Short answer: In Prefect 3.x, caching is on by default for tasks. When caching is enabled, Prefect tries to compute a cache key by serializing your task inputs. That happens even if you didn’t set

persist_result=True

. If any input isn’t serializable (e.g., objects holding thread locks like your

ResourceNotifyDecorator

), Prefect logs the message you’re seeing and skips persistence for that run. More detail: - By default, tasks use a cache policy that includes inputs, source, and run id. Computing that cache key requires serializing your inputs. - When a cache policy is active (default), Prefect will auto-enable result persistence to support caching. - If serialization fails, you’ll see “Error encountered when computing cache key – result will not be persisted…”, but the task still runs normally; it just won’t reuse or persist a cached result. How to fix it (pick one): 1) Don’t need caching? Disable it for that task

from prefect import task
from prefect.cache_policies import NO_CACHE

@task(cache_policy=NO_CACHE)
def my_task(...):
    ...

Or explicitly turn off result persistence (also disables caching):

from prefect import task

@task(persist_result=False)
def my_task(...):
    ...

2) Want caching but exclude the problematic parameter - Using the cache policy to ignore a specific input name:

from prefect import task
from prefect.cache_policies import INPUTS

@task(cache_policy=INPUTS - "resource")  # replace "resource" with your param name
def my_task(data, resource):
    ...

- Or define a custom cache key function that only hashes safe inputs:

from prefect import task
from prefect.utilities.hashing import hash_objects

def my_cache_key_fn(ctx, params):
    safe = {k: v for k, v in params.items() if k != "resource"}  # exclude non-serializable
    return hash_objects(safe, raise_on_failure=True)

@task(cache_key_fn=my_cache_key_fn)
def my_task(data, resource):
    ...

3) Restructure so non-serializable objects aren’t passed as task parameters - Initialize those resources inside the task, or pass just primitive config needed to construct them. References: - Task decorator parameters (including

cache_policy

,

cache_key_fn

,

persist_result

): Task API reference - Cache policies and helpers: Cache policy API reference If you can share your task signature (which parameter holds

ResourceNotifyDecorator

), I can show the exact

INPUTS - "param_name"

or

cache_key_fn

you’d need.