PostgreSQL 19 Released With Online Table Repack

PostgreSQL 19 has been released, the PostgreSQL Global Development Group announced on 19 November 2026. The new version of the world's most widely used open-source database arrives later than usual, after a release cycle in which AI-assisted code review uncovered bugs faster than the project's volunteers could fix them, and it brings a set of features aimed at keeping large databases running without downtime.
Recent major versions have shipped in late September, and PostgreSQL 18 came out on 25 September 2025. This time the general availability date was moved to 19 November, and three large features were pulled from the release late in its beta period. The headline additions that did ship focus on maintenance, query-plan stability and replication, rather than new syntax.
What is new in PostgreSQL 19
The most consequential change is the new REPACK command. Tables that see many updates and deletes collect dead space, known as bloat, which slows queries. Until now the fix was VACUUM FULL or CLUSTER, and both lock the table for the whole run. With REPACK (CONCURRENTLY), reads and writes keep working while the table is rebuilt, and rows can also be reordered to match an index, as CLUSTER did.
Administrators also gain finer control over query performance. The new pg_plan_advice extension lets database teams pin the planner's choices for specific queries, and the companion pg_stash_advice extension stores that advice and applies it automatically, so a query that suddenly slows down can be fixed without changing application code. Autovacuum can now process indexes with parallel workers, and a new scoring system decides which tables to vacuum first.
For applications that read from replicas, the new WAIT FOR LSN command delivers read-your-writes consistency: a standby session waits until it has replayed the write-ahead log up to a given log sequence number before running the query, so a read on the standby is guaranteed to see a row just written on the primary. Logical replication can now be enabled without a server restart, and developers get INSERT ... ON CONFLICT DO SELECT, JSON output from COPY, and IGNORE NULLS in window functions.
Why the release was late
The delay has a public explanation. Robert Haas, a long-time PostgreSQL committer, wrote in October that large language models let people outside the core team find bugs in the decades-old codebase far faster than before, while every fix still needs human review. Comparing the period from the April feature freeze to late September across two years, he found commits to the release branch rose from 724 in 2025 to 1,169 in 2026, an increase of more than 60 percent, while commits that also had to be back-patched to the previous stable branch rose from 248 to 433.
Between 27 August and 15 September, three large features were reverted: SQL Property Graph Queries, ALTER TABLE MERGE and SPLIT PARTITION, and UPDATE and DELETE ... FOR PORTION OF. Another committer, Peter Eisentraut, offered a more complicated picture, saying code quality and design problems in the reverted features themselves, and a dense wave of security work from April to August that pulled senior reviewers away, also played a part.
The release also changes several defaults. Just-in-time compilation is now off by default, RADIUS authentication is removed, the default for max_locks_per_transaction doubles from 64 to 128, and new TOAST data is compressed with LZ4. PostgreSQL 19 is supported for five years, until November 2031. Microsoft has already made it available in preview on Azure Database for PostgreSQL, and packages are in the official PGDG repositories. Teams in no rush are advised to test first and wait for the first point release and for extensions such as PostGIS and pgvector to publish builds for version 19.