0.25.3 - Paradedb — ContextQMD

New Features 🎉

Aggregates over NUMERIC columns are now pushed down into the ParadeDB index.
Queries containing Param and SubPlan expressions are now solved on the leader before MPP dispatch, so more plan shapes execute over MPP.
EXPLAIN (ANALYZE, VERBOSE) now attributes vector search buffer hits and reads to cluster routing versus cluster probing.

Faster counts for unfiltered and single-term-filter queries.
Cardinality aggregations over string fields now defer visibility checks, speeding up aggregations over fully-visible segments.

Fixed dropped pdb.agg filters, and improved the warnings emitted when an aggregate cannot be pushed down.
Unsortable top_hits.sort fields are now rejected at plan time instead of failing during execution.
partition_by is now preserved on AggregateScan for range co-partitioned joins.
Scaling a PgSearchScan node now declares only the partition assigned to it.
The final index build merge now runs even when the last insert flushed at the document cap.
Prebuilt binaries are now published for macOS 26 (Tahoe). macOS 14 (Sonoma) binaries are no longer published.

The full changelog is available on the GitHub Release.