Release 1.2.0-rc3

Bug fixes / Improvements

• Fixed a crash caused by lazy expiration during active search commands by using VALKEYMODULE_OPEN_KEY_NOEXPIRE and explicit expiration checks.
• Index Creation Validation: Added check for the limit of 64 text fields per index.
• Timeout Handling: Added timeout logic in aggregation stages and added comprehensive timeout tests for FT.AGGREGATE and all FT.SEARCH execution paths.
• Removed stemmed word length restrictions to support cases where a stem root can be longer than the original word.
• Support was added for the Valkey 9 CME Multi-DB feature by updating the select DB logic during RDB Load.
• Fixed GetSerializationRange logic for Aggregation

Optimizations

• Concurrent Indexing: Implemented a sharded PostingsMutexPool and utilizing Reader-Writer tree locks to improve throughput for parallel writes.
• Lock Contention: Reduced lock contention on per_key_text_indexes_ during query evaluation.
• Stemming Performance: Achieved up to 25-42% peak improvement for Snowball stemming by optimizing compiler flags.
• Tokenization improvements to reduce allocations during normalization, stemming, etc.
• Switched PositionMap to absl::btree_map, stem variants storage to vector and temporary stem mappings to InlinedVector for improved cache locality and reduced memory overhead.
• Optimized raxMutate to complete mutation operations in a single tree traversal.
• Resource Control: Introduced search.max-nonvector-search-results-fetched to limit ephemeral memory usage during non vector query searches.
• Shard Efficiency: Introduced a development (usable in debug mode) fanout-data-uniformity configuration to reduce cross shard network traffic by optimizing fanout fetching.

Full Changelog: 1.2.0-rc2...1.2.0-rc3

Release 1.2.0-rc2

Bug fixes

• Fix mutation sequence number timing bug to prevent race conditions with key deletions
• Fix to ensure an RDB from a search module loaded valkey-server (after deleting the pre-existing indexes) can be loaded onto the valkey servers without the search module
• Add null check for thread_monitor_ to prevent crashes in GetAvgCPUPercentage
• Fix unhandled text path in PreFilter Evaluators for complex OR queries
• Remove unnecessary allocation in text search operations

Optimizations

• Replace linear scan with min-heap approach in TermIterator for faster term searches
• Optimize FlatPositionMap::ReadCounts by replacing memcpy with loop-based byte reading
• Implement Size Estimation of Text Predicate key results size to help with the decision of pre-filtering vs inline-filtering in vector queries

Full Changelog: 1.2.0-rc1...1.2.0-rc2

Release 1.2.0-rc1

Note: Like valkey-search 1.1, this valkey-search 1.2 release also requires Valkey core 9.0.1 or higher.

Major New Features / Updates

• Adding text search capability. This includes updates to the FT.CREATE, FT.SEARCH, FT.AGGREGATE, Ingestion, etc.
  • Refer to documentation on how to use the feature: https://valkey.io/topics/search/
• Support for SKIPINITIALSCAN option on the FT.CREATE command
• Single Slot Query support
• Added support for SORTBY functionality on the FT.SEARCH command
• Enhanced RDB format to eliminate backfill after load
• Increase default for allowed number of indexes to 1000
• Increased limit on number of fields with default updated to 1000 and max to 10000
• FT.SEARCH and FT.AGGREGATE commands are no longer marked as denyoom
• Add Memory Tracking for Index Schema Attributes
• Replica nodes no longer participate in metadata consistency protocol. All mutation information comes from the Primary
• Enhanced FT.INFO command with additional information in Cluster mode

Contributors

• @Aksha1812
• @AlexFilipImproving
• @allenss-amazon
• @baswanth09
• @BCathcart
• @daddaman-amz
• @eifrah-aws
• @eliastam
• @eric-musliner
• @murphyjacob4
• @KarthikSubbarao
• @otherscase
• @boda26
• @Arunsarma07
• @Nivesh-01
• @VoletiRam
• @yairgott
• @yanamolo
• @zvi-code

Full Changelog: 1.1.0...1.2.0-rc1

Release 1.1

Major New Features

• FT.AGGREGATE command is implemented.
• Indexes and query operations no longer require a vector component.
• Consistency of operations across the cluster is substantially improved.
• The RDB format now includes information to avoid generating a backfill on reload when it's not required. (Backward compatible with 1.0 release)
• The build system has switched to cmake.
• A pytest-based integration test environment was added
• Full DBNum support for cluster mode is implemented.
• This release requires 9.0.1 Valkey core release or higher.

