Hi, do we still have SequentialTaskRunner in any form in Prefect 3.0.0? I couldn't find it, and not specifying task runner shows concurrent behaviour.

Starting with 3.0, Prefect flows run sequentially by default. There's no need to specify SequentialTaskRunner explicitly anymore.

@Mihir Thatte I tried running this piece of code:

import time

from prefect import task, flow
from prefect.states import Completed


@task
def task_a(x):
    time.sleep(3)
    print("A"*x)
    return Completed()


@flow
def generic_flow():
    lst = []
    for i in range(1, 6):
        lst.append(task_a.submit(i))
    for f in lst:
        f.wait()


if __name__ == "__main__":
    generic_flow()

and it produced such log:

17:58:25.150 | INFO    | prefect.engine - Created flow run 'warping-hoatzin' for flow 'generic-flow'
17:58:25.158 | INFO    | prefect.engine - View at <http://127.0.0.1:4200/runs/flow-run/ea1484fa-4d54-4d4e-b74c-843ef61fa75d>
17:58:25.380 | INFO    | Flow run 'warping-hoatzin' - Submitting task task_a to thread pool executor...
17:58:25.398 | INFO    | Flow run 'warping-hoatzin' - Submitting task task_a to thread pool executor...
17:58:25.423 | INFO    | Flow run 'warping-hoatzin' - Submitting task task_a to thread pool executor...
17:58:25.442 | INFO    | Flow run 'warping-hoatzin' - Submitting task task_a to thread pool executor...
17:58:25.455 | INFO    | Flow run 'warping-hoatzin' - Submitting task task_a to thread pool executor...
17:58:26.001 | INFO    | Task run 'task_a-2' - Created task run 'task_a-2' for task 'task_a'
17:58:26.104 | INFO    | Task run 'task_a-3' - Created task run 'task_a-3' for task 'task_a'
17:58:26.273 | INFO    | Task run 'task_a-0' - Created task run 'task_a-0' for task 'task_a'
17:58:26.393 | INFO    | Task run 'task_a-4' - Created task run 'task_a-4' for task 'task_a'
17:58:26.472 | INFO    | Task run 'task_a-1' - Created task run 'task_a-1' for task 'task_a'
AAA
A
17:58:30.052 | INFO    | Task run 'task_a-2' - Finished in state Completed()
AA
17:58:30.118 | INFO    | Task run 'task_a-0' - Finished in state Completed()
AAAAA
17:58:30.216 | INFO    | Task run 'task_a-1' - Finished in state Completed()
17:58:30.301 | INFO    | Task run 'task_a-4' - Finished in state Completed()
AAAA
17:58:30.444 | INFO    | Task run 'task_a-3' - Finished in state Completed()
17:58:30.609 | INFO    | Flow run 'warping-hoatzin' - Finished in state Completed()

It seems that 5 task submits with time.sleep(3) each (so 15 seconds at least) are done in 5 seconds. That shows code is executed concurrently. Also visualizer shows this:

hi @Ihor Ramskyi - to add some color here by default if you call a sequence of tasks, they're going to execute sequentially like normal python code

from prefect import flow, task

@task
def foo(): ...

@flow
def some_flow():
  foo()
  foo()
  foo()

some_flow()

by using

submit

, you're opting into the

ThreadPoolTaskRunner

(called

ConcurrentTaskRunner

in 2.x) where you submit task runs to the task runner in a non-blocking fashion, which the task runner will execute concurrently, subject to any state dependencies (set with

wait_for

) so this is not actually different from 2.x at all, and tldr: • to get simple sequential behavior, just call your tasks, no

.submit()

• to get concurrent behavior, using

.submit()

or

.map()

to send work to task runners

@Nate yeah, that clears things for me, thank you for explanations!