Return-Path: owner-new-httpd Received: by taz.hyperreal.com (8.6.12/8.6.5) id IAA08675; Thu, 22 Jun 1995 08:32:55 -0700 Received: from gw0.telebase.com by taz.hyperreal.com (8.6.12/8.6.5) with ESMTP id IAA08669; Thu, 22 Jun 1995 08:32:53 -0700 Received: from wormhole.telebase.com by gw0.telebase.com id HAA10557 for ; Thu, 22 Jun 1995 07:33:14 -0400 Received: from khan.telebase.com (chuck@khan.telebase.com [192.132.57.215]) by wormhole.telebase.com (8.6.12/8.6.9.1) with ESMTP id LAA11805 for ; Thu, 22 Jun 1995 11:43:13 -0400 Received: (from chuck@localhost) by khan.telebase.com (8.6.12/8.6.9.1) id LAA09234 for new-httpd@hyperreal.com; Thu, 22 Jun 1995 11:35:22 -0400 From: Chuck Murcko Message-Id: <199506221535.LAA09234@telebase.com.> Subject: Re: leaking fd's To: new-httpd@hyperreal.com Date: Thu, 22 Jun 1995 11:35:22 -0400 (EDT) In-Reply-To: from "Brian Tao" at Jun 22, 95 11:50:10 am X-Mailer: ELM [version 2.4 PL24] Content-Type: text Content-Length: 1780 Sender: owner-new-httpd@apache.org Precedence: bulk Reply-To: new-httpd@apache.org Brian Tao liltingly intones: > > On Wed, 21 Jun 1995, Rob Hartill wrote: > > > > Re: zombies, after running unattended for 12 hours, 0.7.3 at cardiff > > now has 3 zombies. I remember seeing 50+ the other day, but these two > > observations might not have any real meaning.. we'll see. > > Okay, ran 0.73f for about 8.5 hours last night, 123340 requests > served. Started off with 20 servers instead of 50 (which finally > dawned on me might not be such a good idea with a machine with just 16 > megabytes ;-)), but still had 50 WebHound clients blasting away at it > from other machines. No fd leakage (even with CGI's), was running my > standard kernel that doesn't have special considerations for children > with a zillion open files or processes, but still had the zombie > children. > > ...more about several zombies, and missing active children... > I did some similar testing last night against BSDI 2.0 and FreeBSD 2.0. No fd leaks appeared. I couldn't get any zombies to appear, either, but both machines I had were pretty resource-limited. I'll try today on our BSDI server box here with 50 children and see if anything starts to zombify or disappear. I will also increase the number of connects between child respawns, but this looks better to me. I'll get a Solaris build done tomorrow (our servers are going down today for mondo Oracle database cloning, so who knows when they'll be back), and see what things look like. I'll also try to get some better instrumentation around the running server to see if there are any slow memory leaks not due to OS libs. No apparent fast leaks, though. This looks good so far. chuck Chuck Murcko Telebase Systems, Inc. Wayne PA chuck@telebase.com And now, on a lighter note: "You'll never be the man your mother was!"