Applied patch and it has been running for 15hours. No segfaults
concerning the issue this patch addresses.
BUT, seems like apache does segfault if graceful restart is done:
[Thu Aug 10 16:49:51 2006] [notice] Graceful restart requested, doing
restart
[Thu Aug 10 16:49:51 2006] [notice] SIGUSR1 received. Doing graceful
restart
[Thu Aug 10 16:49:51 2006] [notice] Digest: done
[Thu Aug 10 16:49:51 2006] [notice] LDAP: Built with OpenLDAP LDAP SDK
[Thu Aug 10 16:49:51 2006] [notice] LDAP: SSL support unavailable
[Thu Aug 10 16:49:51 2006] [notice] Apache/2.0.54 (xServuWHS) configured
-- resuming normal operations
[Thu Aug 10 16:49:51 2006] [warn] (peruser: pid=18200 uid=99 child=7)
child_main(): input available ... resetting socket.
[Thu Aug 10 16:49:51 2006] [warn] (peruser: pid=18200 uid=99 child=7)
check_pipe_of_death(): WATCH: die_now=0
[Thu Aug 10 16:49:51 2006] [warn] (peruser: pid=18200 uid=99 child=7)
child_main(): marked jmpbuffer
[Thu Aug 10 16:49:51 2006] [warn] (peruser: pid=18200 uid=99 child=7)
child_main(): calling process_socket()
[Thu Aug 10 16:49:51 2006] [warn] (peruser: pid=18200 uid=99 child=7)
process_socket(): Creating dummy connection to use the vhost lookup api
[Thu Aug 10 16:49:51 2006] [notice] child pid 18200 exit signal
Segmentation fault (11)
About 7-8 of segfaults seem to happen after (few seconds or less)
graceful restart. Probably all processors which are active at the moment
seem to crash and burn.
This does not seem to be big problem, but if someone feels bored then it
could be fixed and if not fixed as a feature , it could be fixed as a
little cleaner notice than segfault.
And Stefan,
many thanks for the patch and your help.
I'll just gone transfer over one 100GB/monthly traffic website and see
what happens. That should double the load apache has at the moment.
For now, let's hope it works and thanks again :)
Janno
Stefan Seufert wrote:
Hi Janno,
again thanks for the debugging and sorry for my sluggish responses. I'm
still away and only have sporadic internet access.
I'm not familiar with metuxmpm ways and gotchas so i can't really say
if peruser is better or worse solution than metuxmpm.
It depends on what you are trying to do. It is thread based which means
that everything using it must be thread-safe (PHP isn't by default).
Apart from that there is a reason why apache still has this
MaxRequestPerChild option. Modules tend to have leak memory. When a
process exits all memory is reclaimed, this is not true in case of a thread.
Back to the problem:
Can you apply the attached patch? (Please revert the previous patch
beforehand). It should make the segfaults go away. But it won't fix the
underlying problem (apache can't create a connection). Most probably
those requests which are segfaulting at the moment are some aborted
requests or similar. So simply closing them is the right thing to do. If
your users should start complaining about empty pages or other errors
then please come back to this list and we'll try to track down the root
cause (Because it seems to be hidden somewhere inside apache itself this
may take a few iterations).
Stefan
------------------------------------------------------------------------
_______________________________________________
Peruser mailing list
[email protected]
http://www.telana.com/mailman/listinfo/peruser
_______________________________________________
Peruser mailing list
[email protected]
http://www.telana.com/mailman/listinfo/peruser
