Hello Nginx experts,

We make heavy use of Nginx as a reverse-proxy/load-balancer. It communicates with Apache and Tornado hosts upstream and proxies them publicly on port 80/443 - pretty standard.

The problem is, when pinging the LB host, every 30 pings or so, a ping is completely dropped and the latency immediately jumps from <20ms to >1000ms, then after a few pings, calms down again.

We are receiving a lot of messages like:

2013/10/28 08:49:05 [error] 20612#0: *77590567 recv() failed (104: Connection reset by peer) while reading response header from upstream, client: 50.xx.xx.169, server: loadbalancer, request: "GET / HTTP/1.1", upstream: "http://10.xx.xx.84:8014/", host: "loadbalancer"

and:

2013/10/28 08:49:05 [error] 20612#0: *77590567 no live upstreams while connecting to upstream, client: 50.xx.xx.169, server: loadbalancer, request: "GET / HTTP/1.1", upstream: "http://api-read-frontends/", host: "loadbalancer"


Is there configuration in Nginx that could be causing this? We have also cusomized sysctl.conf to try and fix it, no luck so far. There's more info, ping dumps, and our sysctl file attached to this question: http://serverfault.com/questions/549273/diagnosing-packet-loss-high-latency-in-ubuntu

Thanks in advance, any help is immensely appreciated :) Nginx is awesome!