Nasty mean evil ugly SIGALRM kills parent, leaving orphaned children

[Wilfredo Sanchez <ws...@apple.com>]

  We have this bug involving Apache (1.3.5 and 1.3.6) on Mac OS X  
Server in which the parent process seems to be going poof with no  
apparent cause.  No message is being logged, and the children are not  
being terminated.

  After setting a breakpoint in the kernel on exit_1(), we finally  
got some data.  It appears that the parent is getting a SIGALRM,  
which it hasn't registered for, and therefore the default action is  
to go poof.  This is proving to be a hard bug to reproduce reliably,  
so I don't have better info yet.  Tracing the evil sender of the  
signal is a bit challenging.

  I'm betting that if there were some spurious alarm() calls in the  
parent, that this would have been noticed on other platforms, so my  
current guess is that there is some signal mangling going on in the  
kernel that's causing this.  That said, I'm wondering if this sound  
familiar to anyone (just in case), or if anyone has some good tricks  
for me to try in debugging it.

	-Fred


--
       Wilfredo Sanchez, wsanchez@apple.com
Apple Computer, Inc., Core Operating Systems / BSD
          Technical Lead, Darwin Project
   1 Infinite Loop, 302-4K, Cupertino, CA 95014