cebernardi

to add some more context and explanation I've opened a draft PR for a possible solution (if this is confirmed as something to change by the maintainers): https://github.com/fluxcd/helm-controller/pull/368

@hiddeco from your comment it looks like you agree that the current behavior should change?

this is unfortunately not enough with AWS NLBs / ALBs as the problem is a long de-registration time, so the NLB would continue to serve traffic to the ingress

Please let me know if I got this correctly: on current master (not released yet), Istio introduced support for draining indefinitely. As it stands now (commit n. e00bce30daf81e6b29119bbe5b01b3f62c8097b3), if using...

Sorry, I'm not sure I understand. The `preStop` will call the endpoint `/drain`, which is served (after some wrapping) by `DrainNow`, which sets skip_exit: https://github.com/istio/istio/blob/master/pkg/envoy/agent.go#L148 would [this call](https://github.com/istio/istio/blob/master/pkg/envoy/agent.go#L148) return immediately,...

We are experiencing similar pain points with soft constraints, but in our use case, strictly honouring soft constraints is not only a cost problem, but it's creating stability issues. When...

> The cases that are really under question are things like preferential pod spread across hostname, or zone. We're definitely in this bucket. All of our use cases revolves around...

@ellistarn sorry, what is your proposal? [This one](https://github.com/kubernetes-sigs/karpenter/issues/666#issuecomment-2016044287)?

What are the downsides of making this behaviour configurable? If none of the options on the table are fully satisfying, maybe users should choose the scenario that best suites their...

I'm re-reading your message @jonathan-innis , and I think I miss some of the context about the original decision. Would you mind to expand a bit about the problem that...