Open sozercan opened 6 years ago
This is difficult and subtle, and the short answer is:
Don't do that, use kubectl plugin helm delete releaseName
instead.
Currently the CRD is treated as the authoritative definition of what should be in the cluster (eg: if someone creates/removes HelmRelease objects while the controller is down, then on restart the controller should tell tiller to install/delete charts until the two match again). If we start trying to also discover what tiller is doing and reflect that back in the HelmRelease state, then now we have no authoritative source and can't reliably resolve differences.
This sucks because it's such an obvious thing to want to support :( I will have to think about other options..
Using
helm delete releaseName --tiller-namespace=kubeapps
from CLI should also delete the helmrelease