I had a flow die 80% of the way a big mapped task. (It died due to a problem with my dask cluster that has been resolved.) Is there a way to restart the flow from where it left off? I'm not using cloud.

Are you using prefect server? Or just prefect core?

just core

Without a backend (like cloud/server) or use of

Result

objects for caching there's currently no way to rerun only the failed tasks in a failed flow if the failed tasks are mapped tasks. You can skip non-mapped tasks by passing in

task_states

to

flow.run

, but mapped tasks are all-or-nothing currently unless you use a backend or use result-based caching.

gotcha, good to know

@Ryan Abernathey This is something we should probably discuss on Monday! @Sean Harkins and I looked at it this week. It might be really useful for large dataset re-runs

if we are using cloud (as we plan), it should be easier