What manticoresoftware/manticoresearch shipped
Written by FoxPlug from the repository's own releases; not affiliated with Manticoresearch. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/manticoresoftware/manticoresearch; Manticoresearch did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of October 5, 2026
What shipped
- Added a per-table knn rescore budget, vector prefetch for columnar rescoring, and
early_terminationon agents. Pull request #4969 - Split ramchunk into equal row ranges when building the hnsw graph and build hnsw in parallel on FLUSH RAMCHUNK. Pull request #4978
- Fixed a performance regression on filtered knn queries. Pull request #4985
- Bounded SST FILE_RESERVE hash metadata by adaptively increasing SST chunk sizes when SHA1 metadata would exceed the cluster packet budget. Pull request #4976
- Fixed merging MVA groups without source attributes in multi-table queries. Pull request #4990
- Fixed a performance regression introduced by commit dbc40aa by updating the columnar submodule hash. Pull request #4973
- Fixed apostrophe escaping in fuzzy SQL queries by updating Buddy to version 4.4.7. Release
- Added a CLT test for fuzzy apostrophe escaping. Pull request #4984
- Introduced a new --realistic mode in manticore-load. Release
- Updated manticore-load from version 1.28.0 to 1.29.0. Pull request #4991
Why it matters
This week brought several performance fixes, including filtered knn queries and a regression in columnar performance. New features such as per-table knn rescore budget and parallel hnsw building improve query efficiency. Several releases include these fixes, making the search experience more stable and faster for users.
Changelog entry
- Feat: per-table knn rescore budget, vector prefetch for columnar rescoring,
early_terminationon agents Pull request #4969 - Fix: split ramchunk in equal row ranges when building hnsw graph; build hnsw in parallel on FLUSH RAMCHUNK Pull request #4978
- Fix: fixed a performance regression on filtered knn queries Pull request #4985
- Fix: bound SST FILE_RESERVE hash metadata Pull request #4976
- Fix: merge MVA groups without source attributes Pull request #4990
- Fix: regression of performance after dbc40aa Pull request #4973
- Fix: apostrophe escaping in fuzzy SQL queries Release
- Ci: add CLT test for fuzzy apostrophe escaping Pull request #4984
- Feat: manticore-load --realistic mode Release
- Update load version from 1.28.0 to 1.29.0 Pull request #4991
- Ci: fix artifact upload verification race in build jobs Pull request #4988
- Moved CLT cases into ubertest 197; added HAVING case to run all CI stages Pull request #4992
New releases: per-table knn rescore budget, parallel hnsw build, fixes for filtered knn regression, MVA merge, apostrophe escaping, and more. See changelog.
This week we shipped several fixes and improvements. Performance regressions in filtered knn and columnar queries were fixed. MVA group merging in multi-table queries now works without source attributes. We added a per-table knn rescore budget, vector prefetch for columnar rescoring, and early termination on agents. HNSW graph building now splits ramchunks into equal row ranges and runs in parallel on FLUSH RAMCHUNK. SST FILE_RESERVE hash metadata is now bounded. Fuzzy SQL apostrophe escaping was fixed via Buddy 4.4.7. manticore-load gained a --realistic mode. Full details in the changelog.
Week of September 28, 2026
What shipped
- Hybrid queries now support facets and JSON aggregations that execute over fused documents with counts reflecting post-fusion filtering. Release
- DDL validation now reports unsupported or missing column types clearly and suggests database types where applicable. Release
- CI candidate cleanup was fixed. Pull request #4955
Why it matters
Facet support in hybrid search enables more sophisticated result analysis on fused datasets. Clearer DDL error messages reduce debugging time when creating tables with unsupported column types.
Changelog entry
- Hybrid queries now execute facets and JSON aggregations over fused documents, with counts reflecting post-fusion filtering Release
- DDL validation reports unsupported or missing column types more specifically and suggests database types where applicable Release
- Improved UUID column type handling in DDL Release
- Fixed CI candidate cleanup Pull request #4955
29.13.1 improves DDL error messages for unsupported column types. 29.13.0 adds facet/aggregation support in hybrid search with RRF fusion.
Two releases this week: 29.13.0 brings facet and JSON aggregation support to hybrid search, with counts properly reflecting post-fusion filtering. 29.13.1 improves DDL validation by reporting unsupported column types clearly and suggesting database types where applicable, making table creation less error-prone.
