PostgreSQL 19 vs PostgreSQL 18: What's New and Should You Upgrade?

PostgreSQL 19 vs. PostgreSQL 18, compared: the new REPACK command, auto-scaling I/O workers, restart-free logical replication, and why SQL/PGQ graph queries were reverted before release.

The official PostgreSQL elephant logo, Slonik, in blue, dark blue, and yellow on a white background
The PostgreSQL elephant logo (Slonik), the project's official mark. Image: PostgreSQL Global Development Group.

PostgreSQL 19 is built around one trade: a more automated, online-maintainable server in exchange for three high-profile beta features — SQL/PGQ graph queries, GROUP BY ALL, and ALTER TABLE ... MERGE/SPLIT PARTITION — that were all pulled before release. Compared with PostgreSQL 18, the headline upgrade reasons are a new online-capable REPACK command that replaces VACUUM FULL/CLUSTER, auto-scaling asynchronous I/O workers, restart-free logical replication, and JIT compilation switching from on-by-default to off-by-default. November 19, 2026 is the general-availability date the PostgreSQL release team set for PostgreSQL 19 — but as of this writing the project's own site still listed PostgreSQL 19 RC 2 (November 12, 2026) as the newest build, so treat GA as "today, pending final confirmation" rather than a done deal, and double-check the official release notes before you schedule a production upgrade.

What's actually new in PostgreSQL 19

Per the official PostgreSQL 19 release notes, the release's real headline is operational: PostgreSQL has shipped a new REPACK command that "reclaims disk space and reorganizes table contents, which can improve query performance," explicitly as a replacement for VACUUM FULL and CLUSTER (both of which are retained for compatibility). Critically, REPACK (CONCURRENTLY) "allows repacking without blocking reads and writes to the table" — something neither VACUUM FULL nor CLUSTER could ever do on PostgreSQL 18 or earlier.

The second major theme is I/O automation, extending the asynchronous I/O subsystem PostgreSQL 18 introduced. PostgreSQL 19 removes the single io_workers setting and replaces it with four variables — io_min_workers, io_max_workers, io_worker_idle_timeout, and io_worker_launch_interval — that let the server scale its I/O worker pool automatically under the io_method=worker setting, instead of running a fixed number of workers around the clock.

Other confirmed additions in the official notes: optimizer work that lets more LEFT JOINs become ANTI JOINs and lets Memoize apply to anti-joins with unique inner sides; faster foreign-key checks; a switch of the default TOAST compression method from pglz to lz4; parallel autovacuum of indexes; a new WAIT command on standbys for read-your-writes consistency; IGNORE NULLS/RESPECT NULLS support for lead(), lag(), first_value(), last_value(), and nth_value(); and INSERT ... ON CONFLICT DO SELECT ... RETURNING. Our earlier coverage tracked this feature set as it firmed up across the beta and RC cycle.

The SQL/PGQ graph-query reversal — confirmed, not rumor

This is the fact that matters most if you've been planning around PostgreSQL 19's graph capabilities: SQL/PGQ property graph queries did not ship. Early PostgreSQL 19 beta builds did include CREATE PROPERTY GRAPH and GRAPH_TABLE syntax for querying relational data as a graph, and that preview is why so much "coming in PostgreSQL 19" content online still describes graph querying as a flagship feature. It no longer is.

We verified the removal directly against the primary source: the official pgsql-committers mailing list archive shows a commit from PostgreSQL contributor Peter Eisentraut, dated September 7, 2026, titled plainly "Revert SQL Property Graph Queries (SQL/PGQ)." The commit reverts 43 prior commits — starting with the original feature commit itself — and removes 15,859 lines of code, including the feature's documentation and regression tests, from the PostgreSQL master branch. Consistent with that revert, SQL/PGQ, CREATE PROPERTY GRAPH, and GRAPH_TABLE do not appear anywhere in the current official PostgreSQL 19 release notes.

SQL/PGQ wasn't the only late casualty. The same mailing list archive shows two more reverts on the REL_19_STABLE branch during the beta cycle:

  • GROUP BY ALL — reverted on July 17, 2026 after a post-commit review found it produced wrong results when a grouped column also appeared in an ORDER BY clause using non-default equality semantics. The commit message says the fix needed "too much code churn for late beta," so the feature was pulled with a note to "try again in v20."
  • ALTER TABLE ... MERGE/SPLIT PARTITION(S) — reverted on August 27, 2026, with the commit citing "multiple design issues which are too late to address in this release cycle."

If you've seen articles or AI-generated summaries (including some published before this site's own prior coverage finalized) that still list graph queries or GROUP BY ALL as PostgreSQL 19 features, they're describing a beta snapshot that no longer matches what shipped. None of the three reverted features are in the final PostgreSQL 19 release notes, and all three are explicitly pointed at a future PostgreSQL 20 instead.

