LoginSign Up
← Back to all sparks
M

Manticore Search

DEVOPS
Velocity6.3

Open-source database for full-text search and analytics

manticoresearch.com ↗

Three patch releases in eight hours: Manticore's 29.0 line is being stabilized in public.

search engineshardingpatch cadencequery correctnessvector search
◆Current state
Manticore shipped 29.0.3, 29.0.4 and 29.0.5 inside a single day, three days after the 29.0.2 release notes consolidated the sharding work. Two are correctness fixes in query handling — internal sort helper columns leaking into LEFT JOIN output, and NEAR and proximity distances being reset by repeated JSON query fixup. The third hardens the binary API against malformed search counts under a reported advisory.
◆Where it's heading
The pattern after a major line opens is holding: the commit-level train keeps running at multiple releases a day while the newly exposed surfaces — sharded tables, LEFT JOIN, JSON query parsing — report their edge cases back. The bugs being caught are shaped by what 29.0 changed rather than by new work, and each one arrives with regression coverage attached, which is why they land as separate patch releases rather than accumulating. Search-side work continues to split between classic full-text concerns and the vector and conversational paths.
◆Prediction
The patch cadence should keep compressing toward the ordinary rhythm as the 29.0 edge cases drain; the next substantive item is more likely to come from the columnar and KNN thread than from sharding, which has just had its release.

◆Recent moves

  1. 1d ago

    NEAR and proximity distances survive repeated JSON query fixup

    Null-sweep bookkeeping was sharing state with operator arguments, so repeated JSON query_string fixup reset NEAR and proximity distances. A relevance-affecting parser bug, fixed with HTTP regression coverage for negated positional queries.

    View source ↗
  2. 1d ago

    Internal sort columns no longer leak into LEFT JOIN results

    Internal sort helper columns were identified by schema position, so LEFT JOIN appending its null mask pushed them into client output. Filtering them by identity instead keeps them available for distributed sorting — an edge case created by the join work itself.

    View source ↗
  3. 1d ago

    Binary API rejects malformed search counts before allocating

    The binary API now rejects negative and impossible SEARCH parser counts before allocating vectors, closing a reported advisory. It also validates the top-level query count when batching is disabled, and ships daemon-survival coverage — hardening the protocol boundary rather than the query engine behind it.

    View source ↗
  4. 4d ago

    Manticore 29.0 makes sharded tables operable and changes the SHARD_WRITE protocol

    ⚡ SPARK

    The 29.0 release notes consolidate the sharding work the commit train has been building: synchronous OPTIMIZE across distributed and sharded tables, settings inspection on native shards, and a breaking change to SHARD_WRITE that streams heartbeat replies during long shard writes and replication SST. It is the point where sharded tables stop being a creation-time feature and become an operable one.

    View source ↗
  5. 4d ago

    29.0.2 bumps manticore-backup to 1.10.3 and Buddy to 4.4.1

    Component version bumps for manticore-backup and Buddy, shipped as their own tagged release hours before the 29.0.2 notes. Routine train mechanics.

    View source ↗
  6. 6d ago

    SHOW TABLE SETTINGS now works on sharded tables

    Settings inspection arrives for native sharded tables, one of the operator controls the 29.0 notes later gathered up. Each of these commits handed operators something that previously existed only for single local tables.

    View source ↗