Oct 4, 2026 · 17 min · 10 segments
In this episode of Scaling Postgres, we discuss the reasons for all the PG 19 reversions, why open source wins, the cost of REPACK CONCURRENTLY and hint implementations. To get the show notes as well…
Creston JamisonHost
Our first piece of content: Are we reverting patches because of bugs found by AI? This is from vondre.me, and the first thing he mentions is that, you know, there was this blog post by Elizabeth Christensen that was saying basically AI is helping find bugs that's causing these features to get reverted.

Although he presents some data that shows, of course, a more nuanced view than that.

So he talks about the different patches reverted from Postgres 19, and he talks about the reasons why they were reverted, as well as a conclusion on whether it was AI or not.


And ultimately, I don't think it matters whether AI was the reason or not necessarily, but I think he represents a pretty good interpretation of what happened.

So we did look at reverts in the past, so we looked at PG14 through PG19 releases and when things tended to be reverted.

Now, what you can see is the slope of the PG19 is quite sharp, so there were...

So the significance of this is that it is a pretty steep slope and how late it happened.

So basically, people had already assumed features would probably get in at this time.

And in this early phase, when other releases had sharp increases in the number of reverts, PG19 was relatively stable during this time.

And they said day 275, which is around here, is when the feature freeze happens for each of the versions.

And then he looks at this other chart of the cumulative revert changes with regard to insertions and deletions, and you can see [chuckles] PG19 is definitely off the chart.

So my interpretation is this is lines of code, so at least the magnitude of PG19 has been significant.

It seems more than twice as much as any previous release, as well as a lot of the increase happened rather late.

Now, he ends with a question: Is it related to AI? But from that same post from Elizabeth Christensen, he points out the August patch for Postgres had 28 CVEs, so this is probably the primary reason that there was relatively low activity at the feature freeze mark compared to other releases, because people were probably busy reviewing and implementing security patches that the work that normally got completed here got shifted late.


Because maybe more features would've been able to reach a stable state and not be reverted if committers didn't have to focus on so many CVEs that were happening.

Our first piece of content: Are we reverting patches because of bugs found by AI? This is from vondre.me, and the first thing he mentions is that, you know, there was this blog post by Elizabeth Christensen that was saying basically AI is helping find bugs that's causing these features to get reverted.

Although he presents some data that shows, of course, a more nuanced view than that.

So he talks about the different patches reverted from Postgres 19, and he talks about the reasons why they were reverted, as well as a conclusion on whether it was AI or not.


And ultimately, I don't think it matters whether AI was the reason or not necessarily, but I think he represents a pretty good interpretation of what happened.

So we did look at reverts in the past, so we looked at PG14 through PG19 releases and when things tended to be reverted.

Now, what you can see is the slope of the PG19 is quite sharp, so there were...

So the significance of this is that it is a pretty steep slope and how late it happened.

So basically, people had already assumed features would probably get in at this time.

And in this early phase, when other releases had sharp increases in the number of reverts, PG19 was relatively stable during this time.

And they said day 275, which is around here, is when the feature freeze happens for each of the versions.

And then he looks at this other chart of the cumulative revert changes with regard to insertions and deletions, and you can see [chuckles] PG19 is definitely off the chart.

So my interpretation is this is lines of code, so at least the magnitude of PG19 has been significant.

It seems more than twice as much as any previous release, as well as a lot of the increase happened rather late.

Now, he ends with a question: Is it related to AI? But from that same post from Elizabeth Christensen, he points out the August patch for Postgres had 28 CVEs, so this is probably the primary reason that there was relatively low activity at the feature freeze mark compared to other releases, because people were probably busy reviewing and implementing security patches that the work that normally got completed here got shifted late.


Because maybe more features would've been able to reach a stable state and not be reverted if committers didn't have to focus on so many CVEs that were happening.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.