如何设置 Kubeflow 使用抢占式虚拟机？

Question

I'm trying to set up a new Kubeflow deployment to use preemptible VMs. Referring to this doc , I see that there is some YAML to specify preemptible nodes but I am not sure where to put it in the myriad of YAML files generated by kfctl.sh . Meanwhile, going through the console, I see Preemptible nodes = disabled, but there is no way to change it.

How do I set the Kubeflow node pools to be preemptible?

Answer 1

There's no way to change a node-pool from non-preemptible to preemptible.

What you can do is create a new preemptible node-pool and migrate the workloads from the non-preemptible node-pool to the new one.

By the other hand, you can go ahead and recreate the GKE cluster with a preemprible node-pool.

Hope this helps!

Answer 2

The kfp.gcp extension module has gcp.use_preemptible_nodepool() . Check the preemptible sample .