What didn't make it into PostgreSQL 19
  • SQL/PGQ property graph queries (CREATE PROPERTY GRAPH, GRAPH_TABLE) — reverted September 7, 2026
  • GROUP BY ALL — reverted July 17, 2026 (correctness bug with ORDER BY semantics)
  • ALTER TABLE ... MERGE/SPLIT PARTITION(S) — reverted August 27, 2026 (design issues)

All three are described in their revert commits as candidates to retry in PostgreSQL 20.

Feature comparison: PostgreSQL 19 vs. PostgreSQL 18

The table below lines up the two releases on the changes that most affect day-to-day operation, based on the official release notes for PostgreSQL 19 and PostgreSQL 18.

AreaPostgreSQL 18PostgreSQL 19
Table/index maintenanceVACUUM FULL / CLUSTER (exclusive lock, blocks reads/writes)New REPACK command; REPACK (CONCURRENTLY) avoids blocking reads/writes
JIT compilation defaultEnabled by defaultDisabled by default (costing model found unreliable); re-enable manually for heavy analytics
Async I/O workersAsynchronous I/O subsystem introduced; fixed io_workers settingio_workers removed; auto-scaling via io_min_workers/io_max_workers
max_locks_per_transaction default64128 (lock size allocation changed — effectively double old settings to match prior capacity)
MD5 password authMarked deprecatedStill supported, but now warns after every successful MD5 login
Default TOAST compressionpglzlz4
Logical replication + wal_level=replicaRequires a restart to enableNo restart required
Sequence replicationNot supported via logical replicationSupported, including ALL SEQUENCES in publications
Graph queries (SQL/PGQ)Not availableStill not available — reverted before release; targeted for v20
GROUP BY ALLNot availableStill not available — reverted in beta; targeted for v20
UUID v7 (uuidv7())IntroducedCarried over unchanged
Virtual generated columnsIntroduced as the default behaviorCarried over unchanged

Performance and storage changes worth planning around

Three changes in this release are easy to miss in a changelog skim but will show up in production metrics. First, JIT is now off by default, reversing PostgreSQL 18's default; the release notes say this "requires sites that are doing many large analytical queries to manually enable JIT" via the jit setting if they were relying on the previous default. Second, default TOAST compression moves from pglz to lz4, which generally trades a little CPU for better compression ratios and faster decompression on large column values — existing tables keep their stored compression method until rewritten, so the change mainly affects new data and tables rebuilt with REPACK. Third, parallel autovacuum of indexes is now supported, which can shorten vacuum windows on tables with many indexes.

The REPACK command deserves its own callout because it changes a long-standing operational trade-off. VACUUM FULL and CLUSTER have always required an exclusive lock, so DBAs typically scheduled them for maintenance windows. REPACK (CONCURRENTLY), per the official notes, reclaims space and reorders table contents "without blocking reads and writes to the table," plus a new max_repack_replication_slots setting governs how it interacts with replication. That doesn't make it free — it still does real I/O and WAL work — but it removes the hard requirement to take tables offline for bloat cleanup.

Replication and availability

Two changes reduce operational friction around logical replication and standbys. Enabling logical replication no longer requires a server restart when wal_level is already set to replica — previously a restart was mandatory even if replication was only being turned on, not configured for the first time. Sequences can now be replicated logically, including an ALL SEQUENCES option for publications, closing a long-standing gap where sequence state had to be handled separately from table data. On the standby side, a new WAIT command gives applications a way to get read-your-writes consistency against a replica without rolling their own polling logic.

Security and authentication changes

PostgreSQL 19 continues the project's multi-release push away from MD5 password authentication, which PostgreSQL 18 first marked deprecated. PostgreSQL 19 adds a warning that fires after every successful MD5 login (disable it via md5_password_warnings if you're not ready to migrate yet), plus a new password_expiration_warning_threshold setting with a default seven-day warning window. The release notes also list the removal of RADIUS authentication support, forcing standard_conforming_strings permanently to on, and removal of the MULE_INTERNAL encoding — none of these are exotic in most modern deployments, but any instance still relying on RADIUS or non-standard string escaping will need to change configuration before upgrading.

Breaking changes to check before you upgrade

Beyond the security removals above, two changes can quietly break existing deployments if you don't re-check your configuration:

  • Lock table sizing. max_locks_per_transaction's default moved from 64 to 128, but the underlying lock size allocation also changed — the official notes describe this as effectively needing double the old setting value to retain the same lock capacity as before. If you explicitly set max_locks_per_transaction (common on systems with many partitions or large transactions), audit that value rather than assuming the new default covers you.
  • get_relation_info_hook removal. Any extension that hooked into this internal API will need updating or will fail to build/load against PostgreSQL 19.
  • btree_gist opclass change. GiST becomes the default opclass for inet/cidr types from btree_gist, which can change index choice for existing queries on those columns.

