Mailing-List: contact new-httpd-help@apache.org; run by ezmlm
Precedence: bulk
Reply-To: new-httpd@apache.org
Date: Tue, 16 May 2000 18:33:04 -0400
Message-Id: <200005162233.SAA07023@k5.localdomain>
From: Jeff Trawick <trawickj@bellsouth.net>
To: rbb@covalent.net
CC: new-httpd@apache.org
In-reply-to: <Pine.LNX.4.21.0005161407220.2839-100000@koj.rkbloom.net>
	(rbb@covalent.net)
Subject: Re: ap_wait_all_procs(), recent "reliable piped logs" changes to
 MPMs
Reply-to: trawickj@bellsouth.net
References: <Pine.LNX.4.21.0005161407220.2839-100000@koj.rkbloom.net>

> From: rbb@covalent.net
> X-Authentication-Warning: koj.rkbloom.net: rbb owned process doing -bs
> Date: Tue, 16 May 2000 14:09:54 -0400 (EDT)
> X-Sender: rbb@koj.rkbloom.net
> Content-Type: TEXT/PLAIN; charset=US-ASCII
> X-UIDL: 43d62eaaecc279012d21174519f381b0
> 
> 
> > Currently, the status parameter to wait_or_timeout() is never
> > initialized and so the accidental contents of that storage is used as
> > an array index in process_child_status().  mpmt_pthread has been
> > broken on OS/390 for the last couple of weeks due to this, and I just
> > got a chance to look at it.
> > 
> > (This little note is just to give Ryan a chance to say "I ran into a
> > little problem with that... you need to do XXX" before I go hacking.)
> 
> Actually, I removed the status variable all together for dexter, because
> it wasn't necessary.  It may not be necessary for the other MPM's either,
> I don't remember right now.  

It isn't "necessary" but it is useful for logging any non-zero exits.
Both prefork and mpmt_pthread have logic to log signals and bad exit
codes.  Why waste it?

>                              I believe the problem with returning a status
> from wait_all_procs is that some platforms don't support that option, I'm
> thinking Windows, but I could be wrong about that.

Maybe this Unix-specific code shouldn't be calling a routine that is
subject to such issues?

> This is one of the places where bringing all of the common code from the
> mpms together will help us in the long run.  I have no problem adding a
> status variable to ap_wait_all_procs, but we need to make sure it will
> actually do something on Windows and OS/2 first.  And I want to make sure
> we need to status variable before we begin on this project.
> 
> Ryan
> _______________________________________________________________________________
> Ryan Bloom                        	rbb@apache.org
> 406 29th St.
> San Francisco, CA 94131
> -------------------------------------------------------------------------------

What was improved with the change from waitpid() to
ap_wait_all_procs()?

We picked up the following negative characteristics:

1) occasional (?) segfaults
2) lack of trace of bad exit status or signals
3) storage leak (ap_proc_t created by ap_wait_all_procs)

-- 
Jeff Trawick | trawick@ibm.net | PGP public key at web site:
     http://www.geocities.com/SiliconValley/Park/9289/
          Born in Roswell... married an alien...