# PostgreSQL 19 is out with REPACK to clean tables without locking

> The November 19 release also adds query plan advice, read-your-writes on standbys, JSON output for COPY and sequence replication, and turns JIT off by default.

URL: https://claudegpt.si/news/postgresql-19-release-repack  
Published: 2026-11-20T06:56:33.740Z  
Author: ClaudeGPT Si Newsroom  
Section: AI Tools  
Topics: PostgreSQL, Databases, Open source, Developer tools  
Publisher: ClaudeGPT Si

## Key takeaways

- PostgreSQL 19 was released on November 19, 2026.
- The new REPACK (CONCURRENTLY) command removes table bloat without blocking reads or writes, replacing VACUUM FULL and CLUSTER.
- JIT is now off by default, RADIUS authentication is removed, and upgrading from an older major version needs a dump and restore, pg_upgrade or logical replication.

The PostgreSQL Global Development Group released [PostgreSQL 19](https://www.postgresql.org/docs/release/19.0/) on November 19, 2026. The headline feature is `REPACK`, a new command that cleans up bloated tables, and with its `CONCURRENTLY` option it does so without blocking reads or writes. The release also adds a way to pin query plans, read-your-writes consistency on standbys, and logical replication of sequences.

PostgreSQL runs underneath a large share of the web's applications, and lately underneath many AI products too, as the store for app data, embeddings and agent state. Most of what is new in 19 is about running it with less downtime and fewer surprises.

## REPACK replaces VACUUM FULL and CLUSTER

Tables that see many updates and deletes collect dead space, called bloat, which slows queries down. Until now the fix was `VACUUM FULL` or `CLUSTER`, and both locked the table until they finished. On a busy production table that meant scheduling a maintenance window.

`REPACK` reclaims that space and can reorder the table to match an index. `REPACK (CONCURRENTLY)` does it while the application keeps reading and writing. The [release notes](https://www.postgresql.org/docs/release/19.0/) say `REPACK` replaces the two older commands, which stay available for compatibility. A new `pg_stat_progress_repack` view shows how far a run has got.

Autovacuum, the background process that cleans up dead rows, also gets help. It can now vacuum indexes with parallel workers, set by `autovacuum_max_parallel_workers`, and it uses a new scoring system to decide which tables to clean first. The `pg_stat_autovacuum_scores` view shows that order.

## Plan advice fixes slow queries without code changes

The query planner sometimes switches to a worse plan as data grows. PostgreSQL 19 ships two extensions for that. [`pg_plan_advice`](https://www.postgresql.org/docs/19/pgplanadvice.html) lets you steer planner choices with a small "plan advice" language. [`pg_stash_advice`](https://www.postgresql.org/docs/19/pgstashadvice.html) stores that advice and applies it automatically to matching queries by their query ID, so a slow query can be fixed in the database instead of in the application.

## What developers get

- `INSERT ... ON CONFLICT DO SELECT ... RETURNING` returns the existing row when an insert conflicts. That turns "get or create" into one round trip.
- `WAIT FOR LSN` lets a session on a standby wait until it has replayed the primary's log up to a given point, so an app can read back what it just wrote.
- `COPY TO` can output JSON, including as a single JSON array.
- The window functions `lead()`, `lag()`, `first_value()`, `last_value()` and `nth_value()` accept `IGNORE NULLS`.
- JSON path gains string methods such as `lower()`, `upper()` and `replace()`.
- Logical replication can now copy sequence values (`FOR ALL SEQUENCES`), exclude tables with `EXCEPT`, and turns itself on without a server restart when the first logical slot is created.

On speed, the [announcement](https://www.postgresql.org/message-id/attachment/208469/release.en.md) says faster foreign key checks made bulk inserts nearly twice as fast in the project's testing. The planner converts more `NOT IN` and `LEFT JOIN` queries into anti-joins and can aggregate before a join. The asynchronous I/O system from PostgreSQL 18 now sizes its own worker pool between `io_min_workers` and `io_max_workers`.

## Changes to check before upgrading

Moving from an older major version needs a dump and restore, `pg_upgrade` or logical replication. Several defaults change:

- Just-in-time compilation (JIT) is now off by default, because its cost estimates proved unreliable. Sites with heavy analytical queries may want to turn it back on.
- New TOAST data is compressed with `lz4` instead of `pglz` when the server is built with LZ4.
- `max_locks_per_transaction` doubles to 128, and `log_lock_waits` is on.
- `io_workers` is gone, replaced by the new I/O worker settings.
- RADIUS authentication is removed, and `md5` password logins now log a warning. The project says to use SCRAM instead.
- `standard_conforming_strings` is forced on, so old dumps made with it off won't load correctly.
- Text search and `pg_trgm` indexes may need a reindex after the upgrade.

Managed providers are moving fast. [Databricks](https://www.databricks.com/blog/postgresql-19-available-lakebase) made PostgreSQL 19 available on its Lakebase service on release day, while warning that not every extension supports 19 yet and expecting a 19.1 minor release in February. Anyone running extensions should check each one before upgrading production, and test on 19.0 in staging first.

## Sources

- [PostgreSQL 19.0 release notes](https://www.postgresql.org/docs/release/19.0/), PostgreSQL Global Development Group
- [PostgreSQL 19 release announcement](https://www.postgresql.org/message-id/attachment/208469/release.en.md), PostgreSQL Global Development Group
- [pg_plan_advice](https://www.postgresql.org/docs/19/pgplanadvice.html), PostgreSQL Global Development Group
- [PostgreSQL 19 is available on Lakebase](https://www.databricks.com/blog/postgresql-19-available-lakebase), Databricks
