PostgreSQL 19 has been available since November 19, 2026. The new major release introduces a table-reorganization command meant to replace VACUUM FULL and CLUSTER without blocking production, and lets teams turn on logical replication without restarting the instance. Several defaults change too, with a direct impact on applications that take a lot of locks.
REPACK replaces VACUUM FULL and CLUSTER
REPACK reorganizes a table’s contents to reclaim disk space and rewrite rows in a given order, without the limits of the two commands it replaces. Used with the CONCURRENTLY option, it runs without blocking reads or writes, while VACUUM FULL takes an exclusive lock for the whole operation. For a production table spanning tens of gigabytes, that is the difference between a reorg scheduled during an off-peak maintenance window and one that can finally run during business hours.
-- Reorganize a table without blocking reads or writes
REPACK (CONCURRENTLY) orders;
-- Reorganize while keeping a physical order close to an index
REPACK (CONCURRENTLY, USING_INDEX) orders USING INDEX orders_date_idx;
Logical replication, enabled without a restart
Through PostgreSQL 18, turning on logical replication on an instance already running with wal_level = replica required a restart to flip that setting to logical. PostgreSQL 19 drops that requirement: the change now applies live, which simplifies setting up a migration or a data pipeline triggered after the fact, with no dedicated maintenance window. Sequence values can now be replicated the same way table rows are, closing a frequent blind spot in low-downtime migrations.
Autovacuum, per-transaction locks, and the planner
Autovacuum gains a priority scoring system that determines the order in which tables are processed, rather than a plain sequential sweep: tables most prone to bloat or under the heaviest load move to the front of the queue. max_locks_per_transaction doubles by default, from 64 to 128, cutting the risk of the out of shared memory error on transactions that touch many partitions or indexes. On the planner side, pg_plan_advice lets teams steer some planner decisions declaratively, and pg_stash_advice applies that advice automatically to matching queries, a finer-grained alternative to the third-party hinting extensions used on other engines.
| Setting or behavior | Before PostgreSQL 19 | PostgreSQL 19 | Practical impact |
|---|---|---|---|
max_locks_per_transaction | 64 | 128 | Fewer out of shared memory errors on multi-partition transactions |
| JIT (just-in-time compilation) | On by default | Off by default | Must be re-enabled explicitly if analytical queries depended on it |
standard_conforming_strings | Configurable | Always on | Strings with an unescaped backslash now raise an error instead of being misread |
| RADIUS authentication | Supported | Removed | Migration to LDAP or an SSO provider is required before upgrading |
With REPACK (CONCURRENTLY), reorganizing a production table no longer means choosing between blocking writes and letting bloat pile up.
jit = on after upgrading, or risk a silent performance regression.The upgrade path still relies on the same tools as before: pg_dumpall, pg_upgrade, or logical replication, the latter now more relevant than ever for a low-downtime cutover since PostgreSQL 19 lets teams enable it without a prior restart. Teams already running their own self-hosted infrastructure should test pg_upgrade on a staging environment before any cutover, and check which third-party extensions are installed, since compatibility with version 19 isn’t always immediate right after release.
Key takeaways
PostgreSQL 19 targets the two operational pain points most often raised by teams running large tables: reorganizing a table without downtime via REPACK (CONCURRENTLY), and turning on logical replication without a restart. The doubling of max_locks_per_transaction and JIT being off by default both call for a targeted check before upgrading, especially for analytical workloads and multi-partition transactions.
JIT being off by default is the kind of change that breaks nothing visibly, but quietly drives up response time on two or three analytical queries long flagged as sensitive. Better to check it in cold blood against a known query set before upgrading, rather than find out through a latency alert in production — Simon Janvier.
Further reading: the official PostgreSQL 19.0 release notes on postgresql.org.