None of these are exotic edge cases — they're the kind of thing that shows up in a staging-environment dry run, which is the right place to catch them before a production cutover.

Should you upgrade to PostgreSQL 19?

For most teams already comfortable running recent PostgreSQL, the case for upgrading is operational rather than feature-driven. REPACK (CONCURRENTLY) alone is a meaningful win if table/index bloat has forced maintenance windows in the past, and the restart-free logical replication change removes a real source of downtime for teams that enable replication after initial deployment. Teams running JIT-dependent analytical workloads need to explicitly re-enable it post-upgrade, and anyone who built roadmap plans around native graph queries in Postgres needs to revise them — that capability is not in this release, full stop.

If you're currently on PostgreSQL 14, upgrading is no longer optional on any timeline: PostgreSQL 14 stopped receiving fixes on November 12, 2026, the same day PostgreSQL 19 RC 2 shipped, so you're already past end-of-life if you haven't moved. See our breakdown of what PostgreSQL 14's end of life means and how to upgrade for migration paths that apply whether you land on PostgreSQL 18 or 19.

As always with a .0 release, waiting for a first point release (19.1) before migrating business-critical production databases is a reasonable, conservative default — PostgreSQL's own release history, including the mid-cycle fixes bundled into the November 12 update release alongside RC 2, shows that early builds of a major version routinely pick up additional corrections shortly after the initial cut.

What's next

Two things to watch. First, final confirmation that PostgreSQL 19 has actually reached general availability: the RC1 announcement set November 19, 2026 as the planned GA date, conditional on no further delay or additional release candidate being needed, and our team tracked the date's earlier moves in our coverage of the RC2-driven schedule change — but as of publication, postgresql.org's own latest news listing still showed PostgreSQL 19 RC 2 (November 12, 2026) as the newest build, not a GA announcement. We'll update this piece once the official release notes carry a dated, non-placeholder release date. Second, expect SQL/PGQ graph queries, GROUP BY ALL, and partition merge/split to resurface as PostgreSQL 20 proposals — all three revert commits explicitly frame the cuts as "try again" rather than abandonment, so the design discussions that produced this release's reverts are the best early signal for what actually lands next time.

Frequently asked questions

See the FAQ section below for quick answers on the graph-query reversal, the confirmed GA date, JIT defaults, REPACK, lock settings, and PostgreSQL 14's end of life.

Frequently asked questions

Did PostgreSQL 19 ship with native graph query support (SQL/PGQ)?

No. SQL/PGQ property graph queries (CREATE PROPERTY GRAPH, GRAPH_TABLE) were part of early PostgreSQL 19 betas but were reverted from the master branch on September 7, 2026, per the official pgsql-committers mailing list archive. The feature does not appear in the final PostgreSQL 19 release notes, and the revert commit points to a possible retry in PostgreSQL 20.

What is the official PostgreSQL 19 general availability (GA) date?

The PostgreSQL 19 RC1 announcement set November 19, 2026 as the planned GA date. As of this article's publication, postgresql.org's news listing still showed PostgreSQL 19 RC 2 (November 12, 2026) as the newest build rather than a GA announcement, so confirm the dated release notes at postgresql.org/docs/19/release-19.html before relying on GA having shipped.

Is JIT compilation still enabled by default in PostgreSQL 19?

No. PostgreSQL 19 disables JIT by default, reversing PostgreSQL 18's on-by-default setting, because the project found its costing model unreliable. Sites running large analytical queries that benefited from JIT need to manually set jit=on after upgrading.

What does the new REPACK command do, and is it the same as VACUUM FULL?

REPACK is a new PostgreSQL 19 command that reclaims disk space and reorganizes table contents, explicitly replacing VACUUM FULL and CLUSTER (both are kept for compatibility). Unlike those commands, REPACK (CONCURRENTLY) can run without blocking reads and writes to the table.

Do I need to change max_locks_per_transaction after upgrading to PostgreSQL 19?

Check it. The default rose from 64 to 128, but PostgreSQL 19 also changed the underlying lock size allocation, so the official release notes say settings effectively need to be doubled to match the lock capacity they provided in previous releases.

Is PostgreSQL 14 still supported?

No. PostgreSQL 14 stopped receiving fixes on November 12, 2026, the same day PostgreSQL 19 RC 2 was released, according to postgresql.org's official release announcement. Anyone still on PostgreSQL 14 should plan an upgrade to PostgreSQL 18 or 19 immediately.

Sources

More on PostgreSQL 19 →PostgreSQLPostgreSQL 19PostgreSQL 18DatabaseOpen Source
Felix Moreau
Written byFelix Moreau

Felix Moreau writes Pandromeda's software coverage and how-to guides. He covers Windows, macOS and Linux updates, the apps people rely on, emulators and developer tools, and turns official documentation into clear, numbered steps that work on the current version.

More from Software & Guides

See all