Hey Flavien - there’s a longer discussion here around why this isn’t as common a thing in Dagster as it is in Airflow:
https://dagster.slack.com/archives/C01U954MEER/p1625004255241500
We do support ways of running every solid in a pod (by configuring your pipeline to use a k8s executor) and you can definitely have a solid that runs a k8s pod. But it’s more common in Dagster pipelines to separate out the business logic of the step from the execution environment of the step