I think you may be on to something, we are definitely seeing some weirdness
at a network level. I found an older thread of mine, forgot about it, this
is the same problem.

http://nginx.org/pipermail/nginx/2008-December/008667.html

We're looking at the tcp/ip traffic in more detail, and things do not look
correct . . .

On Tue, Dec 7, 2010 at 8:18 PM, Mikhail Mazursky <ash2kk@gmail.com> wrote:

> 2010/12/8 Neil Sheth <nsheth@gmail.com>:
> > Hello,
> >
> > Something I've been seeing for a long time here - I've posted long ago,
> but
> > was never able to find a solution. Looking at our access logs, I see
> tons
> > of lines like (multiple times a minute)
> > 123.45.6.789 - - [07/Dec/2010:19:05:51 -0600] "-" 400 0 "-" "-"
> >
> > I changed my error_log to info, and I see a lot of corresponding entries
> > like:
> > recv() failed (104: Connection reset by peer) while reading client
> request
> > line
> >
> > I also set the following, but it doesn't seem to have helped, as I see a
> lot
> > of posts about this issue potentially being caused by large cookies:
> > large_client_header_buffers 8 4k;
> >
> > Thoughts?
>
> Hello,
>
> we used to have issues like this on two of our servers when the router
> was setup incorrectly and some ip packets of a TCP connection were
> routed not to the right server. Check your network.
>
> _______________________________________________
> nginx mailing list
> nginx@nginx.org
> http://nginx.org/mailman/listinfo/nginx
>
_______________________________________________
nginx mailing list
nginx@nginx.org
http://nginx.org/mailman/listinfo/nginx