Hi, my task is mapped to some children tasks. if some of the children tasks fail in a run, is there a way to skip the successful ones and only re-run the failed? Can this be achieved by state handlers? thanks

When you mean re-run, are you clicking the restart button in the UI? Or are you kicking off a new flow run and only want to re-run the failed?

Also, if you configure automatic retries on your tasks, this should retry failed child task runs automatically without restarting successful child task runs

the later. i am thinking of the situation where there might be issues specific (like missing input files) to certain child tasks, so re-run is needed after manually fix that issue

You need caching , which will take effect across flow runs. You need to use templating so each mapped task has a unique file

so for @task(target="func_task_target.txt", checkpoint=True, result=LocalResult(dir="~/.prefect"), if there is no such func_task_target.txt in ~/.prefect, then prefect will create that txt file and write the return value of my task into it?

Yes exactly but for mapped tasks, you want different file names for each task

ok, for some reason, i don't see the txt being created in the dir

Are you using

flow.run()

or an agent run?

flow.run()

It won’t persist for

flow.run()

. Maybe if you have an env variable

PREFECT__FLOWS__CHECKPOINTING=true

i see, i tried this flow.run_config=LocalRun(env="PREFECT__FLOWS__CHECKPOINTING": True}), is there another way i can try?

flow.run()

does not respect the RunConfig so you really need an environment variable or to put it in the

config.toml

ok that worked. how does it work with cache_validator and cache_for? say if i pass a different set of inputs, will it bypass the target and invalidate the cache?

Targets and cache_for + cache validator are two independent caching mechanisms are you should only need one

got it, thank you

Yes it’s with a backend run

got it, thanks