You’re very close! A few notes for Prefect 3.x: -

run_deployment

is sync-compatible and will wait for completion by default. Passing

timeout=0

is the right way to “fire-and-return” immediately with a

FlowRun

so you can wait concurrently yourself. -

wait_for_flow_run

is async and returns the final

FlowRun

(with its terminal state). - In your snippet, you forgot to await the second

gather

— you’re checking a Future instead of the finished runs. - Prefer checking

state.is_completed()

instead of comparing to

StateType.COMPLETED

. Here’s a clean pattern you can drop in:

import asyncio
from prefect import flow
from prefect.deployments import run_deployment
from prefect.flow_runs import wait_for_flow_run

@flow
async def launch_and_wait(x: int):
    # Trigger child runs; timeout=0 returns immediately (no waiting)
    children = await asyncio.gather(
        run_deployment(name="flow-a/dep-a", parameters={"p": x}, timeout=0),
        run_deployment(name="flow-b/dep-b", parameters={"p": x}, timeout=0),
        run_deployment(name="flow-c/dep-c", parameters={"p": x}, timeout=0),
    )

    # Wait for all children to finish concurrently
    finished = await asyncio.gather(
        *(wait_for_flow_run(r.id, log_states=True) for r in children)
    )

    # Fail the parent if any child did not complete successfully
    if not all(fr.state and fr.state.is_completed() for fr in finished):
        raise ValueError("Not all sourcing flows successfully completed")

Notes and options: - If you want the parent to wait inline for a single deployment, just call

await run_deployment(..., timeout=<seconds>)

without using

wait_for_flow_run

. - If you want a global timeout for all children, wrap the second

gather

with

asyncio.wait_for(...)

. - By default,

run_deployment

links runs as subflows (

as_subflow=True

) so you’ll see the hierarchy in the UI. Docs for reference: - Run deployments (how-to) - run_deployment API - wait_for_flow_run API - States overview Want the parent to fail fast and cancel the other children if any one fails? If so, I can share a variant that cancels outstanding runs as soon as the first failure is observed.