Re: parallel regression test failure

Bruce Momjian <pgman(at)candle(dot)pha(dot)pa(dot)us> writes:
> I think you need a dual cpu machine to see the failures.

I was wondering about that myself, but we shouldn't fixate on that
assumption without more evidence. There could be some other factor
explaining why I can't reproduce it. A couple of questions for both
of you:
- what configure options are you using?
- can you reproduce the problem with serial tests (make installcheck)?
- exactly how repeatable is it --- when it fails, is it always at the
same places, or do the failures move around?

It would also be good to find out exactly where the failures are coming
from. Please try running the tests with LOG_ERROR_VERBOSITY set to
VERBOSE (probably the easiest way to hack this in make check's temp
installation is to modify src/backend/utils/misc/postgresql.conf.sample).
Then the postmaster log file created by make check will show the elog
calls' locations.

regards, tom lane