Envoy Pods take 30 seconds to terminate

[rosskukulinski]

Is this a BUG REPORT, PERFORMANCE REPORT or FEATURE REQUEST?:

Bug report

What happened:

Deleting an Envoy pod as part of the Gimbal daemonset takes a long time to exit (appears to take 30-seconds, the default grace period). Once the grace period ends, the process is SIGKILLED

What you expected to happen:

I expected the Envoy process to respect the SIGTERM and shutdown.

[stevesloka]

I did some testing with this by customizing the terminationGracePeriodSeconds for the pod. I found if I set it to zero the pods immediately terminate, but it doesn't give envoy time to drain connections. Any other value seems to be ignored.

[rosskukulinski]

I think we should avoid modifying terminationGracePeriodSeconds -- the default 30 seconds is probably fine. I think this issue would be fixed by #144.