https://prefect.io
Join Slack
Hi! I think I found a bug (it happens since 2.6.1 version), when a flow has consecutive `.map` calls...
# ask-community
m
Hi! I think I found a bug (it happens since 2.6.1 version), when a flow has consecutive 
.map
calls, if one task fails in the first map, then all dependent tasks stay in NotReady state. Example flow in the comments
✅ 1
Copy code
@task
def get_list():
    return [1, 2, 3]


@task
def task1(i):
    if i == 1:
        raise ValueError
    return i


@task
def task2(i):
    return i


@task
def final_task(i_list):
    logger = get_run_logger()
    <http://logger.info|logger.info>(i_list)


@flow
def failing_flow():
    result = get_list.submit()
    result = task1.map(result)
    result = task2.map(result)
    result = task1.map(result)
    final_task.submit(result)
🙏 1
And this is radar view and tasks states
a
Afaik, this is intentional to ensure that we don't proceed with the next tasks if the first one fails. Starting from 2.6 you can modify that behavior by wrapping the future using 
allow_failure
annotation
great MRE btw! You might be right, this works with a for loop, but not sure if this works with mapping:
Copy code
from prefect import task, flow, allow_failure, get_run_logger


@task
def get_list():
    return [1, 2, 3]


@task
def task1(i):
    if i == 1:
        raise ValueError
    return i


@task
def task2(i):
    return i


@task
def final_task(i_list):
    logger = get_run_logger()
    <http://logger.info|logger.info>(i_list)


@flow
def failing_flow():
    numbers = get_list.submit()
    for i in numbers.result():
        flaky_result = task1.submit(i)
        flaky_result = task2.submit(allow_failure(flaky_result))
        final_task.submit(allow_failure(flaky_result))


if __name__ == "__main__":
    failing_flow()
thanks so much for reporting this, we can check with Michael and follow up on the issue
m
Oh great, thanks for opening the issue, I'll be waiting for the resolution! Thank you Anna 😬
🙌 1
7 Views
Open in Slack
PreviousNext
Powered by

Prefect Community

Bring your towel and join one of the fastest growing data communities. Welcome to our second-generation open source orchestration platform, a completely rethought approach to dataflow automation.

Powered by