As I've mentioned before on these bugs there's almost nothing we can do
because there are C library routines that require fds but do nothing
intelligent to report they can't get them. All the gethostby* functions
for example. They're likely going to find that it was one of my "continue
after errors" patches, so that'd be a good starting point.
I'm guessing something else is causing apache to allocation more fds than
it would otherwise. I ran into this recently. mod_log_hotwired was based
on the old mod_log_common and used the same per-server config to open
logs. Then I added a per-directory config to it and was surprised to
discover that even on the vhosts which didn't touch the per-directory
config I was getting a new per-server config for each of them. That meant
that instead of one site-wide log file handle I now had one for each
vhost. My solution was to indirectly point at the log file/fd and
preserve that across per-server merges.
I diagnosed this problem using lsof. Maybe they should try that and see
just what it has open.
Dean
On Tue, 29 Apr 1997, Brian Behlendorf wrote:
>
> This bug continues in 1.2b10. 179 domains each with an access and error log
> cause Apache to freeze up - not dump core, not fail to start, just freeze. On
> BSDI 2.1. After this message they confirmed that commenting out the
> access/error logs for each domain (i.e. having one log) fixed the problem.
> This was introduced in b7, and was not present in b6. They are downloading the
> CVS repository and are attempting to step through to see which patch broke it,
> but I thought you should know.
>
> Brian
>
> --=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
> brian@organic.com www.apache.org hyperreal.com http://www.organic.com/JOBS
>
> ---------- Forwarded message ----------
> Date: Tue, 29 Apr 1997 11:56:49 -0700 (PDT)
> From: Brian Behlendorf <brian@organic.com>
> To: Brandon Lee Poyner <poyner@neca.com>
> Cc: jmcfarli@neca.com
> Subject: Re: Apache 1.2b7 and 8 under BSDi 2.1
>
> On Tue, 29 Apr 1997, Brandon Lee Poyner wrote:
> > Thanks for the response.
> >
> > I grabbed 1.2b10 and unfortunately it's doing the same thing. The
> > only options I turned on were mod_status, mod_info, and mod_rewrite.
> >
> > I'm not sure if there is a threshold where this occurs. We do have
> > seperate logs for each domain, so that's 179 domains with an
> > access_log and error_log per domain, plus the main log files.
> >
> > We've tried single-threaded mode and setting MaxClients to 15 with
> > the same results. There isn't a core file, I end up killing the
> > parent httpd process.
> >
> > I've even tried 'limit'ing every option to unlimited.
> >
> > It seems as if my setup (many domains with log files) and some change
> > in beta7 has caused hangage. But that's just a guess :) I've
> > even increased kernel limits that beta6 didn't require.
> >
> > If you'd like to see the configuration files or twiddle with the
> > server to figure out why, that can be arranged.
>
> If you're sure you've unlimited everything (descriptors, openfiles, maxproc),
> then this is really weird. There are two things we can do to track this down:
>
> 1) Download the CVS repository and track every change between 1.2b6 and 1.2b7,
> testing each one to see if it's the cause of the problem.
>
> 2) Run it -X under gdb and watch for when it hangs.
>
> Does it say what system call it's waiting for when it freezes?
>
> If you want to give me an account on this system (which I'll probably share
> with one or two other apache developers) then we can assist in this process.
>
> What version of BSDI? The apache reference box is BSDI 2.1, so I'm surprised
> there are problems, but then again we only have 40 vhosts.
>
> Brian
>
> --=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
> brian@organic.com www.apache.org hyperreal.com http://www.organic.com/JOBS
>
>
>
|