[OSM-dev] 500 errors during TIGER upload
tom at compton.nu
Wed Sep 5 19:25:57 BST 2007
In message <1189012764.21340.20.camel at localhost>
Dave Hansen <dave at sr71.net> wrote:
> On Wed, 2007-09-05 at 18:13 +0100, Tom Hughes wrote:
> > In message <1189010129.21340.9.camel at localhost>
> > Dave Hansen <dave at sr71.net> wrote:
> > > Here's a more extensive list. I'm not sure, but I think these started
> > > around the time that the AND upload began. Let me know if you need more
> > > examples. I'm sure I'll see a few more.
> > My working hypothesis is that they are happening each time the
> > server restarts, which is every 10000 requests or something.
> Would this affect just a single request, or does it continue during a
> restart where the action may take several seconds? From what I just
> sent, it looks like they occurred over a 140 second period of time.
The problem is that lighttpd will assign each incoming request to
one of the backend fastcgi daemons and start writing it to the
network connection to that daemon even if the daemon is still
processing another request.
So when a daemon dies any request that is currently queued for
that daemon will be returned with a 500 error by lighttpd without
rails ever having seen it.
You can recognise this case because the 500 errors show up in
the lighttpd log but not in the rails log. That is what I was
seeing with the ones I looked at last night.
> > Unfortunately I'm not sure there is anything that can be done
> > about it while we're using FastCGI. I don't know if switching
> > to Mongrel would be any better or not without looking into it.
> Where does that 10,000 request limit come from? I've heard it
> mentioned, but I don't know about its origin.
config/environment.rb in the server code is where it is defined.
Tom Hughes (tom at compton.nu)
More information about the dev