8 (#23407)
- Update ASM to 9.11.1 (#23378)
- Update Jacoco and Byte Buddy (#23115)
- Update Mockito to 5.24.0 (#23135)
- Align commons-compress dependency versions (#23568)
- Remove usage of deprecated
ClientTlsStrategyBuilder::setTlsDetailsFactory(#23464)