The documentation here describes that node autoreplacement is not recommended for stateful services, because:
“when a node in a node pool loses connectivity with the cluster, its persistent volumes are destroyed, resulting in data loss for stateful applications”
Is this correct for stateful services that are not using host-mounted volumes? Specifically, our cluster has 1) EBS volume mounts, and 2) EFS integration. In both cases it seems the volumes would not be destroyed (presumably Rancher has marked the volumes to not be destroyed when the attached instance is).
Is auto replacement acceptable for these conditions, or am I missing something?