Fabrice Scemama <fabrices@ximmo.ftd.fr> writes:
> [ vacuum appears to be coredumping ]
That's odd ... not so much that vacuum could be failing, which is
probably a garden-variety bug; but it sounds like the postmaster is
failing to do the system restart that it should do after one of the
backends fails. Is there anything showing up in the postmaster log
when this happens?
As for tracking the immediate vacuum problem, the failed backend
should have left a core dump file in the database directory
(.../data/base/DBNAME/core). Can you get a backtrace from that
with gdb? Something likegdb path/to/postgres path/to/corebtquit
should do.
regards, tom lane