Manticore Search 17.5.1: Maintenance Release with Breaking Changes and Enhanced Features
#Backend

Manticore Search 17.5.1: Maintenance Release with Breaking Changes and Enhanced Features

Backend Reporter
3 min read

Manticore Search 17.5.1 brings bug fixes, memory optimizations, and new model support for auto-embeddings, along with two breaking changes requiring attention during upgrades.

The Manticore Search team has released version 17.5.1, a maintenance update that addresses numerous bugs while introducing several improvements to the search engine's functionality. This release focuses on stability and performance enhancements, making it an important update for production deployments.

Breaking Changes to Consider

Two breaking changes require attention when upgrading from older versions:

MCL Interface Update: The Manticore Columnar Library (MCL) has been updated to version 10.0.0, which adds support for the DROP CACHE command. This changes the interface between the daemon and MCL, meaning older Manticore Search versions won't support the newer MCL. If you're managing your own library installations, ensure compatibility.

Percolate Query Response Format: JSON responses from percolate queries now return _id and _score fields as numbers instead of strings. This aligns percolate query behavior with regular search responses but may break clients that explicitly expect string types for these fields.

The release recommends updating companion libraries:

  • MCL (Manticore Columnar Library): 10.2.0
  • Manticore Buddy: 3.41.0

Users following the official installation guide don't need to manually manage these versions, as the correct dependencies will be installed automatically.

New Features and Improvements

Enhanced Auto-Embedding Support

The updated MCL brings expanded support for modern language models in auto-embeddings, including Llama, Qwen, Mistral, Gemma, and other popular models. This enhancement allows Manticore Search to better integrate with contemporary AI workflows and semantic search applications.

Memory Optimization

A significant memory optimization has been implemented for Jieba morphology instances. Previously, each table using Jieba morphology created separate instances, consuming considerable memory when multiple tables were configured. Now, instances are shared across tables with identical configurations, substantially reducing memory usage in multi-table deployments.

Simplified Table Creation

Inline configuration support has been added for several text processing components. You can now specify stopwords, wordforms, exceptions, and hitless_words directly within CREATE TABLE statements. This eliminates the need for external configuration files, simplifying table deployment and management.

Critical Bug Fixes

This release addresses 47 bugs across various components of the search engine:

JOIN Operation Fixes: Several issues with JOIN operations have been resolved. Queries returning empty or duplicated values when a column served as both a string attribute and stored field now return correct attribute values. Additionally, JOIN operations on JSON string attributes (e.g., j.s) now function properly, matching the behavior of plain string attributes.

Highlighting Improvements: The highlight() function with html_strip_mode=strip previously corrupted content by decoding HTML entities and altering tag structures. This has been fixed to preserve the original entity form while still providing proper highlighting.

Table Management Fixes: Multiple table management operations have been stabilized. ALTER TABLE REBUILD SECONDARY now handles tables with multiple disk chunks correctly. Table rename operations properly migrate external stopwords, wordforms, and exceptions files when using ATTACH TABLE. Distributed queries now return stored fields from the correct local index, even when agent tables contain duplicate document IDs.

Query Logic Corrections: The MATCH operator with OR conditions over the same phrase in different fields now returns matches only from the intended fields, rather than including matches from other fields. ALTER TABLE operations with table-level settings now work correctly on tables using auto-embeddings, with proper serialization that omits knn_dims when model_name is set.

Compatibility and Upgrade

Manticore Search 17.5.1 maintains strong backward compatibility with existing data and queries. The only breaking changes are the MCL interface update and percolate query response format change noted above. Users upgrading from previous versions should review these changes and test their applications accordingly.

For installation instructions and detailed upgrade procedures, consult the official installation guide. The Manticore team provides multiple support channels including Slack, community forums, GitHub issue tracking, and direct email support at [email protected].

For the complete list of changes, improvements, and bug fixes, refer to the official changelog accompanying this release.

Comments

Loading comments...