ArangoDB version 3.12.10 released

[vadim]

We are happy to announce the release of ArangoDB version 3.12.10!

This patch version is ready for production deployment and is a recommended upgrade for all 3.12 users.

Apart from some basic cleanup and code improvements related to found issues, the following notable things have been added to the 3.12.9.4 version of ArangoDB:

• Updated ArangoDB Starter to v0.19.25.
• COR-726/COR-787/COR-792: Removed the deprecated replication applier's RESTAPI, along with the arangosh client methods that used them.
• BTS-2400: Added the `--vector-index-build-retry-backoff` startup option (seconds) to configure the vector index build retry backoff (default: 60).
• Updated node modules:brace-expansion to 5.0.8
• Updated node modules: fast-uri to 3.1.4, js-yaml to 4.3.0.
• Upgraded included rclone to v1.74.4 with go1.25.12 and non-vulnerable dependencies.
• COR-613: Create Response Code Metrics arangodb_http_response_http_response_code_total{code="<response-code>"}.
• COR-728: Removed the telemetrics API and the arangosh telemetrics client.
• COR-473: Enabled linear scanning of shards that are untrained and in unusable state for having insufficient documents.
• COR-75: Removed RocksDBKeyLeaser, making RocksDBKey own its leased string buffer directly.
• COR-471: Improved filter optimization, rewriting `x IN [a]` to `x == a`, removing always-false/true branches, and deduplicating identical conditions.
• Fix shipped cmake file to make the build work on cmake >= 4.
• Upgraded FAISS to v1.14.2.
• COR-320: Vector search now supports projections, allowing covered queries to skip the document materialization step.
• COR-519: Vector index `factory` strings may now be combined with the scaling`nLists` specification (previously rejected).
• COR-529: Track vector index training memory against the global resource monitor and fail with `TRI_ERROR_RESOURCE_LIMIT` instead of OOM-killing the process.
• COR-467: Extended `compareAstNodes` to compare AST nodes that are not directly VPack-serializable (references, attribute accesses, function calls, operators).
• COR-466: Added constant folding empty array comparisons in `FILTER` nodes.
• COR-439: Make --server.storage-engine option obsolete.
• COR-452: Vector search filters may now reference variables from outside the FOR loop (subquery results, LET bindings, bind parameters).
• COR-444: Vector index training now draws a uniformly random training set via reservoir sampling (Algorithm L) over a full iteration of the collection, instead of keeping the first `nLists * 256` documents.
• COR-411: Introduced optimizer rule `replace-any-eq-with-in` that maps filter nodes of the form FILTER [exp1] ANY == exp2 into FILTER exp2 IN [exp1] to make them visible to downstream optimizer rules, including use-indexes, optimize-traversals, and sort-in-values.
• The vector index API now exposes a `resolvedNLists` field that shows the actual nLists value used by the trained index.
• COR-231: Vector index `nLists` now supports a tiered scaling specification alongside fixed integer values.

Full changelog can be found here: https://github.com/arangodb/arangodb/blob/v3.12.10/CHANGELOG.
 
Version 3.12.10 distributives can be obtained at https://www.arango.ai/downloads/.