Blocked Pod on recreation

[siryur]

TL;DR

When the chart tries to replace the running pod, it first tries to create the new pod before destroying the current one (which is the default behaviour of a Deployment without an explicit "strategy").

Since by default metric ports are allocated on the node, if a new pod is allocated to the same node, the deployment won't continue.

Expected behavior

Have a way to configure the existing pod to terminate instead of waiting first for the new pod to come up

Observed behavior

The deployment is on hold until some event removes the running pod.

Minimal working example

1. Create a new cluster with a single node
2. Deploy the chart
3. Edit the deployment to have a new pod come up

Log output

Additional information

Possible solution:

hcloud-cloud-controller-manager/chart/templates/deployment.yaml

Lines 7 to 12 in 0064e44

	spec:
	replicas: {{ .Values.replicaCount }}
	revisionHistoryLimit: 2
	selector:
	matchLabels:
	{{- include "hcloud-cloud-controller-manager.selectorLabels" . | nindent 6 }}

☝️ could become 👇

spec:
  {{- with .Values.strategy }}
  strategy:
    {{- toYaml . | nindent 8 }}
  {{- end }}
  replicas: {{ .Values.replicaCount }}
  revisionHistoryLimit: 2
  selector:
    matchLabels:
      {{- include "hcloud-cloud-controller-manager.selectorLabels" . | nindent 6 }}