manticoresoftware/manticoresearch release-28.6.6

Manticore Search 28.6.6

on GitHub

Version 28.6.6

Released: July 31st 2026

This release adds UUID document IDs and ordered, limited GROUP_CONCAT(), with fixes across replication, backups, query processing, SQL compatibility, and secondary indexes.

New Features and Improvements

• 🆕 v28.6.0 Issue #4746 Added ordered, limited GROUP_CONCAT() for explicit SphinxQL GROUP BY queries: GROUP_CONCAT(expression ORDER BY ... [SEPARATOR '...'] LIMIT N) keeps the top N values per group in the requested order.
• 🆕 v28.5.0 Issue #1429 Added UUID document IDs for real-time tables via id uuid, including explicit or automatically generated IDs, UUID equality and IN filters, and UUID-based REPLACE, UPDATE, and DELETE.

Bug Fixes

• 🪲 v28.6.6 Issue #4741 Fixed USER() parsing after authentication support made USER a keyword, restoring MySQL client status command compatibility.
• 🪲 v28.6.5 Issue #4764 Fixed CALL SNIPPETS taking seconds on complex boolean queries when boolean_simplify is enabled, and restored expected highlighting for affected queries.
• 🪲 v28.6.4 Issue #4737 Fixed OPTIMIZE TABLE parsing for qualified system.<table> names, allowing physical sharded tables to be optimized.
• 🪲 v28.6.3 Issue #138 Updated manticore-backup to 1.10.2, fixing successful backups leaving RT tables frozen and subsequent writes blocked indefinitely.
• 🪲 v28.6.2 Issue #4758 Fixed obsolete RT external files being left behind after ALTER, TRUNCATE, chunk removal, or optimization, and prevented unrelated ALTER TABLE changes from losing Jieba tokenization settings; unsupported effective Jieba setting changes now fail explicitly.
• 🪲 v28.6.1 Issue #4750 Fixed authenticated replication state transfer so ALTER CLUSTER ... ADD can add existing populated RT tables instead of failing with retries exhausted.
• 🪲 v28.5.3 Issue #4730 Fixed long exact and phrase-like queries potentially corrupting memory or crashing searchd by accounting for their full term width in the query stack guard.
• 🪲 v28.5.2 PR #4729 Improved local_df query processing for multi-chunk RT and distributed tables by deduplicating repeated keyword statistics, avoiding excessive pre-search work for large repeated OR and phrase queries.
• 🪲 v28.5.1 Issue #4726 Fixed a SphinxQL connection-handling race that could spin workers at 100% CPU and leave searchd permanently unresponsive when clients connected during startup or aborted an early SSL upgrade.
• 🪲 v28.4.11 PR #4725 Improved indextool --check to validate secondary-index structure and payloads, reporting corrupted .spidx files as check failures instead of crashing.
• 🪲 v28.4.10 PR #4735 Fixed the Ruby API fixture's response parsing to preserve valid "0" chunks, matching the PHP client behavior.
• 🪲 v28.4.9 Issue #4595 Updated MCL to 13.8.2, fixing a StreamVByte decoder over-read that could crash queries using secondary indexes, including GROUP BY on JSON fields.
• 🪲 v28.4.8 PR #4703 Fixed PHP API response parsing when a valid response chunk ends with the string "0", preventing false failed to read searchd response errors for affected packet sizes.
• 🪲 v28.4.7 PR #4721 Fixed a concurrent SHOW INDEX ... STATUS race that could corrupt query-stat percentile calculations (p95/p99).
• 🪲 v28.4.6 Issue #4677 Fixed document-ID IN(...) filters silently returning no rows for valid unsigned 64-bit IDs above Long.MAX_VALUE; invalid out-of-range document IDs now fail validation without restricting valid multi64 values.
• 🪲 v28.4.5 Issue #4633 Fixed a crash when a LEFT JOIN query uses multiple FACET clauses by preventing join multi-facet sorters from being cleaned up twice.