That actually may work well, since dagster could be run by a job system with retries itself, that tracks the run ID and other params. Dagster being the “head” job