The PostgreSQL Global Development Group released PostgreSQL 19 on November 19, 2026. The update targets administrators running large production systems: it lets them reclaim table space without taking tables offline, pin down slow query plans, and keep standby servers consistent with what an app just wrote.
Core team member Jonathan Katz said the release lets people run their workloads without interruption. Here is what stands out.
REPACK ends the VACUUM FULL trade-off
Tables with heavy update and delete traffic pile up unused space, called bloat, and bloat slows queries. The old remedies were VACUUM FULL and CLUSTER, and both blocked access to the table until they finished.
The new REPACK command reclaims that space and can reorder a table to match an index. Run it as REPACK (CONCURRENTLY) and it works without blocking reads or writes. REPACK replaces VACUUM FULL and CLUSTER, though both commands remain for compatibility.
Fixing a slow plan without touching the app
The query planner usually picks an efficient plan. As data changes, though, the chosen plan can shift, and sometimes the new one is slower.
The new pg_plan_advice extension lets you stabilize plan choices and try alternatives using a small “plan advice” language. A companion extension, pg_stash_advice, stores that advice and applies it automatically by query identifier. A DBA can fix one bad plan without editing application code.
Autovacuum and monitoring upgrades
Busy systems get help keeping up with cleanup:
- Autovacuum can vacuum indexes with parallel workers, set through the new autovacuum_max_parallel_workers setting.
- A new scoring system decides which tables autovacuum handles first.
- Queries that read a table can mark pages all-visible in the visibility map as they run, which leaves future vacuums less to do.
Monitoring gains three views. pg_stat_lock breaks statistics down by lock type, pg_stat_recovery reports recovery status, and pg_stat_autovacuum_scores reveals how autovacuum ranks tables. EXPLAIN ANALYZE gains an IO option, with a matching auto_explain setting, to report asynchronous I/O activity. The WAL written for full-page images now shows up in pg_stat_wal and EXPLAIN (ANALYZE, WAL).
Logical replication gets easier to start
Enabling logical replication used to mean restarting the server in many cases. In PostgreSQL 19, creating the first logical replication slot automatically raises the effective WAL level to logical. The new effective_wal_level setting shows what is in use.
The release also replicates sequence values, which you publish with FOR ALL SEQUENCES in CREATE PUBLICATION. That command gains an EXCEPT clause to leave out tables when publishing FOR ALL TABLES. CREATE SUBSCRIPTION can pull connection settings from a foreign server through a SERVER clause. A new retain_dead_tuples subscription option spots updates that collide with a delete happening at the same time, and pg_stat_subscription_stats tallies them.
Read-your-writes on standbys
Apps that send reads to standbys can now ask a standby to catch up first. A session runs WAIT FOR LSN with a log sequence number and pauses until the standby has replayed the write-ahead log that far. It can then query data it just wrote on the primary.
Developer features
Several additions trim code and round trips:
- INSERT … ON CONFLICT DO SELECT lets RETURNING hand back the existing row, which simplifies the “get or create” pattern and saves a network round trip.
- COPY can output JSON, including a single JSON array.
- SQL/JSON path picks up string methods: lower(), upper(), replace(), split_part() and btrim().
- Window functions lead(), lag(), first_value(), last_value() and nth_value() accept IGNORE NULLS to skip NULL values.
- random() can now produce dates and timestamps within a range.
For foreign tables, the new import_stats option lets postgres_fdw import statistics from the remote server during ANALYZE instead of sampling rows, which helps the planner choose better plans.
Speed gains and changed defaults
Some workloads get faster with no application changes. Faster foreign key checks made bulk inserts nearly 2x faster in testing. The planner can turn NOT IN clauses into anti-joins when no NULL values are present, and it can aggregate eagerly before joins to cut the number of rows processed.
The asynchronous I/O subsystem from PostgreSQL 18 also grows. With io_method set to worker, the default, the server now scales its I/O worker count between two new limits, io_min_workers and io_max_workers. They take over from io_workers. LISTEN and NOTIFY scale better for multi-channel workloads.
Two defaults change. Just-in-time compilation is now off, because the cost estimates that trigger it turned out to be unreliable. Teams that run many large analytical queries may want to turn it back on. And when built with LZ4 support, TOAST compression defaults to lz4.
Security changes to plan for
PostgreSQL 19 supports Server Name Indication on the server side, so one server can present different TLS certificates depending on the hostname a client uses. A new pg_hosts.conf file ties each hostname to its certificate and key. Another addition, password_expiration_warning_threshold, warns users ahead of a password expiring.
The release warns after a successful md5 password login. That warning can be switched off with md5_password_warnings. md5 authentication was marked deprecated in PostgreSQL 18 and is slated for removal later, so teams that need passwords should move to SCRAM. RADIUS authentication support is gone entirely.
What to check before upgrading
Anyone on RADIUS authentication has to pick another method first. Teams running big analytical queries should test with JIT enabled and disabled. Those still on md5 should plan the SCRAM migration. The full list of changes is in the project’s release notes, and PostgreSQL 19 ships with HTML documentation and man pages.
Source: PostgreSQL News