Hello! On Mon, Mar 09, 2015 at 04:28:22PM +0100, Wiebe Cazemier wrote:
> I have a question about sporadic long upstream response times > I'm seeing on (two of) our Nginx servers. It's kind of hard to > show and quantify, but I'll do my best. > > One is a Django Gunicorn server. We included the upstream > response time in the Nginx access log and wrote a script to > analyze them. What we see, is that on the login page of a > website (a page that does almost nothing) 95%-99% of 'GET > /customer/login/' requests are processed within about 50 ms. The > other few percent can take several seconds. Sometimes even 5s. > Our Munin graphs show no correlation in disk latency, cpu time, > memory use, etc. > > I also added an access log to Gunicorn, so that I can see how > long Gunicorn takes to process requests that Nginx thinks take > long. Gunicorn has 8 workers. It can be seen that there is > actually no delay in Gunicorn. For instance, Nginx sees this > (the long upstream response time is marked red, 3.042s): The 3s time suggests there is a packetloss somewhere. -- Maxim Dounin http://nginx.org/ _______________________________________________ nginx mailing list [email protected] http://mailman.nginx.org/mailman/listinfo/nginx
