v3.0.7

Performance: multi-hop $relation queries expand from their targets instead of checking every record of the label (2 hops on a 100K-node graph: minutes → milliseconds); filtered vector search is routed on the matching-record count and filters during the ANN walk; exact vector search, grouped aggregates and unindexed filters run in parallel parts with results independent of the thread count; a plain select runs as a listing; lower CSV ingest memory and write volume. Fixes: a crash while a WAL segment was being created no longer blocks the next start; a batch writing one property twice under two types keeps one value. Behaviour changes: $relation paths are trails, vectorSearch.pool is the scorable candidate count, vector-default-ef is 128, budget errors name their limit. Upgrade: the first start rewrites vector binding rows to a new binary form (migration 0001-binary-vector-bindings, one-way): take a backup before upgrading if you may need to return to 3.0.6. See CHANGELOG.

What's Changed

• release: RushDB 3.0.7 — planned traversals, cardinality-routed vector search, parallel reads by @1pxone in #434

Full Changelog: v3.0.6...v3.0.7