kubernetes - ReplicaSet fallback selector -
i'd define replicaset use node selector first incarnation of pods. if pod fails , re-created, i'd replicaset use node selector b new pod. in other words, i'd node selector uses pod generation number parameter.
the use-case fixed-size batch jobs, first incarnation of pod scheduled on cheap-but-unreliable node [think preemptible vms], fallback on expensive-and-reliable node can finish job.
i've looked preferredduringschedulingignoredduringexecution, https://kubernetes.io/docs/concepts/configuration/assign-pod-node, it's not quite need, node pools autoscaling. "preferred" constraint fail due lack of nodes, can brought if there pods scheduled on pool.
Comments
Post a Comment