Week of September 21, 2026
What shipped
- Release 29.12.0 adds direct-to-disk bulk import for real-time tables through SQL and HTTP /bulk, building each batch as a disk chunk and publishing on commit to avoid RAM-chunk staging, with support for rollback, staging cleanup, duplicate-ID handling, and PURGE BULK_IMPORT. Release
- Release 29.12.2 fixes postlimit expressions in distributed tables with multiple locals having different settings. Release
- Debian and RHEL build issues are resolved by replacing std::filesystem with boost filesystem. Pull request #4940
- Fixed a race condition where REPLACE/DELETE operations paused by FREEZE could miss disk chunks being saved during the state transition. Pull request #4925
- Fixed a disk chunk ID race condition between parallel merges caused by non-atomic chunk ID generation in merge workers. Pull request #4924
- Added freeze support for plain tables as a prerequisite for manticore-backup plain mode support. Pull request #4919
- Manual documentation expanded to cover HTTP errors, readiness checks, numerical precision, geo-distance models, KNN, and unsupported mathematical functions. Pull request #4927
- Fixed memory retention issue where MVA group-by source data was not properly released. Pull request #4921
- Fixed geopoly contains operation to preserve EXIST float type correctly. Pull request #4909
- Fixed JSON highlight queries to properly escape apostrophes. Pull request #4913
Why it matters
This week brings significant improvements to real-time table operations with the new direct-to-disk bulk import feature for better performance and build reliability across platforms. Several critical race conditions and data handling issues in distributed tables and chunk management have been fixed, improving system stability.
Changelog entry
- 29.12.2: fixed postlimit exprs in distributed tables of several locals with different settings Release
- Fixed debian and rhel build issues by replacing std::filesystem with boost filesystem Pull request #4940
- 29.12.1: fixed build issues by replacing std::filesystem with boost filesystem Release
- 29.12.0: added direct-to-disk bulk import for real-time tables with SQL and HTTP /bulk support, rollback, and duplicate-ID handling Release
- 29.11.6: fixed REPLACE/DELETE paused by FREEZE missing disk chunk saved meanwhile Pull request #4925
- Fixed disk chunk id race between parallel merges Pull request #4924
- 29.11.4: fixed preserve EXIST float type in geopoly contains Release
- 29.11.3: fixed avoid retaining MVA group-by source Release
- 29.11.2: bumped buddy version from 4.4.4 to 4.4.5 Release
- 29.11.1: fixed escape apostrophes in JSON highlight queries Release
- Added freeze for plain tables Pull request #4919
29.12.2 shipped: direct-to-disk bulk import for RT tables, fixed postlimit in distributed setups, resolved FREEZE/REPLACE races, and improved build compatibility.
Week 29 brought major improvements to Manticoresearch: direct-to-disk bulk import for real-time tables now enables faster large-volume data insertion, build stability improved across Debian and RHEL, and critical race conditions fixed in chunk merging and FREEZE operations. Documentation expanded for HTTP behavior, geo-distance models, and numerical precision.
Week of September 14, 2026
What shipped
- Release 29.10.0 adds support for ONNX models without token types, including BAAI/bge-m3 embeddings. Release
- Fixed a deadlock that could occur when concurrent BERT auto-embedding requests blocked on shared resources. Pull request #4916
- Fixed docid lookup synchronization with index format version during ALTER operations and repaired v70 lookups under v71 headers. Pull request #4858
- Increased implicit
max_matchesin non-rescored KNN queries to match the k parameter. Pull request #4900 - Fixed an implicit sorter tdigest memory leak. Pull request #4912
- Added test coverage for CHAT MODEL to verify it correctly accepts
api_key,base_url, and timeout parameters. Pull request #4896 - Added MariaDB 13.0 mysqldump coverage to CI testing. Pull request #4911
- Updated buddy dependency to version 4.4.4. Pull request #4901
Why it matters
This week's updates address critical concurrency issues in embedding operations and fix synchronization problems with index format versions. The addition of BAAI/bge-m3 support expands embedding model compatibility for users working with semantic search.
Changelog entry
- Feat: support ONNX models without token types (e.g. BAAI/bge-m3) Release
- Fix: prevent deadlock in concurrent BERT auto-embeddings Pull request #4916
- Fix: keep docid lookup in sync with index format version on ALTER; repair v70 lookups under v71 header Pull request #4858
- Fix: increase implicit
max_matchesto match k in non-rescored KNN queries Pull request #4900 - Fix: implicit sorter tdigest leak Pull request #4912
- Fix: CHAT MODEL now correctly accepts
api_key,base_url, and timeout Pull request #4896 - Ci: add MariaDB 13.0 mysqldump coverage Pull request #4911
- Deps: update buddy to 4.4.4 Pull request #4901
29.10.0 ships support for BAAI bge-m3 ONNX models, fixes a deadlock in concurrent BERT embeddings, and resolves index format sync issues on ALTER operations.
29.10.0 is available with several important fixes. Concurrent BERT auto-embedding requests no longer deadlock. Docid lookup now syncs correctly with index format versions during ALTER operations. ONNX models without token types are now supported, including BAAI/bge-m3. KNN queries also get improved max_matches handling.