Question for the Prefect gurus around performance and overheads. A work colleague has seen my adventures with prefect and was impressed. He had a pure dask pipeline that would build a large graphy consisting over ~40k function calls, each taking 1 to 5 seconds. When porting this over to a prefect workflow (replacing

@delayed

with

@task

and the like), is there: • an expected hit to the performance? The distributed dask schedular seems to be happy with the workflow, but would prefect? When do the task decorated functions report over the rest api about state transitions? Is this done by the task runner (my case a dask task runner) • any approaches out there to 'batch' the execution a little more, or advice around that?