The human-centric platform for production ML & AI

Outerbounds

Can I confirm that `@retry` (via argo &amp; k8s) should restart a step correctly if the pod is provisioned on a spot instance, and that instance is interrupted/reclaimed?

And a follow-up: is there any mechanism by which we can tell (either within the step runtime, or in other lifecycle hooks) the difference between a code exception and a interruption? If we wanted to retry on one but not the other

yes - @retry retries interrupted spot instances 

to avoid retrying errors in user code, see this thread for an example <https://outerbounds-community.slack.com/archives/C02116BBNTU/p1674195227085449|https://outerbounds-community.slack.com/archives/C02116BBNTU/p1674195227085449>