Re: Recent "could not register background process" buildfarm failures

On Tue, Aug 18, 2026 at 10:47:44PM +0200, Daniel Gustafsson wrote:
>> On 18 Aug 2026, at 19:44, Álvaro Herrera <alvherre(at)kurilemu(dot)de> wrote:
>>
>> On 2026-Aug-18, Tom Lane wrote:
>>
>>> # +ERROR: could not register background process
>>> # +HINT: You may need to increase "max_worker_processes".
>>>
>>> test_shm_mq itself hasn't changed meaningfully in ages,
>>> so it seems like some recent change somewhere else has increased
>>> the probability of not being able to get a worker process.
>>> Maybe autovacuum is eating too many workers? Any thoughts?
>>
>> Hm, autovacuum does not use workers from the bgworker pool AFAIK.
>> Repack concurrently does. However, I don't think we use repack
>> concurrently anywhere in the standard tests; we only run it in the
>> test_decoding and injection_point tests, which have separate instances.
>
> Online checksums also use bgworkers but it too should not be running here, and
> looking at it I didn't see evidence of it interfering. I'll have another look
> in the morning with fresh eyes.

I spent some time looking into this last night. AFAICT this is more likely
a buildfarm machine configuration problem than a code problem. For tayra,
the failures correlated exactly with a bump in meson_jobs to 72. It has
been good since it was lowered to 8. I did recently upgrade akepa's
buildfarm client and teach it to test more branches, but the failures
didn't occur until around a week later, so the correlation is weaker there.
I can't find any clues that point me towards a code problem, either. In
any case, it's the only machine with this problem at this point, so I
lowered its meson_jobs setting and will keep an eye on it.

--
nathan