Re: server crash with "process 22821 releasing ProcSignal slot 32, but it contains 0"

On Mon, Jun 25, 2012 at 9:57 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Merlin Moncure <mmoncure(at)gmail(dot)com> writes:
>> 2012-06-25 09:08:08 CDT [postgres(at)ysanalysis_hes]: LOG:  could not
>> send data to client: Broken pipe
>> 2012-06-25 09:08:10 CDT [postgres(at)ysanalysis_hes]: LOG:  unexpected
>> EOF on client connection
>> 2012-06-25 09:08:10 CDT [postgres(at)ysanalysis_hes]: LOG:  process 22821
>> releasing ProcSignal slot 32, but it contains 0
>> 2012-06-25 09:08:10 CDT [postgres(at)ysanalysis_hes]: LOG:  failed to
>> find proc 0x7f48617e2ab0 in ProcArray
>> [and a bit later]
>> 2012-06-25 09:08:24 CDT [postgres(at)ysanalysis_hes]: FATAL:  latch already owned
>
> I think what we're looking at here is a screw-up in the process shutdown
> sequence.  Perhaps caused by bad recovery from an attempt to send an
> error message to the already-disconnected client; but that's just
> speculation, and it's hard to see how to get more info without a core
> dump.
>
> I wonder whether we shouldn't promote some or all of these three error
> cases to PANIC, as they certainly suggest shared-memory corruption.
> And if it did panic, we could hope to get a core dump for debugging
> purposes.

Ok, I'll look into reproducing the crash conditions. Unfortunately
this is a critical server and it crashed during a time sensitive
process. I can schedule a maintenance window though but it will have
to wait a bit.

merlin