Cross posting here for visibility . https://flyte-org.slack.com/archives/C01P3B761A6/p1684866651162639

Not totaly sure, but I would guess interruptible nodes are what AWS calls spot instances. Those are compute ressources that could be gone before your job is done. But they are cheaper. In some usecases that are short you may prefer those and use the retry functionality of flyte in case a task was aborted because the instance was gone to early. Please correct me if I am mistaken here.

I ended up using default-node-selector with AWS managed node group with spot instances. Couldn’t get Interruptible-node-selector to work.