There is logic in the operator to remove the finalizer on the ConfigMap. But on an undeploy, the Operator gets removed before it can run. There is a pattern to hold off an undeploy and allow code to run for up to 30 secs (not forever), which would fix this. However, if pods are consuming extracted cache, it may not be safe to just delete all the GKMCache and ClusterGKMCache instances. So this requires more investigation.
There is logic in the operator to remove the finalizer on the ConfigMap. But on an undeploy, the Operator gets removed before it can run. There is a pattern to hold off an undeploy and allow code to run for up to 30 secs (not forever), which would fix this. However, if pods are consuming extracted cache, it may not be safe to just delete all the GKMCache and ClusterGKMCache instances. So this requires more investigation.