Retry moveMasterPodsOffNode on failure instead of aborting (#3179)

attemptToMoveMasterPodsOffNode's transient errors (e.g. no synced
standby available yet) were treated as terminal, so the retry loop
gave up after a single ~4-5s attempt instead of retrying over
master_pod_move_timeout. This left masters stuck on draining nodes
indefinitely once a PodDisruptionBudget started rejecting evictions.

Log the failure and return (false, nil) so the retry loop keeps
polling every minute until the timeout expires.

Fixes #3176

pkg/controller/node.go

@@ -156,7 +156,8 @@ func (c *Controller) moveMasterPodsOffNode(node *v1.Node) {
 		func() (bool, error) {
 			err := c.attemptToMoveMasterPodsOffNode(node)
 			if err != nil {
-				return false, err
+				c.logger.Warningf("attempt to move master pods off node %q failed, will retry: %v", node.Name, err)
+				return false, nil
 			}
 			return true, nil
 		},