panic in propagate_priority w/ postgresql under heavy load

[John Baldwin]

On Thursday 01 September 2005 06:04 pm, Koen Martens wrote:
> John Baldwin wrote:
> > On Thursday 01 September 2005 01:02 pm, Koen Martens wrote:
> >>I've had a little chat with neologism on ircnet/#freebsd about this
> >>already, and done as he suggested: compile a debug kernel to obtain
> >>a stack trace.
> >
> > Can you reproduce it with a kernel that has INVARIANTS and
> > INVARIANT_SUPPORT on?  I see that you had WITNESS on, can you check to
> > see if there were any witness messages about sleepign with non-sleepable
> > locks held before the crash?
>
> I will do this when I get back. I did a grep -i on witness in the
> console log but this did not turn up anything suspicious (exact
> output pasted below). Also, i checked again the logs right before
> the crashes, nothing special output to console before the Kernel
> trap 12..
>
>
> voltaire# grep -i witness yin.log
> WARNING: WITNESS option enabled, expect reduced performance.
> witness_get: witness exhausted
> WARNING: WITNESS option enabled, expect reduced performance.
> witness_get: witness exhausted

This last means that witness had turned itself off because it had run out of 
resources.  Try bumping up the WITNESS_COUNT constant in 
sys/kern/subr_witness.c.

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org