Helm uninstall leaves px-operator namespace in (stuck) terminating state
Describe the bug The helm install process followed by a helm uninstall does not fully clean up all pixie resources in the v0.1.7 operator release. The OLM project added a csv-cleanup finalizer in v0.27.0 that causes the px-operator to get stuck in a terminating state.
To Reproduce Steps to reproduce the behavior:
- Install pixie via the helm install instructions
helm -n pl uninstall pixie- Notice that the px-operator is stuck in a terminating state due to the
operators.coreos.com/csv-cleanupfinalizer
Expected behavior
Helm uninstall cleans up the olm, px-operator and pl namespaces.
Logs I don't have the logs handy, but I will attach them to this issue when I trigger the problem again.
As a short term workaround, if OLM is installed prior to using the Pixie helm chart, the issue does not occur. This is because OLM won't be a part of the helm installation, which allows the px-operator finalizers defined within OLM to run successfully on helm uninstall.
I'm evaluating options for addressing this finalizer issue in the meantime.
Can confirm I am experiencing the same issue. Have also found that installing via helm, and when uninstalling, running:
px delete
against the specific cluster before helm uninstall has also worked without finalizers causing operator namespace to get stuck. But yes not ideal, preferable for helm to manage everything
I have a proof of concept on this branch that implements a potential fix.
That change adds an additional k8s Job that will delete the olm operator namespace (px-operator by default) on the helm pre-delete hook. This allows olm to stay running before helm starts deleting resources and ensures the csv-finalizer has time to run.
I'll be working to get a PR open for this once I get feedback on this approach and test out an operator rc build.
#2059 is nearing completion and I hope to have an operator release out with it in the next week.
#2059 is nearing completion and I hope to have an operator release out with it in the next week.
Thanks @ddelnano for the fix. Looking forward to it.
The v0.1.7 release is now available!