A rollout restart, or scale down of haproxy causes 503 connection timeout errors

[pnmatich]

I'm using helm to run HaProxy ingress with autoscaling enabled (chart version 1.21.1). Whenever an HaProxy pod terminates (because of a scale down event, or a rollout restart), I start seeing 503 backend connection timeout errors for a few seconds.

I tried adding the following example config for graceful shutdown, but that did not resolve the issue

## Example preStop for graceful shutdown
lifecycle: {}
    preStop:
      exec:
        command: ["/bin/sh", "-c", "kill -USR1 $(pidof haproxy); while killall -0 haproxy; do sleep 1; done"]

[dkorunic]

This preStop action can't possibly be right due to s6 being used and s6-svc -1 /var/run/s6/services/haproxy should be used instead. Note that this issue is really not related to Helm Chart, but to Ingress Controller project.

[pnmatich]

Thanks @dkorunic I've opened a new issue in the Ingress Controller project. Maybe we should close this issue?

I'm not sure what you're referring to by s6 or s6-svc. Neither of those are in the preStop hook I tried.

[dkorunic]

Closing the ticket due to active https://github.com/haproxytech/kubernetes-ingress/issues/464.