RediSearch 2.8 release notes

RESP3 support. Geo Polygon Search. Performance improvements.

Requirements

RediSearch v2.8.12 requires:

• Minimum Redis compatibility version (database): 7.2
• Minimum Redis Enterprise Software version (cluster): 7.2.4

v2.8.12 (March 2024)

This is a maintenance release for RediSearch 2.8.

Update urgency: MODERATE: Program an upgrade of the server, but it's not urgent.

Details:

• Bug fixes:

  • #4476 Split INFIX and SUFFIX report on FT.EXPLAIN and FT.EXPLAINCLI (MOD-6186)
  • #4467 Memory leak upon suffix query for a TAG indexed with WITHSUFFIXTRIE (MOD-6644)
  • #4403 Clustered FT.SEARCH hangs forever without replying when an invalid topology is found (MOD-6557)
  • #4355 Searching for a synonym will iterate in the same group multiple times, causing a performance hit (MOD-6490)

• Improvements:

  • #4313 Memory allocation patterns on the memory used to query GEOSHAPE types (MOD-6431)

v2.8.11 (January 2024)

This is a maintenance release for RediSearch 2.8.

Update urgency: MODERATE: Program an upgrade of the server, but it's not urgent.

Details:

• Bug fixes:

  • #4324 Internal cluster mechanism not waiting until all replies from shards causing a crash (MOD-6287)
  • #4297 Execution loader when using FT.AGGREGATE with LOAD stage failing to buffer the right results potentially causing a crash (MOD-6385)

• Improvements:

  • #4264 Granularity of the time reporting counters on FT.PROFILE (MOD-6002)

v2.8.10 (January 2024)

This is a maintenance release for RediSearch 2.8.

Update urgency: HIGH: There is a critical bug that may affect a subset of users. Upgrade!

Details:

• Bug fixes:

  • #4287 Re-index process while syncing from the replica causes a crash due to internal index variable initialization (MOD-6337, MOD-6336)
  • #4249 Memory tracking on cluster setups causing high memory usage and potentially Out-of-Memory (MOD-6123, MOD-5639)
  • #4244 Profiling FT.AGGREGATE using the WITHCURSOR flag with a - clause causes a crash due to timeout (MOD-5512)
  • #3916 Expiring JSON documents while querying it causing a crash due to deadlock (MOD-5769, MOD-5895, MOD-6189, MOD-5895)
  • #4235 Memory excessively growing on databases caused by unbalanced nodes on inverted index trie (MOD-5880, MOD-5952, MOD-6003)
  • #4190 Profiling FT.AGGREGATE causes a crash on RESP3 replies (MOD-6250, MOD-6295)
  • #4148, #4038 ON_TIMEOUT FAIL\RETURN policies in the cluster setup not being respected (MOD-6035, MOD-5948, MOD-6090)
  • #4110 Format of error response contains inconsistencies when timing out (MOD-6011, MOD-5965)
  • #4104 FT.SEARCH not responding when using TLS encryption on Amazon Linux 2 (MOD-6012)
  • #4009 In cluster setup does not return a timeout error for FT.SEARCH (MOD-5911)
  • #3920 In cluster setup does not return a timeout error for FT.AGGREGATE (MOD-5209)
  • #3914 FT.CURSOR READ with geo queries causing a crash when data is updated between the cursor reads (MOD-5646)
  • #4220 Server crash when attempting to run the ForkGC (Garbage Collection routine) after dropping the index (MOD-6276)

• Improvements:

  • #3682 Report last key error and field type indexing failures on FT.INFO (MOD-5364)
  • #4236 Adding Vector index parameters at the FT.INFO report (MOD-6198)
  • #4196 Check for timeout after results processing in FT.SEARCH on cluster setup (MOD-6278)
  • #4164 Report TIMEOUT, MAXPREFIXEXPANSION warnings in RESP3 replies (MOD-6234)
  • #4165 Indicate timeout on FT.PROFILE report (MOD-6184)
  • #4149 Indicate timeout from Cursor on FAIL timeout policy (MOD-5990)
  • #4147 Initialization of the maximum numeric value range leading to a better balance of the index leaf splitting (MOD-6232)
  • #3940 Query optimization when predicate contains multiple INTERSECTION (AND) of UNION (OR) (MOD-5910)
  • #4059 Return cursor id when experiencing a timeout, when the policy is ON_TIMEOUT RETURN (MOD-5966)
  • #4006 Possibly problematic index name alias validation (MOD-5945)

v2.8.9 (October 2023)

This is a maintenance release for RediSearch 2.8.

Update urgency: HIGH: There is a critical bug that may affect a subset of users. Upgrade!

