Hello. Have another question while checking on ifelse() flow. What is my best approach if I have a lot of downstream tasks based on a ifelse decision. Do I wrap all the downstream tasks with a ifelse()?

tasks which are downstream of the skipped task will also skip, so you shouldn’t have to worry about them!

piggy-backing on this: is there a way to put multiple tasks behind an

ifelse

at once? the following runs both

add

tasks and only the

List

is skipped

@task
def add(x, y):
    print(x, y)
    return x + y

with Flow('flow') as flow:
    do_stuff = Parameter('do_stuff')
    ifelse(do_stuff, [add(1, 2), add(2,3)], None)

curious if this was what @Tsang Yong was dealing with as well (maybe w/ a shared upstream dependency for all of the tasks that gets skipped by the ifelse?)

Hi Brett, we used to have a warning for this situation, but it was confusing because it also showed in other situations, so we removed it in https://github.com/PrefectHQ/prefect/pull/1514. Essentially, though it’s surprising, the

add

tasks are actually upstream of the

list

task, and therefore they aren’t gated by the

ifelse

.

makes sense; this seemed to work for me

from typing import List
from prefect import Task

def do_if(condition: Task, *tasks) -> None:
    flag = Task()
    ifelse(condition, flag, None)
    for task in tasks:
        task.set_upstream(flag)

@task
def add(x, y):
    print(x, y)
    return x + y

with Flow('flow') as flow:
    do_stuff = Parameter('do_stuff')
    total_1 = add(1, 2)
    total_2 = add(2, 3)
    do_if(do_stuff, total_1, total_2)

state = flow.run(parameters={'do_stuff': False})

👍 Glad this works! However, interestingly, this could give you unexpected behavior in the future, depending on the complexity of your real-world tasks that are upstream of the

add

. Prefect executes tasks in a non-deterministic topological order. That means that your constant tasks (

1, 2, 2, 3

) MIGHT get executed before even

do_stuff

, simply because the system identifies that they have no upstream blockers. Obviously in this example it doesn’t matter but just wanted to call it out as an implementation detail for your real-world cases!

👍 should be fine here but good to keep in mind, thanks!