About replica rebuilding


#1

How controller detects replica’s status? If the controller who owns the volume failed at the same time with one replica, who will chooses the new replica and triger rebuilding?
Thanks.


#2

@ludehp

I assume you’re talking about Longhorn Engine controller, not the Kubernetes controller?

If the controller failed, the volume will be detached, we won’t able to rebuild the replica.

Controller is on the same node of the consumer of the volume, we don’t expect it to fail unless it’s a host down, in which case nothing on that node would survive anyway.

If Kubernetes detects the failure of the node and decide to restart the pod on another host, the controller should be started on that node as well. We’ve implemented this with CSI but haven’t release the feature yet.


#3

If Kubernetes detects the failure of the node and decide to restart the pod on another host, the controller should be started on that node as well. We’ve implemented this with CSI but haven’t release the feature yet.

who decides to restart pod on another host? auto by kubernetes?