[nginx] worker-shutdown-timeout causes issues for request longer than 10s

[Congelli501]

I had issues with the nginx ingress controller starting from 0.9-beta.12. Requests lasting more than 10 seconds were closed with a 499 status code in the nginx logs.

In fact, this was caused by the worker-shutdown-timeout config param, which defaults to 10s.

The ingress controller reloads every 3 seconds, and, because of this param, any request lasting more than 10s after a reload will be killed (the worker process will be shut down).
As a result, long running requests were killed after 10s (if the request starts right before a reload) to 13s (if the request starts right after a reload).

This also affects TCP connections (for TCP services, making them completely pointless).

In my opinion, it would be better to completely disable this parameter by default (the default in upstream nginx).

MR causing the problem: #1088 & #1261

[aledbf]

Closing. You can set a custom timeout using the key worker-shutdown-timeout in the configuration configmap. We cannot set this to 0 (by default) because this means some workers could live forever. This introduces new issues like hitting cpu and memory limits.

[aledbf]

The ingress controller reloads every 3 seconds, and, because of this param, any request lasting more than 10s after a reload will be killed (the worker process will be shut down).

This is not normal. Please open an issue with the log showing this behavior