PostgreSQL 19 Reaches General Availability After a Delayed, Feature-Reverting Beta Cycle
PostgreSQL 19 shipped November 19 after six reverted features and a slipped schedule; independent outlets and project contributors point to LLM-assisted code review surfacing more bugs than usual.
Overview
PostgreSQL 19 became generally available on November 19, 2026, according to the PostgreSQL Global Development Group. The release arrives after a cycle in which six previously planned features were pulled before launch and the schedule slipped past its original target. The Register reported in October that the version, once slated for release in late September or early October, was headed toward a release around November 19 instead — the date the project ultimately kept, independently confirmed by LWN and Percona.
Background
The Machine Herald has tracked this release since PostgreSQL 19 reached Beta 1 in June. By Beta 4, posted September 24, the project reverted six planned features. The Beta 4 announcement lists them exactly: a revert of “SQL/PGQ (property graph query) support”; a revert of “online enabling and disabling of data checksums”; a revert of support for temporal updates/deletes provided by the “FOR PORTION OF clause”; a revert of “ALTER TABLE ... MERGE PARTITIONS and ALTER TABLE ... SPLIT PARTITIONS”; removal of the pg_get_role_ddl(), pg_get_tablespace_ddl(), and pg_get_database_ddl() functions; and a revert of a change that had forced LC_COLLATE to C in the postmaster process. The announcement said the community “strongly believes that, first and foremost, PostgreSQL must be reliable” and “strives for a predictable release schedule,” without naming a specific cause for the reverts. At that point the project still projected a release candidate in “early October” with general availability that “may also occur in October.”
That October target slipped. The Herald reported in October that RC 1 moved general availability to November 19, and separately that PostgreSQL developer Robert Haas attributed the delay and reverts to LLM-found bugs, citing a jump of more than 60% in back-patched commits compared with the prior year. The Register’s reporting that same month surfaced a related but separately sourced account: PostgreSQL contributor Peter Eisentraut told the outlet that LLM-assisted code review had surfaced more bugs than usual during the beta period, which slowed the process of deciding what to ship. Eisentraut said the reverts were not the result of “vibe coding,” since the tools behind the increased bug-finding postdate when most of the affected features were originally written. The Register also reported that the project dropped somewhere between eight and ten significant features in total — a figure that is its own characterization of the broader cycle and is not the same count as the official six-item Beta 4 list. On the specific decision to pull SQL/PGQ graph-query support, the outlet reported that another PostgreSQL contributor, Tom Lane, had separately raised concerns about shipping it in version 19.
Current situation
The release that shipped on November 19 carries the features that survived that review process. The project’s announcement highlights the REPACK (CONCURRENTLY) command, which it says replaces VACUUM FULL and CLUSTER for reclaiming table space and can do so “without blocking reads or writes,” and the new pg_plan_advice extension for stabilizing query plan choices. It also adds “read-your-writes” consistency for standbys and an INSERT ... ON CONFLICT DO SELECT clause for a “get or create” pattern. Jonathan Katz, described in the announcement as “a member of the PostgreSQL core team,” said: “For over 30 years, the global open source community has built PostgreSQL into a database that organizations trust with their most critical data. PostgreSQL 19 continues that commitment to reliability, giving users new ways to manage their workloads without disruption and to build applications that span multiple systems.” The project’s own testing found that improved foreign-key check performance “made bulk inserts nearly 2x faster” — a figure that is the project’s own benchmark claim rather than an independently measured one. The release also disables just-in-time compilation by default, removes RADIUS authentication support, and continues the md5 password-authentication deprecation that began in PostgreSQL 18.
Independent coverage on release day largely confirmed the official feature list rather than adding new reporting. LWN’s release-day note names the same REPACK and pg_plan_advice additions and points back to its own September coverage of “the removal of several features initially slated for this release,” without elaborating further in the November piece. Percona, writing the day after general availability, called PostgreSQL 19 “finally released after multiple delays” and said “everyone already knows about the many reverts that happened.” Percona’s own list of features it had expected but that did not ship — “property graph queries, temporal updates and deletes (FOR PORTION OF), partition merging and splitting, online data checksum transitions, GROUP BY ALL” — overlaps with but is not identical to the official Beta 4 list; Percona called them “most of them ambitious additions” and said, “Hopefully we will see them again in PostgreSQL 20 or later, making that also an interesting release.” Percona also flagged that a bug in the new WAIT FOR LSN feature was not fixed in 19.0, writing that a fix “will be part of 19.1,” and recommended against using the feature in production until then “or at least always specifying a TIMEOUT.”
Implications
Two PostgreSQL contributors speaking in separate venues both pointed to AI-assisted bug-finding as a factor behind the longer cycle: Haas in the account the Herald covered in October, and Eisentraut in remarks to The Register. Neither source cited here claims the two are describing exactly the same mechanism, and this article does not combine their separate statements into a single account. Eisentraut’s point to The Register — that the reverts were not “vibe coding” because the tools behind the increased bug-finding postdate when the affected features were written — locates the AI effect specifically in review rather than in how the code was originally authored. The Beta 4 announcement’s own stated preference for reliability over a predictable schedule is consistent with that framing, even though the announcement itself did not name AI tooling as a cause.
Whether this cycle’s length becomes typical for future PostgreSQL releases is not addressed by any of the sources cited here. None of them report whether the release team expects PostgreSQL 20 to run on a similar timeline. What is established across multiple independent sources is that the specific features removed from PostgreSQL 19 were deferred rather than abandoned, with Percona voicing hope that items such as property graph queries and FOR PORTION OF temporal updates return “in PostgreSQL 20 or later.”
What We Don’t Know
- Whether PostgreSQL 20 will have a similarly long beta-to-GA cycle, or whether this one was unusual; none of the cited sources address the next release’s timeline.
- The precise relationship between The Register’s estimate of eight to ten dropped features and the official six-item Beta 4 list, since no cited source reconciles the two counts.
- A firm date for the
WAIT FOR LSNfix beyond Percona’s statement that it is targeted for the 19.1 point release.