If I use EKS on EC2 Spot instances with

k8s_job_executor

, and the spot instances get interrupted, how will the executor respond -- retry the step on a new instance? Treat it as a failure? (Is that something I would expect to be handled at the k8s control plane, or the job executor?)

op

/

step

pod interruption can be handled by setting a retry policy https://docs.dagster.io/concepts/ops-jobs-graphs/op-retries#retrypolicy run pod interruption can be handled by run level retries https://docs.dagster.io/deployment/run-retries

Thanks! So it will be seen as a failure from Dagster's point of view, but I can also use standard Dagster tools to make execution resilient to those failures.

ya i would expect our features around this to continue to improve, but generally we want an explicit opt in to retrying computations since its not implicit that they are safe to redo if interrupted unexpectedly