PostgreSQL 19 ships two months late, and AI bug reports get the blame
PostgreSQL 19 is out, two months after the September slot its users had come to rely on. According to Simon Willison, the delay mostly comes down to a flood of AI-generated vulnerability reports the team had to work through before shipping.

PostgreSQL 19’s release was flagged on 19 November 2026 by Simon Willison on his blog. That post is the only source for this article, and it doesn’t cover everything: Willison comments on the schedule and one feature, not the full list of what’s new. So we’re staying inside those lines.
A September fixture, pushed back two months
Most years, Willison points out, PostgreSQL ships its new version in September. This one lands two months later. The explanation on offer, and it remains Willison’s own reading: the team had to absorb a burst of issues to fix, thrown up by the wave of AI-reported vulnerabilities, which the post calls, with no hedging whatsoever, an apocalypse.
The post doesn’t say how many reports came in, or what share of them turned out to be real. It does say those issues had to be dealt with, which is all it takes to move a release date.
WAIT FOR LSN, or reading back what you just wrote
Willison’s favourite new feature is aimed at applications that send their reads to standby servers. The problem is a classic: you write to the primary, you read from a replica, and the data hasn’t arrived yet. Your user has just updated their profile and gets the old one back, which is nobody’s idea of a confidence builder.
PostgreSQL 19 brings so-called “read-your-writes” consistency with the WAIT FOR LSN command. A session open on a standby can wait until that standby has replayed the write-ahead log (the WAL) up to a given position, the LSN, before running its query.
In practice, after a write on the primary, the application notes the current position:
SELECT pg_current_wal_insert_lsn();
It gets back a value like 0/0306EE20, which it stashes somewhere (Willison suggests a signed cookie). On the next request, over on the replica:
WAIT FOR LSN '0/0306EE20' WITH (TIMEOUT '0.1s');
The timeout is optional. If you set one and the replica hasn’t caught up in time, the command returns an error and the application decides what to do next.
Wikipedia was already doing it by hand
None of this is exotic. Willison describes first coming across the mechanism at Wikipedia, whose MediaWiki software carries a home-grown version built on MySQL replication: the primary’s binary log position is recorded in the session, then each request starts by waiting for the replica to reach it before any read. The current code relies on the MySQL function WAIT_FOR_EXECUTED_GTID_SET(), with SOURCE_POS_WAIT() as a fallback.
Put another way, PostgreSQL 19 builds into the engine what the teams who needed it had been duct-taping together in their application code.
What the delay tells you
Willison reckons the release was worth the wait. That leaves the reason for the delay, which is more instructive than the delay itself. A tool that can churn out vulnerability reports on a production line doesn’t, in the same breath, churn out the people who have to read them, reproduce them and fix them. Time spent on that triage is time not spent elsewhere, and a project whose calendar held from one year to the next has just demonstrated it, to the tune of two months.
The irony is that the release that finally shipped lets you wait gracefully for a replica to catch up. The maintainers are still waiting for a command that does the same for them.
Sources (1)
- PostgreSQL 19 Releasedsimonwillison.net
Written with AI assistance from the sources cited above, then reviewed and approved before publication by Sébastien Soulier.