Details:

• Bug fixes:

  • #3874 Heavy document updates causing memory growth once memory blocks weren't properly released (MOD-5181)
  • #3967 Resharding optimizations causing the process to get stuck (MOD-5874, MOD-5864)
  • #3892 After cleaning the index the GC could cause corruption on unique values (MOD-5815)
  • #3853 Queries with WITHCURSOR making memory growth since CURSOR wasn't invalidated in the shards (MOD-5580)

• Improvements:

  • #3938 Propagating error messages in multiple shards database, instead of failing silently (MOD-5211)
  • #3903 Added support for Rocky Linux 9 and RHEL9 (MOD-5759)

v2.8.8 (September 2023)

This is a maintenance release for RediSearch 2.8.

Update urgency: SECURITY: There are security fixes in the release.

Details:

• Security and privacy:

  • #3788 Don’t expose internal cluster commands (MOD-5706)
  • #3844 Limits maximum phonetic length avoiding to be exploit (MOD 5767)

• Bug fixes:

  • #3771 Broken lower() and upper() functions on APPLY stage in FT.AGGREGATE in DIALECT 3 (MOD-5041)
  • #3752 Setting low MAXIDLE parameter value in FT.AGGREGATE cause a crash (MOD-5608)
  • #3780 Wrong document length calculation causing incorrect score values (MOD-5622)
  • #3808 LOAD step after a FILTER step could cause a crash on FT.AGGREGATE (MOD-5267)
  • #3823 APPLY or FILTER parser leak (MOD-5751)
  • #3837 Connection using TLS fail on Redis 7.2 (MOD-5768)
  • #3856 Adding new nodes to OSS cluster causing a crash (MOD-5778)
  • #3854 Vector range query could cause Out-of-Memory due to a memory corruption (MOD-5791)

• Improvements:

  • #3534 Vector Similarity 0.7.1 (MOD-5624)

v2.8 GA (v2.8.4) (July 2023)

This is the General Availability release of RediSearch 2.8.

Headlines

RediSearch 2.8 introduces support for RESP3, new features, performance improvements, and bug fixes.

What's new in 2.8.4

This new major version introduces new and frequently asked for Geo Polygon Search, adding the GEOSHAPE field type that supports polygon shapes using WKT notation. Besides the current GEO (alias for GEOPOINT) used in geo range queries, we add support for POLYGON and POINT as new geo shape formats (new GEOSHAPE). In addition, 2.8 brings performance improvements for SORT BY operations using FT.SEARCH and FT.AGGREGATE, and new FORMAT for enhanced responses on FT.SEARCH and FT.AGGREGATE in RESP3 only.

Features:

• Introduce support for geo polygon shapes and queries:

  • Adding GEOSHAPE field type to map polygons in the SCHEMA on FT.CREATE (MOD-4798)

  • Support for polygons POLYGON and POINT using WKT notation, for example POLYGON((lon1 lat1, lon2 lat2, ...))

  • Adjust the query syntax on FT.SEARCH for polygons using the predicate @geom:[OPERATOR $poly] and defining polygon in WKT format as PARAMS 2 poly "POLYGON((10 20, ...))" using DIALECT 3

  • Initially WITHIN and CONTAINS operators with GEOSHAPES for now

  • Support multiple coordinate systems: cartesian (X,Y) with the flag FLAT for flat earth and geographic (lon, lat) using the flag SPHERICAL (MOD-5303). Geographic coordinate system using spherical indexing as default (SPHERICAL)

  • Add memory usage per Geometry Index in the FT.INFO response report (MOD-5278)

• Introduce performance optimization for sorting operations on FT.SEARCH and FT.AGGREGATE as default on DIALECT 4. It will improve performance in 4 different scenarios, listed below:

  • Skip Sorter - applied when there is no sort of any kind. The query can return once it reaches the LIMIT requested results.

  • Partial Range - applied when there is a SORTBY a numeric field, with no filter or filter by the same numeric field, the query iterates on a range large enough to satisfy the LIMIT requested results.

  • Hybrid - applied when there is a SORTBY a numeric field in addition to another non-numeric filter. Some results will get filtered, and the initial range may not be large enough. The iterator then is rewinded with the following ranges, and an additional iteration takes place to collect LIMIT requested results.

  • No optimization - If there is a sort by score or by non-numeric field, there is no other option but to retrieve all results and compare their values.

• Add WITHCOUNT argument that allow return accurate counts for the query results with sorting. This operation processes all results in order to get accurate count, being less performant than the optimized option (default behavior on DIALECT 4) (MOD-5311)

• New FORMAT argument in FT.SEARCH and FT.AGGREGATE to retrieve the results as JSON strings or RESP3 hierarchical structures (RESP3 only) (MOD-5390)

Improvements (since 2.8.3):

• #3717 - Polygon shapes validation and orientation correction when clockwise (MOD-5575)

• #3534 - Vector Similarity [0.7.0]

• #3657 - Allow GC calls for all tiered indexes in the schema

• #3701 - HNSW is now using data blocks to store vectors and metadata instead of array

Changed behavior:

• #3355, #3635 Expired keys deleted from replica's index, returning an empty array instead of nil (MOD-4739)