What's Changed

• Add integration test framework by @murphyjacob4 in #20
• CMake support by @eifrah-aws in #33
• Ensure that Index names and Prefixes do not have a slot tag, closes #31 by @allenss-amazon in #57
• Enforce vector indexed attribute by @yairgott in #63
• Add ACL support to Valkey Search. by @jeon1226 in #49
• Automatically configure threadpool size based on physical cpu core count. by @dpbnasika in #60
• Implement new RDB format by @murphyjacob4 in #68
• Add ASAN build option by @eifrah-aws in #100
• adding support for blocking clients during keyspace notification by @yairgott in #95
• FT.SEARCH - Update neighbor sorting to always be asc by @KarthikSubbarao in #104
• Rename ValkeySearch to valkey-search by @yairgott in #119
• Dynamic thread pool by @eifrah-aws in #126
• Command registration permissions alignment with ACL prefix permissions by @yairgott in #125
• Introduce RFC for RDB format by @murphyjacob4 in #28
• Add OOM checking to index backfill by @murphyjacob4 in #135
• renaming total_indexed_hash_keys to total_indexed_documents by @yairgott in #137
• Fixing empty vector query clause ft.search cmd parsing by @yairgott in #138
• Allow changing the reader/writer threads count at runtime by @eifrah-aws in #129
• Module-Config: fix a possibility of a double free by @eifrah-aws in #149
• Enable integration tests + ASan by @eifrah-aws in #162
• Search trigger for Valkey Extension Automation by @Nikhil-Manglore in #163
• Fix bug resulting in no ingestion to indexes with the same prefix as a dropped index by @eifrah-aws in #166
• Alpine Linux related fixes by @eifrah-aws in #168
• Fix GetRedisLocalPort function to support TLS by @yanamolo in #144
• Add coordinator port info metric by @yanamolo in #177
• The module can now be built on macOS by @eifrah-aws in #178
• We renamed Valkey-Extensions to Valkey-Bundle by @Nikhil-Manglore in #189
• Valkey Test Framework: load JSON module by @eifrah-aws in #198
• Revision of info framework. by @allenss-amazon in #202
• Add FT._DEBUG command by @allenss-amazon in #207
• Limit FT.SEARCH query string parsing recursive depth by a module config by @KarthikSubbarao in #182
• Add network byte counting to coordinator client by @yulazariy in #212
• Add operational limits to create command by @aradz44 in #214
• Add CPU metrics for reader and writer threads by @yanamolo in #217
• Add query string bytes limit using a module config by @KarthikSubbarao in #222
• Add a limit on the number of terms in a search query by @aradz44 in #229
• Add limit to the K value in FT.SEARCH cmd by @aradz44 in #231
• Limit the result record in ft.search cmd by @aradz44 in #239
• Supporting tsan build and addressing the time slice mutex test issues by @yairgott in #236
• Add metric for reclaimable memory by @aradz44 in #240
• Adding support for non vector queries by @KarthikSubbarao in #218
• Fix ACL bug where commands where allowed when user has no permissions for one of the index prefixes by @yanamolo in #246
• Reimplement timeout handling. by @allenss-amazon in #249
• Change visibility for some of info fields to developer only by @yulazariy in #281
• Handle OOM during search command fanout by @yanamolo in #268
• Reject queries when no valid pre-filter and no vector filter is provided by @KarthikSubbarao in #283
• Add coordinator metadata health monitoring info field by @yulazariy in #284
• Added support for cluster with replicas by @eifrah-aws in #313
• Properly wait for replication group to reach stabilization status by @eifrah-aws in #315
• Fanout operation base and cluster-wide primary and cluster fanout operation modes by @boda26 in #310
• Strip the complete build path from logs by @Shivshankar-Reddy in #308
• feat: memory scope & memory usage for string interning store by @proost in #153
• Expose global symbols so that crash dump backtraces are useful by @allenss-amazon in #329
• Add a configuration for resuming workers after fork by @eranfrie in #301
• Introduced new "debug-mode" configuration entry. by @eifrah-aws in #339
• Add support to retrieving JSON values via the JSON shared API by @yairgott in #287
• Add configurable thread pool fairness mechanism with comprehensive testing by @yulazariy in #362
• Respect valkey and json module versions defined in build and integration tests by @ChristophKoerner in #363
• Remove command alloed acl checks is it is already done by Valkey by @yulazariy in #387
• added and removed stats and attributes from ft.info by @Nivesh-01 in #370
• tag casesentive updated to be key-value pair by @Nivesh-01 in #395
• Add --format option to build script by @BCathcart in #384
• FT.CREATE consistency check by @boda26 in #386
• FT.DROPINDEX consistency check by @boda26 in #393
• Implement server version checking on load by @allenss-amazon in #408
• Implementation of FT.AGGREGATE Command by @allenss-amazon in #345
• addressing non-vector search issues by @yairgott in #400
• prevent Valkey crash when module is loaded twice by @dmitrypol in #415
• Prevent crash on Valkey 7.2 when loading valkey-search by @dmitrypol in #427
• Cluster Map Implementation by @boda26 in #435
• Replace multiple utility libraries with unified vmsdklib dependency by @eifrah-aws in #477
• Update ProcessKeyspaceNotification to not return before trying out all attributes by @KarthikSubbarao in #489
• Revise Save/Restore for true pit snapshot. by @allenss-amazon in #401
• Add support for checking ACL key prefix permissions using the new Valkey module API. by @eifrah-aws in #479
• Populate fingerprint and version to IndexSchema by @boda26 in #480
• feat: set command info for ft.create by @proost in #299
• FT.SEARCH partition and consistency controls by @boda26 in #464
• FT.INFO partition and consi...

