Every other 60 second timeouts

Hi,

We're using NGINX as a reverse proxy and are having difficulty getting consistent throughput. The service we have stood up behind the proxy is a NodeJS service running on CoreOS on an AWS ec2 instance, sitting behind an ELB. For the proxy, we're using NGINX as part of KONG running on a CoreOS Docker container. The version is openresty/1.9.3.1. We're using a simply proxy_pass to route the traffic though.

We have a few of these NGINX instances all configured the same. They'll work great for a while (1-4 days) and then the traffic passing through will enter a pattern where every other call proxies through instantaneously, the other calls will delay in NGINX for 60 seconds before making it out to the upstream service and getting a response back to proxy back to the caller.

I've configured other test upstream services to see if there's anything tied to the service being fronted, however as soon as I apply the config changes to NGINX, the report goes away for a while, so this path is a slow process and also hard to validate once we think we have a fix as we basically need to just wait 2-3X times the standard failure time to see if it fails again.

Has anyone seen this before or know how to fix it?

Thanks!
-Brian