v0.4.0

0.4.0 (2026-09-13)

Features

• Allow keys to be rotated and protect against accidental removals of last key (5a850df)
• Answer 304 when a GET carries an ETag and the resource is unchanged (3517e54)
• Answer every failed request with a standard JSON error body (96bb3f3)
• Filter and page the index, key and reindex listings (a9b9080)
• Generate real query and field types in OpenAPI clients (6104987)
• Improve batch indexing error handling and allow skipping docs with errors (f626075)
• Introduce shared cache settings for the entire node (983b10a)
• Let a deployment cap how many facet values and suggestions a request may ask for (012c483)
• Measure facet value and explain requests (dcedc22)
• Normalize error codes for numeric ranges (7d9f1e7)
• Publish a named Interpret union in the OpenAPI document (7f8acf7)
• Read and cancel a reindex job under /admin/reindexes (00ff93d)
• Reindex jobs are now identifiable and carry more metadata (e9bbf70)
• Rename every error code by family, request part and defect (fd4cb1a)
• Rename index:not-found to index:not_found (045382b)
• Rename some request:update errors to index:update (67774a4)
• Rename the live generation liveGeneration when listing indexes (493d02f)
• Report enums in the REST API as lowercase (2c1ac3c)
• Report every error path in the REST API using dotted form (4af6fd5)
• Report which index generation answered a search (51663ce)
• Require all parameter to delete all documents (21d1e36)
• Require indexes.promote permission when auto-promoting reindex (eeb47ec)
• Return missing primary keys as text (561faac)
• Return the index with its ETag from commit and pull (068f6cb)
• Stronger If-Match application in the REST API (3be9a94)
• Support for ranking signals with non-indexing updates (402b301)

Bug fixes

• Accept a document key written as another JSON type in a PATCH body (4a6e2f4)
• Answer index:not_found when trying to reindex a non-existent index (1980fbe)
• Avoid sending back empty tieBreakers for rankings (53c6755)
• Cancel a queued reindex without a 15 second wait (b062e1a)
• Count the clauses of interpret targets against the search clause caps (a68d202)
• Ignore paging, sort and rescore in explain instead of validating them (8c89ca4)
• Index documents forwarded over HTTP/2 instead of reporting zero indexed (6933dbc)
• Keep a closing index copy from overwriting files (64051b3)
• Keep a registry repair from undoing a delete that lands while it runs (993f066)
• Keep a rescored search from paging into an empty page (33328ae)
• Keep a slow registry read from hiding an index another thread just created (1036215)
• Keep attempting to push indexes if a push fails (830c102)
• Keep registry entries a newer node wrote when an older node changes it (87ff74e)
• Keep the definition and change log intact when a node stops mid-write (7ce5358)
• Mark storage as removed when a delete cannot clear a local copy (6cf9dbf)
• Refuse a boost weight of NaN or Infinity instead of scoring with it (3204515)
• Report NDJSON document errors without a documents prefix (f314c49)
• REST API now answers more calls with 4xx instead of 500 (2f9aad7)
• Return 400 for an invalid after parameter when reading documents (b4edb2e)
• Stop a refused reindex from leaving an empty generation behind (b2f2008)
• Stop omitting score from a search ranked by signals or a second pass (9812afd)
• Tell apart an empty matched and a blank field name, and point highlight errors at the field (9a44f8b)

Performance

• Cache typo and prefix matches on small segments too (ffba8f2)
• Reuse prefix automata across fields and requests (fe3257e)
• Speed up long and typo tolerant searches by reusing term lookups (1facf2d)