[ic] Occasional/random "Connection reset by peer" errors

[Daniel Davenport]

> -----Original Message-----
> From: interchange-users-bounces at icdevgroup.org 
> [mailto:interchange-users-bounces at icdevgroup.org] On Behalf 
> Of Jon Jensen
> Sent: 2006 September 12 -- Tuesday 5:17 PM
> To: interchange-users at icdevgroup.org
> Subject: Re: [ic] Occasional/random "Connection reset by peer" errors
> 
> On Tue, 12 Sep 2006, Daniel Davenport wrote:
> 
> > From time to time, some requests to IC pages cause a page to be 
> > displayed that reads "We're sorry, the Interchange server 
> is available 
> > due to a system error.  <snip> connection reset by peer".  
> (I'd paste 
> > the exact message, but someone here already knows the text of it -- 
> > plus, i can never seem to _make_ it show the error 
> message.)  No error 
> > appears in either the IC logs or in apache's log for that 
> site.  There 
> > were a few mentions of "interrupted system call" in the IC 
> error log, 
> > and setting $ENV{PERL_SIGNALS} to 'unsafe' seems to have 
> stopped that, 
> > but the "connection reset by peer" message still shows occasionally.
> 
> Where do you set PERL_SIGNALS to unsafe?

I put a line at the beginning of the first BEGIN block in
bin/interchange.


> > Using IC 5.3.1 (unix mode, via cgi + mod_rewrite -- not 
> > Interchange::Link), perl 5.8.8 (unthreaded), apache 2.0.53, mysql 
> > 3.23.58, and whatever perl module versions were available 
> about 2 weeks 
> > ago.
> 
> Is Interchange in PreFork, or fork-on-demand? What is the MaxServers 
> setting?

Appears to be fork-on-demand.  There's only one IC running, unless it's
serving a request.
Hrm.....it's in "low" traffic mode, which at the moment sets MaxServers
to 5.  I don't know whether i've ever seen it busy enough to need 5
copies running...but considering the machine's serving 15
catalogs...maybe turning it up isn't too bad an idea.

Another box, which seems to be running great, has MaxServers set to 0.
As far as i remember, i didn't do that (i've never had a reason to tweak
those settings, as until now they've always just worked).  I assume that
unlimits MaxServers -- which, although it works a lot better than the
current state of things, doesn't seem too safe to me.

Going to look at changing the MaxServers setting, and probably turn on
PreFork as it sounds like it'd help speed things up a bit.


> What operating system, distribution, kernel version?

I'm running Fedora Core 3 (Linux 2.6.12, with SMP) on that box.   


> Are you using suEXEC or a set-UID CGI binary? It sounds like 
> you're using the compiled C-based vlink, right?

Yep.  It's setuid to IC.  I have suexec, but it's generally too
restrictive for the setup we have here, so i don't use it.


> Can you see the "connection reset by peer" message in the 
> logs? If so, is 
> there any pattern to the URLs that it happens on?

The only messages i saw in the logs were things about "interrupted
system call".  That seems to have been cleared up by setting
PERL_SIGNALS to "unsafe", but the connection resets still happen -- and
no mention of them ever seems to appear in the logs.

And i haven't noticed any real pattern to when or where the problem
occurs.  At one time i thought it might be related to pages that hit the
database, but that's almost every page in most catalogs.


--
Daniel Davenport
New Age Digital
http://www.newagedigital.com