Longhorn v0.6.0 Upgrade: caveats about rolling back to v0.5.0

Background

1. Some users may fail to re-upgrade Longhorn from v0.5.0 to v0.6.x after rollback succeeds.

• It's due to a Helm bug: when a resource should be introduced by the new version already exists in the current cluster (e.g. due to the failure of installing v0.6.0 in the first place), the Helm upgrade will fail.

2. The uninstaller may get triggered during rollback and upgrade, which can cause potential loss of data.

• See issue#783.

Instruction

The instruction is applied only if:

1. Longhorn is rolled back/downgraded from v0.6.x to v0.5.0 or earlier.

• Don't matter if the previous v0.6.x upgrade succeeded or not.

Step:

The user need to clean up the resources (introduced by v0.6.x upgrade process):

kubectl patch -p '{"metadata":{"finalizers": null}}' crd instancemanagers.longhorn.rancher.io
kubectl delete crd instancemanagers.longhorn.rancher.io
kubectl -n longhorn-system delete cm longhorn-default-setting