Nginx + fair load balancer patch looping

[Grzegorz Nosek]

On Wed, Apr 16, 2008 at 11:17:09PM +0200, Alexander Staubo wrote:
> On 3/28/08, Grzegorz Nosek <grzegorz.nosek at gmail.com> wrote:
> > Looking at the strace (desperately trying to connect to a malfunctioning
> >  backend -- it closes the connection without sending a response) it
> >  certainly looks like the bug I fixed in the last commit.
> 
> Unfortunately, we are experiencing the problem even with the latest
> snapshot. What can we do?

That's a good question unfortunately ;)

I could use the debug_http output but that's an insane amount of data.
One thing to try would be increasing the priority of all upstream_fair
messages, e.g. by running

sed -i '-es/ngx_log_debug[0-9]*(NGX_LOG_DEBUG_HTTP/ngx_log_error(NGX_LOG_EMERG/g' \
 ngx_http_upstream_fair_module.c

(not tested, but that's the basic idea).

I can cook a patch if you aren't comfortable messing with C code. This
will cause all debug messages to get logged at emerg level so you'll be
able to collect it without enabling debug_http.

The upstream_fair module is pretty chatty too (several lines per
request) so the logs can grow but hopefully the amount of data will be
manageable.

Best regards,
 Grzegorz Nosek