Hello, Have anyone experienced problems when using:

prefect.engine.signals.LOOP

, where the task is suddenly is missing the context on retries. e.g:

index = prefect.context.get('task_loop_result', {}).get('index', 0)

See more code in the thread 🧵

Can you try to sort of mock the behavior in a simpler hello-world flow (without the request to an external API) and see if it works locally? The agent doesn't matter in this context - the end outcome should be the same since this is purely Prefect Core code. This code from the docs is quite similar:

import requests
from datetime import timedelta

import prefect
from prefect import task, Flow, Parameter
from prefect.engine.signals import LOOP


@task(max_retries=5, retry_delay=timedelta(seconds=2))
def compute_large_fibonacci(M):
    # we extract the accumulated task loop result from context
    loop_payload = prefect.context.get("task_loop_result", {})

    n = loop_payload.get("n", 1)
    fib = loop_payload.get("fib", 1)

    next_fib = <http://requests.post|requests.post>(
        "<https://nemo.api.stdlib.com/fibonacci@0.0.1/>", data={"nth": n}
    ).json()

    if next_fib > M:
        return fib  # return statements end the loop

    raise LOOP(message=f"Fib {n}={next_fib}", result=dict(n=n + 1, fib=next_fib))


with Flow("fibonacci") as flow:
    M = Parameter("M")
    fib_num = compute_large_fibonacci(M)

Thanks for replying! :catjam: I can't reproduce it locally that is actually the problem. It's also weird because sometimes it works and sometimes it doesn't. Can it be something regarding task.result (PrefectResult, etc), where it for some reason isn't able to find the

prefect.context.get()

inside the task?

can you share your code that uses that context and result type? you haven't shared that piece of code yet and it indeed might be helpful

sometimes it works and sometimes it doesn't

as stupid as it may sound, this type of problem can often be solved by: