DEV Community

Floor Drees
Floor Drees

Posted on Edited on

It’s been a year: reflections on PostgreSQL 19

I’m not a hacker but this year’s PostgreSQL release is more discussed than any before. In my role I’m privileged enough to talk to the people that make PostgreSQL, both colleagues at EDB and community members I meet at conferences, or work with in the various committees and working groups I’m involved with.

I wanted to take a moment and look back.

Everyone and their agent

I gave a keynote presentation at Devopsdays Barcelona last week about the “contribution firehose of the Claude era”. Not to pick on Claude, these days everyone and their agents can submit bug reports and patches to open source projects, whether they know the library (and its stack) or not. And while models find very legitimate issues, a lot of the fixes are difficult to understand or merge without previous project alignment.

In the CloudNativePG project we adopted an AI Policy at the start of the year to try and discourage low effort / low value PRs.

We haven’t maybe seen quite as many random contributions to PostgreSQL because development doesn’t happen on GitHub, but we were certainly hit by the “vulnpocalypse”* as hard as other projects.

The update that was posted on postgresql.org after the publication of the second release candidate for PostgreSQL 19, details CVEs and bugs addressed and fixed, and changes since the last RC, for those interested. Yesterdays post focuses more on the high-level changes.

What happened to PostgreSQL?

Anyway, back to my keynote. I told the audience about how PostgreSQL follows a predictable annual release cycle for major versions and a quarterly schedule for minor updates, except that this year the release is delayed. And the release is Lite. Many highly anticipated features have been reverted, some of them quite late in the beta period. My colleague here at EDB, Core member Peter Eisentraut explained what happened on his blog.

What happened is that there was a long tail of relatively harmless issues that in past times would get fixed as they surface, but with LLM-assisted code review get found much faster. Even if each of them requires only a three-line fix, there is a fixed overhead to dealing with each of these.

Because of the aforementioned vulnpocalypse, many senior developers were busy between April (the start of feature freeze for PostgreSQL 19) and August (the most recent PostgreSQL security release) working on fixes for security issues. For context: there were 7 CVEs fixed in 2023, the same number in 2024 and in 2025, and in 2026 it was more like 95 since the start of the year.

Plus the complexity of a 30 year-old system has made some kinds of features extremely complex to get right. No one can manually test all edgecases, but LLM-assisted fuzzing can find problems really quickly. The community preferred trying to stick (close) to the release schedule and thus drop features, rather than delaying to get more fixes in.

I can highly recommend reading Peter’s blog post for his suggested structural changes for the PostgreSQL project moving forward. There’s a cascading challenge with more frequent releases as a suggested way forward, because of the work it puts on package managers and extension developers (mostly single maintainers made evident again by the recent pgBackRest scare).

How to deal with the new reality has been the subject of discussion at PGConf.DEV last May, and several workshops, mailing list threads and hallway chats after. It’s not hard to imagine it will be one of the main topics of discussion at FOSDEM during the committers meeting coming January.

Andrew Dunstan writes about how PostgreSQL security fixes have a home of their own since June 2026: a private git repository holding every pending fix as a stack of patches, and a private buildfarm that builds and tests that stack across every supported branch, continuously, before anything goes public. Timely work.

The good stuff

The above is just my observation as a concerned community member. For more information about some of the features and improvements spearheaded by my colleagues (there are 29 named contributors, people currently working at EDB), allow me to offer some suggestions for further reading/watching:

In “OAuth in Postgres 19 (and beyond)”, Jacob Champion looks at the incremental changes made this year to OAuth, the feature he worked on that got shipped with PostgreSQL 18. He and I had a conversation last year about how sometimes getting something into the project can be a marathon (not a sprint). In his post Jacob also looks forward to the next major release in terms of the improvements he’s planning for OAuth.

The REPACK (CONCURRENTLY) feature introduced in PostgreSQL 19 reclaims disk space and eliminates table bloat without causing application downtime. EDB’s Álvaro Herrera worked on the feature alongside Antonin Houska and Mihail Nikalayeu. His recent blog post takes you on the development journey - it’s as if you were there!

pg_plan_advice lets database teams describe, reproduce and constrain specific planner decisions, and the companion pg_stash_advice extension can store that advice and apply it automatically. Robert Haas implemented pg_plan_advice during his time at EDB and a recording of a talk he did about it earlier in the year lives on the PGConf.DEV channel for your on-demand viewing pleasure.

The great thing about working in the community is that I get to work with Robert on a project I can’t divulge details about yet, even if we don’t work at the same company anymore.

The other great thing is being able to go into other software ecosystems, sharing challenges we’re facing, and sharing best practices and solutions, or at least understanding.

We’ll be back at this again because insanity is doing the same thing over and over again and expecting different results ;)

Excuse me while I tell my kiddo he’s too young for the “other release” (GTA 6).

*Term describing the massive surge in software vulnerability discoveries and potential exploits driven by generative AI models.

Top comments (0)