While doing opsweek, twice I found myself in a position that a Spark job was OOMing, yet the controlling Airflow job had no varprops (or DagProperties) for me to be able to easily tune the job. ( See here and here )
In this task we should:
- Identify which DAGs are missing the following Spark tunings:
driver_memory driver_cores executor_memory executor_cores
- Modify the dags so that these tunings are available.
Side note: It seems to me like the best time to do this would be as part of T336738, but leaving this as a separate ticket as this issue affects the opsweek sanity.