Can retries of a flow use no cache? I have a retry set in my flow decorator that always fails because we don't do any task result caching I get the error:

raise ValueError(f"Path {path} does not exist.")
ValueError: Path /root/.prefect/storage/<ID> does not exist.

This is with

prefect==2.18.3

- worker on k8s

hi @Leon Kozlowski - are you using containers here?

Yes

ope sorry, i missed that last sentence

can I just set

os.environ["PREFECT_TASKS_REFRESH_CACHE"] = "true"

at runtime?

there's a util,

temporary_settings

for patching settings at runtime, but that seems like maybe a bit more heavy-handed than should be necessary the best course of action imo, assuming you need to persist results for the sake of retries, is to define a

PREFECT_DEFAULT_RESULT_STORAGE_BLOCK

like

s3/my-bucket

so that your results aren't persisting to a container that dies (which is what I would guess is happening now)

I don't need to persist results

do you need retries that start from the last completed state?

I want retries to restart entire flow ideally

oh okay, i guess im confused why it's retrying to read a result path at all then

I think that is the default behavior

ah yeah in 2.x if you have retries, you are persisting results, thats why

FYI on this -- we overhauled this behavior in 3.x so now the non-existence of a result in storage is equivalent to the lack of a cache, and the task will rerun instead of error out. In general result storage and caching is more transparent and robust in 3.x

I'm not in a place to migrate to 3.0 unless I can guarantee backward compatibility with my worker - with this being a major version upgrade I would not suspect that to be true