[Feature Request] Configuration options for upstream keepalive requests and timeout

[ElvinEfendi]

Nginx 1.15.3 introduced keepalive_requests and keepalive_timeout for ngx_http_upstream_module. We currently have upstream-keepalive-connections to configure the number of keepalive connections to upstream. We should also support these two new directive.
One of the use cases is when the client keepalive timeout in the backend is less than the time Nginx waits for upstream keepalive connections. In this scenario the upstream can close the connection but Nginx would still think it's open and proxy a request through it, which result in "Connection reset by peer" error. https://theantway.com/2017/11/analyze-connection-reset-error-in-nginx-upstream-with-keep-alive-enabled/ has more info on this.

NGINX Ingress controller version:

Kubernetes version (use kubectl version):

Environment:

• Cloud provider or hardware configuration:
• OS (e.g. from /etc/os-release):
• Kernel (e.g. uname -a):
• Install tools:
• Others:

What happened:

What you expected to happen:

How to reproduce it (as minimally and precisely as possible):

Anything else we need to know:

[JordanP]

I am not sure If that could fix what I am experiencing in prod (Nginx behind a Google Cloud LB), but I have a lot of 502 errors (in the "stackdriver logging" console) with the following reason "backend-connection-closed-before-data-sent-to-client". I read and applied everything mentioned in this article: https://blog.percy.io/tuning-nginx-behind-google-cloud-platform-http-s-load-balancer-305982ddb340 (section 3. NGINX timeouts: fix a nasty 502 Bad Gateway race condition) but still I got many (2/3 per s) 502 errors

[ElvinEfendi]

I’d be surprised if these directives helped with that case since these are for upstream keepalive (in your case upstream is nginx).

But the issue sounds like the same just between GCLB and Nginx. I’d double check and make sure timeout in gclb is less than configured timeout in nginx.

Does it correlate with ingress-nginx deploys?

[JordanP]

I have a couple of high load microservices that use the "gce-ingress" and don't experience this issue. Yeah I'd love this to be a misconfig but so far nothing pops out, I just wanted to see if any body else experienced this issue. Anyway, I now get that this feature request is for upstream servers, not clients, thanks.

[ElvinEfendi]

@JordanP for downstream connections you can already do it using https://kubernetes.github.io/ingress-nginx/user-guide/nginx-configuration/configmap/#keep-alive

[diazjf]

@ElvinEfendi I can add this via a ConfigMap change.

[ElvinEfendi]

@diazjf that would be great!