Re: BUG #16665: Segmentation fault

On Sun, Oct 11, 2020 at 6:37 PM Pavel Stehule <pavel.stehule@gmail.com> wrote:
> ne 11. 10. 2020 v 3:13 odesílatel Raphael Megzari <raphael@megzari.com> napsal:
>>
>> ok here is the backtrace that I got
>>
>> Let me know if you need anything else
>
> please, can you install debug symbols for Postgres. In this backtrace is not detail information.
>
> It's strange so Postgres crashes in LWLockAttemptLock

Yeah that's pretty weird.  It's as if it has a bad address for the
lock (core data structures corrupted, somehow we got a crazy buffer
number from a corrupted buffer mapping table?), or the mapping just
went away (systemd removed it?  docker/lxc/whatever removed it?), but
then I suppose there'd be lots of different failure modes, so it's
probably not that.  Hmm.  In the case we see, it's a parallel query,
crashing in the leader process under ExecGather().  Is it always like
that?  Is it possible to see any variables in the core file?  For
example, looking in the frame for ReadBuffer_common:

frame 2
print (unsigned int) blockNum
print &BufferDescriptors
print bufHdr
print (int) NBuffers

> Maybe there is problem in https://github.com/postgres/postgres/commit/bb16aba50c9492490a0b57e600a932798f45cd4f
>
> Do you use serializable level of transaction isolations?

Hmm, I don't see anything in the stack trace or config file that
implies that.  It was released in 12.