On Fri, May 15, 2009 at 5:26 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Dave Page <dpage@pgadmin.org> writes:
>> Couldn't the callback have been called by another process though?
>
> Hmm, maybe, if the messages got to the log out of order. =A0Try
> reproducing it with %p added to log_line_prefix.
I've been playing with this for the last couple of hours, to no avail.
Looking at the log with PIDs, it certainly appears to be the crashing
backend that calls the atexit callback. I can't get a backtrace though
- if I attach the debugger before crashing, it breaks out at the
exception and won't continue. If I add an infinite loop in
atexit_callback to 'pause' the process I can attach and break, but for
some reason the debugger doesn't recognise the loop control variable,
so I can't modify the value to make it continue. That stops me even
getting the server to complete startup :-(
> Seems it's crashing inside perl_run(), which leaves us little wiser than
> before :-(
Well, there's no symbols included with ActivePerl, I can't get any
more info from there. I don't know if there are any with perl.org's
builds - but I'm too tired to look now anyway.
Any other ideas would be welcome!!
--=20
Dave Page
EnterpriseDB UK: http://www.enterprisedb.com