3.8.0

Version 3.8.0 Release Notes

Compatible with OpenSearch and OpenSearch Dashboards version 3.8.0

Features

• Add API to modify a data stream's backing indices (POST /_data_stream/_modify) (#22487)
• Add multivalue_doc_count aggregation that returns the number of documents with two or more values (#20472)
• Add pull-based ingestion from Hive tables via new ingestion-hive plugin (#21700)
• Add support for HTTP/3 on the client side via JDK HttpClient (#21718)
• Add gRPC support for Point-in-Time (PIT) create and delete operations (#22105)
• Add gRPC API support for extra_field_values in bulk requests (#21907)
• Add cross-cluster streaming support to the remote cluster client via Arrow Flight (#22359)
• Add coordinator-side index-level field domain pruning before can_match for faster time-range queries (#21865)
• Add bulkscorer script processing interface in ScriptScore query for bulk scoring optimizations (#22423)
• Add process-wide ObjectInputFilter to reject Java deserialization by default, gated behind bootstrap.serial_filter setting (#22073)
• Add cluster-level default setting for delayed shard allocation timeout (cluster.routing.allocation.unassigned.node_left.delayed_timeout) (#22379)

Enhancements

• Add HTTP request body decompression (gzip/deflate) to reactor-netty4 transport (#22382)
• Add cross-setting validator to prevent cluster-manager crash loop when both balance factors are set to zero (#22391)
• Add support for double, int, and long types in ExtraFieldValue for gRPC bulk indexing (#22058)
• Add timing logs for repository create, update, and delete operations (#22489)
• Add parent action name to ProcessorGenerationContext for conditional system-generated processor logic (#22195)
• Add getter for remote translog transfer tracker to enable plugin access to tracking metrics (#22453)
• Deprecate RestClient in favor of the internal HTTP client (#22116)
• Disable Mustache partial template resolution in search templates (#22438)
• Use binary serialization for resource usage headers to reduce CPU overhead on the search path (#21230)
• Preserve resolved search pipeline id and inline pipeline source through query rewrite (#22501)
• Fork GET _aliases serialization to MANAGEMENT thread pool to avoid blocking transport threads (#21954)
• Strengthen scroll_id validation to prevent OOM from crafted scroll IDs (#22396)
• Add depth tracking to recursive deserialization to prevent StackOverflowError (#22404)
• Validate base_path in FsRepository to prevent path traversal outside path.repo (#22328)
• Add path boundary validation in resolveAnalyzerPath to prevent directory traversal (#22094)
• Register snapshot resilience settings (snapshot.repository.io_timeout, max_outstanding_ops, cleanup_stale_blobs) (#22516)
• Wire request_timeout into S3 sync client to prevent indefinite hangs on degraded stores (#22479)
• Separate internal ignore settings from user ignore settings during snapshot restore (#20494)
• Validate total_primary_shards_per_node on index templates against the cluster instead of index-local flag (#22203)
• Rollover checkBlock now scoped to write index only, skipping non-write alias members (#21838)

Bug Fixes

• Fix OpenSearchTimeoutException to return HTTP 504 instead of 500 (#22064)
• Fix ClassCastException on malformed mappings in create index to return 400 instead of 500 (#22371)
• Fix NPE for search-only indices without primaries in cluster allocation explain and bulk shard selection (#22097)
• Fix NPE in S3 multipart upload when provideStream() throws, and fix shared segment array corruption (#22309)
• Fix ReplicationCheckpoint.compareTo to return 0 for equal checkpoints, preventing TimSort crash during shard allocation (#22376)
• Fix negative os.cpu.percent on cgroup v2 containers by granting read access to /sys/fs/cgroup/cpu.stat (#22408)
• Fix snapshot listing failing on repositories containing legacy Elasticsearch snapshots (#22193)
• Fix default SSE type to send AES256 header, preventing silent snapshot failures on buckets requiring encryption (#22144)
• Fix default replica count resolution from node settings (opensearch.yml) (#22334)
• Fix filter rewrite crash with sub-aggregations in date_histogram when segment contains out-of-range documents (#22390)
• Fix V1 template merge regression that broke dynamic object inheritance for dotted field paths (#22515)
• Fix disable_objects array parsing to preserve dotted field names (#22127)
• Fix field resolution for dotted fields under disable_objects (#21929)
• Fix NestedQueryBuilder.visit() to recursively visit child query tree (#22196)
• Fix InputStream leak when loading hyphenation patterns in analysis-common (#22508)
• Fix _cat/nodes API showing negative values in cpu stats when perf counters are unavailable (#22074)
• Fix global ordinals to properly trip fielddata circuit breaker (#22129)
• Restore default max nesting depth to 1000 and decouple stream depth limit from XContent limit (#22486)
• Fix unbounded recursion in deserialization that can cause StackOverflowError (#22404)

Maintenance

• Bump bc-fips to 2.1.3 to fix CVE-2026-8149 (#22537)
• Upgrade Jackson 2 to 2.22.1 to resolve CVE-2026-54515 (#22476)
• Bump Hadoop from 3.4.2 to 3.5.0 to resolve CVE-2026-2332 (#22362)
• Update Jackson to 3.2.1 (#22497)
• Update Netty to 4.2.16.Final (#22403)
• Update OpenTelemetry to 1.63.0 (#22111)
• Update Project Reactor to 3.8.6 and Reactor Netty to 1.3.6 (#22059)
• Upgrade Lucene to 10.5.0 (#22322)
• Accommodate JDK-26 related changes with respect to HttpClient behavior (#22386)
• Remove OpenSearchYamlFactory / OpenSearchYamlParser after Jackson 3.2.0 update (#22147)