Frank van Vugt <ftm(dot)van(dot)vugt(at)foxi(dot)nl> writes:
>> You may care to try your test with CVS tip and see if it works better.
> I've retried this with a cvs-version of this afternoon (local
> time). The build stranded in plpgsql (I seem to need a more recent
> bison), but at that point libpq was already finished. This still
> resulted in more or less the same kind of errors.
Are you sure that each thread is accessing only its own PGconn object?
Other than that, not sure what to suggest. I get the impression that
something is clobbering memory with abandon. Do you have any
Purify-like tools to look for memory stomp problems?
regards, tom lane