PaveDB 0.9.4
English
PaveDB 0.9.4 hardens multi-worker operation, archive recovery, and search correctness.
Highlights
- Worker-safe metrics. Stores counters and bounded latency samples in the catalog database so multiple workers expose one consistent metrics view and share reset state.
- Safer archives. Preflights catalog, collection metadata, FAISS indexes, dimensions, and metrics before restore, and rolls back incomplete swaps.
- Concurrency coverage. Adds the scheduled chaos harness, controlled ingest shedding checks, catalog-to-disk invariants, and automatic failure triage.
- Search correctness. Preserves list-valued metadata filters and document/chunk override precedence, adds the SQL parity harness, and forwards raw-vector queries through shared search.
- Operational cleanup. Bounds hot collection caches, improves issue reconciliation, selects a supported Python interpreter automatically, and documents the Python SDK and Elixir client entry points.