I think a pod disruption budget might actually work here. It can select the spark driver pod using a label. Using that with a minAvailable value that's appropriate here could do it.
In a more general sense, we do plan on some future work to support driver recovery which should help long running jobs to restart without losing progress. On Tue, May 22, 2018, 7:55 AM purna pradeep <purna2prad...@gmail.com> wrote: > Hi, > > What would be the recommended approach to wait for spark driver pod to > complete the currently running job before it gets evicted to new nodes > while maintenance on the current node is goingon (kernel upgrade,hardware > maintenance etc..) using drain command > > I don’t think I can use PoDisruptionBudget as Spark pods deployment > yaml(s) is taken by Kubernetes > > Please suggest ! > > >
