Message-Id: <199704210220.VAA20708@sierra.zyzzyva.com>
To: new-httpd@apache.org
Subject: Re: [STATUS] Sat Apr 19 21:05:18 PDT 1997 
In-reply-to: dgaudet's message of Sun, 20 Apr 1997 15:55:29 -0700.
         <Pine.LNX.3.95dg3.970420155238.14623D-100000@twinlark.arctic.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Sun, 20 Apr 1997 21:20:17 -0500
From: Randy Terbush <randy@zyzzyva.com>
Sender: new-httpd-owner@apache.org
Precedence: bulk
Reply-To: new-httpd@apache.org

> On Sun, 20 Apr 1997, Randy Terbush wrote:
> > >   * Listen and serialization problem should at least be documented,
> > >     perhaps solved.  See
> > >     <Pine.LNX.3.95dg3.970418205548.14942C-100000@twinlark.arctic.org>
> > >     Possibly related PR#s: 361, 299, 256
> > 
> > Serialization of the server on FreeBSD 2.2 has _reduced_ the amount 
> > of swap growth mainly through reduction of processes. I do think 
> > that some children are getting buried by being left in a queue 
> > state that does not allow them to accept.
> 
> I've asked the submitter of 361 if USE_FCNTL_SERIALIZED_ACCEPT helps them.
> They say they get a hang every 24 hours so I should know something tomorrow.

FWIW - FreeBSD needs USE_FLOCK_SERIALIZED_ACCEPT if you try it.

> Memory leaks -- you might want to try my graceful restart take 4.

I'm running it. What I am seeing is a little different than just 
process size growth. Some children seem to get in a state where 
they never accept a connection. The process gets swapped out and 
another created in its place. Eventually, I have lots of inactive 
children that just consume swap. Sending a HUP to the server will 
clean things up and reduce the swap usage by over 100MB. I'm 
running around 50 separate server parents on a single server.
If I were to speculate, it seems that resources used by children 
that are killed off do not get freed.

> > >   * SEGV under FreeBSD with lingering_close [and proxy?]
> > > 	Randy hasn't seen any coredumps in the last 24 hours.
> > 
> > Still no problems in this area.
> 
> I forget what fixed it.  Is it something we need to doc, or just close
> this out?

My problems were caused by a PHP problem that Rasmus has fixed.