1.0.2

Valkey Search 1.0.2 - Released Tue 25 November 2025

Bug fixes

• Disabled creation of indexes on database numbers != 0 in cluster mode.

1.0.1

Valkey Search 1.0.1 - Released Wed 11 June 2025

Bug fixes

• Module-Config: fix a possibility of of a double free (#149)
• More resilient builds on different Linux distros (#152, #154, #158, #168, #172)
• Use a managed pointer for thread safe context in coordinator (#159)
• Enable integration tests + ASAN (#162)
• Fixed drop index bug (#166)
• Fix GetRedisLocalPort function to support TLS (#144)

1.0.0

Valkey Search 1.0.0 GA - Released Wed 28 May 2025

This is the first official release of Valkey Search 1.0. This release is fully compatible with Valkey 8.1.1 and later releases.

Behavior changes

• Adding support for blocking clients during keyspace notification (#95)
• Make index backfill batch/block size configurable (#97)
• FT.SEARCH - Update neighbor sorting to always be asc (#104)
• Add OOM checking to index backfill (#135)

Bug fixes

• Avoid unsafe code in INFO during crash dump (#98)
• Support counting cores on aarch64 (#107)
• Create index with the correct DB number (#120) (#123)
• Fix dumping corrupted rdb file (#131)
• Fix empty vector query clause ft.search cmd parsing (#138)

Performance/efficiency improvements

• Allow dynamic adjustment of thread count at runtime (#126) (#129)

Build and packaging changes

• Allow building with external libs (#85)
• Don't suppress error when running python integration tests (#89)
• Add ASAN build option (#100)

Valkey Search 1.0.0 RC1 - Released Fri 28 Mar 2025

This is the first release candidate of valkey-search 1.0 that is a high-performance Vector Similarity Search engine optimized for AI-driven workloads. It delivers single-digit millisecond latency and high QPS, capable of handling billions of vectors with over 99% recall.

Valkey-Search allows users to create indexes and perform similarity searches, incorporating complex filters. It supports Approximate Nearest Neighbor (ANN) search with HNSW and exact matching using K-Nearest Neighbors (KNN). Users can index data using either Valkey Hash or Valkey-JSON data types.

Major API and Functionality

• Add the search module data type which can handle RDB load, RDB save, free, and memory usage
• Add the following search module commands:
  • FT.CREATE
  • FT.DROPINDEX
  • FT.INFO
  • FT._LIST
  • FT.SEARCH
• Supported indexes
  • Vector: HNSW and Flat
  • Non-vector: Numeric and Tag
• Index data types include Valkey Hash and Valkey JSON.
• Cluster support including cross-shard search via coordinator mode
• Linear scaling of keyspace and compute
• ACL support
• RDB serialization of metadata including the search index
• Hybrid queries combining vector and non vector indexes
• Handle key space events for data mutation
• Expose statistics and reporting memory usage to the core valkey engine

New configurations

• Add support for the following configurations: reader-threads, writer-threads, use-coordinator, log-level

1.0.0-rc1

Valkey Search 1.0.0 RC1 - Released Fri 28 Mar 2025

This is the first release candidate of valkey-search 1.0 that is a high-performance Vector Similarity Search engine optimized for AI-driven workloads. It delivers single-digit millisecond latency and high QPS, capable of handling billions of vectors with over 99% recall.

Valkey-Search allows users to create indexes and perform similarity searches, incorporating complex filters. It supports Approximate Nearest Neighbor (ANN) search with HNSW and exact matching using K-Nearest Neighbors (KNN). Users can index data using either Valkey Hash or Valkey-JSON data types.

Major API and Functionality

• Add the search module data type which can handle RDB load, RDB save, free, and memory usage
• Add the following search module commands:
  ** FT.CREATE
  ** FT.DROPINDEX
  ** FT.INFO
  ** FT._LIST
  ** FT.SEARCH
• Supported indexes
  ** Vector: HNSW and Flat
  ** Non-vector: Numeric and Tag
• Index data types include Valkey Hash and Valkey JSON.
• Cluster support
• Linear scaling of keyspace and compute
• ACL support
• RDB serialization of metadata including the search index
• Hybrid queries combining vector and non vector indexes
• Handle key space events for data mutation
• Expose statistics and reporting memory usage to the core valkey engine

New configurations

• Add support for the following configurations: reader-threads, writer-threads, use-coordinator, log-level