PostgreSQL 19 will not ship this fall as planned: 53 features were reverted during an unusually long review cycle that started with last year's first beta, including SQL/PGQ property graphs and the long-awaited ALTER TABLE ... MERGE/SPLIT PARTITION(S) commands. The reversions were driven partly by AI-assisted tooling that found defects, including a correctness bug in GROUP BY ALL, that earlier review cycles likely would have missed until after release. For anyone planning a PostgreSQL upgrade this year, the question is no longer whether to move to 19, but whether to wait for it at all.
What changed
PostgreSQL 19 entered beta in June 2025 carrying an ambitious feature list, and the project spent the subsequent review cycle finding that several of the largest additions were not ready for a stable release. The 53 reversions announced ahead of Beta 4 (scheduled September 24, 2026) span everything from marquee features to small correctness fixes.
The highest-profile casualty is SQL/PGQ property graph query support, a SQL standard feature that would have let PostgreSQL run graph queries natively, pulled over concerns that its design and implementation weren't ready as a unit, not over any single isolated bug. ALTER TABLE ... MERGE/SPLIT PARTITION(S), a partition-management capability administrators have asked for across multiple release cycles, was abandoned after design issues in how it interacted with existing partition strategies surfaced too late to fix safely before release. Temporal range support via UPDATE/DELETE ... FOR PORTION OF was reverted for similar readiness reasons.
Some reversions are pure correctness fixes rather than scope cuts: GROUP BY ALL was found during testing to produce incorrect results in combination with certain ORDER BY clauses, the kind of bug that silently corrupts query results rather than crashing, exactly the category a database project cannot ship with. A planned default switch of TOAST compression to lz4 was pulled for unrelated build-infrastructure reasons rather than a defect in the feature itself.
What makes this cycle unusual is the tooling behind the discoveries. The post crediting AI-assisted analysis with surfacing several of these defects and generating reproducible test cases points to PostgreSQL 18's August 2026 patch release, which fixed 28 CVEs against a historical average of a couple per release, as the clearest evidence that this isn't a one-off: the project's review process is catching more before release, which means more work before ship, not fewer total defects over the software's life.
Who this affects
Anyone with a PostgreSQL 19 migration scheduled this year, for the property graph support, the new partition-management commands, or the temporal range features specifically, needs to replan around PostgreSQL 18, since none of those capabilities will ship until a still-undetermined future release. Teams that were simply planning to take the next major version for its usual mix of performance and replication improvements are less affected, since Beta 4 continues regardless and most of PostgreSQL 19's non-reverted work is intact.
Anyone maintaining a PostgreSQL fork, extension, or tooling that depended on the reverted features' in-progress APIs, including early adopters who built against beta behavior for GROUP BY ALL or the partition commands, should expect that code to need rework regardless of when 19 eventually ships, since the reverted implementations are not simply delayed but pulled pending redesign.
Verdict
Wait. PostgreSQL 18, patched as recently as August 2026, remains the version to deploy for anything going to production this year, nothing about the PostgreSQL 19 delay makes 18 less stable, and the review process that caused the delay is, if anything, a point in the project's favor rather than against it.
For teams specifically waiting on SQL/PGQ property graphs, the new partition-merge commands, or temporal range support, there is no substitute for those features on 18, and no committed date for when 19 will actually ship them, budget for months, not weeks, and don't schedule dependent work against a hard date.
Everyone else evaluating a PostgreSQL major-version upgrade should treat this delay as routine project hygiene rather than a red flag: a database project finding and fixing correctness bugs before a stable release, rather than after, is the outcome to want.
Tracked daily from official release feeds and vendor changelogs. Full archive: https://media.patentllm.org
Top comments (0)