serviceaccounts "my-release-spark-operator" already exists

[Oneal65]

I reinstall operator, the error happen. Before that happened , I use helm uninstall spark-operator/spark-operator.
the command is :
helm install my-release spark-operator/spark-operator --namespace spark-operator --set enableBatchScheduler=true --set enableWebhook=true
the error message is:

Error: failed pre-install: warning: Hook pre-install spark-operator/templates/serviceaccount.yaml failed: serviceaccounts "my-release-spark-operator" already exists

[TiansuYu]

Second this issue. Major road blocker for short-lived deployments.

I also tried k delete serviceaccount my-release-spark-operator and k delete namespace my-namespace but it seems this error could not go away. The reason could be that for now helm does not mutate certain k8s objects properly (see helm/helm#5473 and related issues).

Would love to know any workaround. Cuz currently even if I manually delete everything under my namespace, including the service account, helm still complains about the issue shown above when I try to helm install xxx again. This applies to the spark-pi example in the quick-start tutorial.

An unmutable / non-upgradable deployment is by all means should be avoided. I think this issue should be addressed somehow.

[swartz-k]

Looks like those command need to exec.

kubectl -n RELEASE-NAMESPACE delete sa  my-release-spark-operator 
kubectl -n RELEASE-NAMESPACE delete clusterrole  my-release-spark-operator
kubectl -n RELEASE-NAMESPACE delete clusterrolebinding  my-release-spark-operator

[swartz-k]

I think templates/rbac.yaml need add before-hook-creation in annotation helm.sh/hook-delete-policy.

[TiansuYu]

Thx, I have found out the same yesterday.

[swartz-k]

Thx, I have found out the same yesterday.

Maybe you can make a PR to fix it.

[TiansuYu]

Thx, I have opened the PR here #1384.

[echang0929]

@swartz-k @TiansuYu @liyinan926, I found this as well, but there is no issue in the tags/v1beta2-1.2.2-3.0.0, i.e.

• remove operator's sa, clusterrole and clusterrolebinding when uninstalling this helm chart
• keep two crds, sparkapp and scheduledsparkapp, so that all of their resources in job namespace are kept.

Why? I mean, could we remove them when uninstallation rather than creation, just like previous tag versions?

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[github-actions]

This issue has been automatically closed because it has not had recent activity. Please comment "/reopen" to reopen it.