# PostgreSQL 19 is out: REPACK CONCURRENTLY, plan advice, JIT off by default. What changes before you upgrade

> The PostgreSQL project released version 19 on November 19, weeks later than its usual autumn slot. It brings a non-blocking replacement for VACUUM FULL, stored query-plan advice and faster foreign-key bulk inserts, but also turns JIT off, drops RADIUS auth and starts warning on md5 logins.

Published Nov 21, 2026 by TechMeasured Staff · Repos · Canonical: https://techmeasured.com/repos/postgresql-19-released-repack-concurrently-jit-off-what-changes-before-upgrading

**PostgreSQL 19 is generally available.** The PostgreSQL Global Development Group [announced the release](https://www.postgresql.org/about/news/postgresql-19-released-3449/) on November 19, 2026. The headline additions are `REPACK`, including a non-blocking `REPACK CONCURRENTLY`, a new way to steer the query planner with stored advice, and a batch of autovacuum, replication and monitoring improvements.

Why it matters: PostgreSQL is the default database for a huge share of new applications, and this release changes several defaults that can affect performance or logins after an upgrade. It also arrived late. [Linux Compatible](https://www.linuxcompatible.org/story/postgresql-19-released-key-features-changes-and-upgrade-guide/) puts GA about six weeks behind the usual mid-October window, after RC1 on October 15, and reports that committers blamed AI-assisted testing for surfacing more bugs than usual.

## The headline features

According to the official announcement:

- **REPACK and REPACK CONCURRENTLY.** A new command that rebuilds bloated tables, replacing `VACUUM FULL` and `CLUSTER` (both remain for compatibility). The concurrent variant avoids blocking normal traffic. Linux Compatible adds that it uses logical decoding to capture changes during the rebuild, needs a primary key or index-based replica identity, and needs roughly twice the table's disk space while running.
- **pg_plan_advice and pg_stash_advice.** Extensions that let you give the planner advice and store it so it is applied automatically by query identifier, which helps pin down plans that regress.
- **Autovacuum.** Parallel index vacuuming via a new `autovacuum_max_parallel_workers` setting, a new prioritization score, and a `pg_stat_autovacuum_scores` view.
- **Replication.** The effective WAL level now rises to `logical` automatically when the first logical slot is created, so enabling logical replication no longer needs a restart. Sequences can be replicated (`FOR ALL SEQUENCES`), and `FOR ALL TABLES` gains an `EXCEPT` clause.
- **Read-your-writes on standbys** with `WAIT FOR LSN`.
- **SQL additions:** `INSERT ... ON CONFLICT DO SELECT` for get-or-create patterns, JSON output from `COPY`, `IGNORE NULLS` for window functions, and new SQL/JSON path string methods.
- **Performance:** foreign-key checks make bulk inserts "nearly 2x faster" in the project's testing, `NOT IN` can become an anti-join, and asynchronous I/O workers now scale between new `io_min_workers` and `io_max_workers` settings, which replace `io_workers`.
- **Observability:** new `pg_stat_lock` and `pg_stat_recovery` views, and an I/O option for `EXPLAIN ANALYZE`.

## What changes on upgrade

These are the behaviour changes the announcement calls out:

| Change | What to check |
|---|---|
| JIT compilation is **off by default** | If analytic queries relied on JIT, benchmark and set `jit = on` explicitly |
| Default TOAST compression is **lz4** (when built with LZ4) | Mostly a win; matters if you depend on pglz behaviour |
| **md5 password logins now warn** after success | md5 was deprecated in 18 and will be removed; move roles to SCRAM |
| **RADIUS authentication removed** | Replace before upgrading or logins will fail |
| `io_workers` replaced by `io_min_workers` / `io_max_workers` | Update tuned configs |

Linux Compatible lists further changes we could not confirm against the official release notes, which we were unable to read: `standard_conforming_strings` is always on (so old dumps made with it off won't load), default `inet`/`cidr` GiST opclasses change in a way that can make `pg_upgrade` refuse a cluster, `max_locks_per_transaction` doubles to 128, and MultiXact members widen to 64 bits. It also says "four" new io_worker settings, where the announcement names two. Read the release notes' migration section before running `pg_upgrade`.

## What didn't make it

Some coverage written during the beta, such as [Devart's overview](https://www.devart.com/blog/postgresql-19-release.html), lists SQL/PGQ graph queries and `GROUP BY ALL` as PostgreSQL 19 features. The final announcement mentions neither, and Linux Compatible reports they were pulled after feature freeze along with online partition merge/split and online checksum toggling, making them PostgreSQL 20 candidates. Don't plan around them yet.

## Who should upgrade, and when

New projects can start on 19. For production, the usual advice applies: wait for a minor release or two unless you need REPACK CONCURRENTLY or the autovacuum changes badly, test with JIT off, and clear md5 and RADIUS dependencies first. If you use a managed Postgres service, check when it will offer 19.


## Sources

1. [PostgreSQL — PostgreSQL 19 Released!](https://www.postgresql.org/about/news/postgresql-19-released-3449/)
2. [Linux Compatible — PostgreSQL 19 Released: Key Features, Changes and Upgrade Guide](https://www.linuxcompatible.org/story/postgresql-19-released-key-features-changes-and-upgrade-guide/)
3. [Devart — PostgreSQL 19 release overview (Beta 1 features)](https://www.devart.com/blog/postgresql-19-release.html)
