Re: logical replication launcher crash on buildfarm

On 16/03/17 04:42, Andres Freund wrote:
> On 2017-03-15 20:28:33 -0700, Andres Freund wrote:
>> Hi,
>>
>> I just unstuck a bunch of my buildfarm animals. That triggered some
>> spurious failures (on piculet, calliphoridae, mylodon), but also one
>> that doesn't really look like that:
>> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=culicidae&dt=2017-03-16%2002%3A40%3A03
>>
>> with the pertinent point being:
>>
>> ================== stack trace: pgsql.build/src/test/regress/tmp_check/data/core ==================
>> [New LWP 1894]
>> [Thread debugging using libthread_db enabled]
>> Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
>> Core was generated by `postgres: bgworker: logical replication launcher '.
>> Program terminated with signal SIGSEGV, Segmentation fault.
>> #0 0x000055e265bff5e3 in ?? ()
>> #0 0x000055e265bff5e3 in ?? ()
>> #1 0x000055d3ccabed0d in StartBackgroundWorker () at /home/andres/build/buildfarm-culicidae/HEAD/pgsql.build/../pgsql/src/backend/postmaster/bgworker.c:792
>> #2 0x000055d3ccacf4fc in SubPostmasterMain (argc=3, argv=0x55d3cdbb71c0) at /home/andres/build/buildfarm-culicidae/HEAD/pgsql.build/../pgsql/src/backend/postmaster/postmaster.c:4878
>> #3 0x000055d3cca443ea in main (argc=3, argv=0x55d3cdbb71c0) at /home/andres/build/buildfarm-culicidae/HEAD/pgsql.build/../pgsql/src/backend/main/main.c:205
>>
>> it's possible that me killing things and upgrading caused this, but
>> given this is a backend running EXEC_BACKEND, I'm a bit suspicous that
>> it's more than that. The machine is a bit backed up at the moment, so
>> it'll probably be a while till it's at that animal/branch again,
>> otherwise I'd not have mentioned this.
>
> For some reason it ran again pretty soon. And I'm afraid it's indeed an
> issue:
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=culicidae&dt=2017-03-16%2003%3A30%3A02
>

Hmm, I tried with EXEC_BACKEND (and with --disable-spinlocks) and it
seems to work fine on my two machines. I don't see anything else
different on culicidae though. Sadly the backtrace is not that
informative either. I'll try to investigate more but it will take time...

--
Petr Jelinek http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services