Re: losing connections to FCGI procs

Ben Davenport (bpd@merc.com)
Thu, 5 Dec 1996 16:25:08 -0700

Message-Id: <v03007806aecd0292db7f@[204.255.152.44]>
In-Reply-To: <Pine.BSI.3.95.961205113306.10021G-100000@get.wired.com>
Date: Thu, 5 Dec 1996 16:25:08 -0700
To: fastcgi-developers@openmarket.com
From: Ben Davenport <bpd@merc.com>
Subject: Re: losing connections to FCGI procs

This is actually what I've been doing as well, though I've been completely
killing the server and restarting.  Doing a kill HUP doesn't seem to cut it
for me.  So what's the consensus on this?  Bug in mod_fastcgi?  Bug in
apache?  Both?

Meanwhile, I'm trying out the OpenMarket server.  It definitely seems less
responsive with the FastCGIs than Apache, though it seems more stable.
Anyone have positive/negative stories of using FastCGI with the OM server?
How does it perform under heavy load?  Yeah, I've seen the marketing hype
on the web site -- i'm much more interested in real-world performance.

Also, I have not seen information anywhere regarding how to determine the
number of OM server procs you should run given cpu power and web traffic.
Obviously it's going to be less than what you'd expect with apache, since
it's multithreaded... but how much less?

Sorry this is not wholly FastCGI related -- is there a mailing list for OM
server discussion?

-Ben

On 12/5/96, you wrote:

> the EXACT same thing happens for me (bianca.com)
>
> here is my short term hack to 'fix' the problem,
> until it is officially fixed.
>
> i have a daemon that continually tails the error_log
> looking for "Interruped system call" and  "Connection refused".
> if found, the daemon SIGHUPS the httpd,
> and continues tailing the error_log.
>
> as you pointed it,
> the problem may appear again within seconds/minutes,
> or not again for hours,
> but atleast hupping the httpd gets things flowing again.
>
> {freeform}
>
>

--
Ben Davenport   <bpd@merc.com>
Experience varies directly with equipment ruined.