No, I can't recreate the problem on demand. It look ~7 weeks running pg19b1 before we hit the issue. I already upgraded the instance, so let's wait to see what happens.
@Chao Li: thanks for pointing out that patch. On Thu, Jul 30, 2026 at 03:09:41PM -0400, Peter Geoghegan wrote: > On Wed, Jul 29, 2026 at 11:12 PM Justin Pryzby <[email protected]> wrote: > > Very possibly. It does seem plausible that the patch mentioned by Chao > > Li could explain this. Do you agree? > > That definitely seems possible, but hard to say with much confidence. > Even if it isn't the bug fix he mentioned, it could be another one. I > noticed that we're vacuuming the catalog index > pg_attribute_relid_attnam_index, and it's possible that that matters > here (this was probably triggered by churning through temp tables or > something, you didn't say anything about the workload you were > testing). > > Can you recreate the problem on demand using the same build? If you > can recreate the problem that way, it would certainly be useful to try > running the same steps against Postgres 19 git tip as of today, to see > if the problem goes away. That experiment seems most likely to be > informative/useful, at least to me.
