Обсуждение: Correct SQLSTATE for ENOMEM in file access

Looking through the logs of some server that were experiencing out of
memory errors, I noticed that errcode_for_file_access() reports
ERRCODE_INTERNAL_ERROR for ENOMEM, while the correct SQLSTATE for this
should probably be ERRCODE_OUT_OF_MEMORY. Attached is a small patch to
fix this.

---
Alexander Kuzmenkov
Timescale

Alexander Kuzmenkov <akuzmenkov@timescale.com> writes:
> Looking through the logs of some server that were experiencing out of
> memory errors, I noticed that errcode_for_file_access() reports
> ERRCODE_INTERNAL_ERROR for ENOMEM, while the correct SQLSTATE for this
> should probably be ERRCODE_OUT_OF_MEMORY. Attached is a small patch to
> fix this.

Hmm, do you think this is actually reachable?  AFAIK we should only be
calling errcode_for_file_access() after functions that are unlikely to
report ENOMEM.

            regards, tom lane

On Fri, Feb 2, 2024 at 8:12 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Hmm, do you think this is actually reachable?  AFAIK we should only be
> calling errcode_for_file_access() after functions that are unlikely to
> report ENOMEM.

It's reachable, that's how I noticed. I'm seeing logs like "XX000:
could not load library \"/usr/lib/postgresql/15/lib/plpgsql.so\": out
of memory" from internal_load_library and so on. Not sure what is the
exact configuration required to reproduce this, probably at least the
overcommit should be disabled.

Alexander Kuzmenkov <akuzmenkov@timescale.com> writes:
> On Fri, Feb 2, 2024 at 8:12 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Hmm, do you think this is actually reachable?  AFAIK we should only be
>> calling errcode_for_file_access() after functions that are unlikely to
>> report ENOMEM.

> It's reachable, that's how I noticed. I'm seeing logs like "XX000:
> could not load library \"/usr/lib/postgresql/15/lib/plpgsql.so\": out
> of memory" from internal_load_library and so on. Not sure what is the
> exact configuration required to reproduce this, probably at least the
> overcommit should be disabled.

OK, can't argue with experimental evidence ;-)

            regards, tom lane