This message was deleted.

That's a complicated question. If the drain was done completely safely, the replicas should have been rebuilt away from the draining node and there should be no failed replicas. The policy for what to do when things are not so simple are called out in https://longhorn.io/docs/1.6.1/maintenance/maintenance/#node-drain-policy-recommendations and there are a variety of options. The leftovers might be a result of the current policy. Likewise, the action to take depends on • whether there are enough or close to enough healthy replicas, • whether the failed ones are nominally on the drained node, • whether the expectation is that the drained node will return or not. (They might be being saved in the anticipation that they will be used for a quicker rebuild than it would be from scratch.) If there are more replicas than the volume spec calls for, then the presence of failed ones that haven't been automatically cleaned up is likely a bug. If, in the user's opinion, there are enough healthy replicas to provide a reasonable level of redundancy, then one could attempt to manually delete a failed replica with

kubectl -n longhorn-system delete replica <name>

. If that results in the count being below the spec and if there is sufficient space elsewhere, the expectation is that Longhorn would start a full rebuild.

Thanks for such a detailled answer mate !