We need an easy way to gracefully migrate pods from one node to another

[mrosic]

oadm manage-node xyz --evacuate immediately kills all pods and will make your services go offline if they are backed by only one pod.

Version

1.1.4

Steps To Reproduce

1. oadm manage-node xyz --schedulable=false
2. oadm manage-node xyz --evacuate

Current Result

All pods are immediately terminated.
The current workaround is tedious: Scale DCs to two replicas, mark old node as unschedulable, delete pods from old node, scale DCs to one replica.

Expected Result

New pods are created on a different node first and only after the readiness checks have succeeded will a pod be terminated on the node that is being evacuated.

[pweil-]

This is not specific to evacuate and rather an issue of redundancy during any node failure that causes pods to terminate. Running multiple pods in a service is the right solution here.