Hi everyone. Is there a good way to gracefully cancel a flows next scheduled run? I have flows scheduled every 5 minutes and when I check a change in, I want the flow to finish the current run, but stop after that so I can execute a git pull and re-deploy the dask workers, etc. Is there a good way to do that, other than watching the session and killing the process when the current flow state is "waiting on next scheduled start time"?

Hi @David N ! Are you using Prefect Server or Cloud? Re-registering the flow will always remove scheduled runs but allow any currently in the

Running

state to finish.

We're using server and not cloud.

Great! So this strategy should work: Whenever you make a change you could either: • Turn the schedule off • Pull your infra changes • Turn the schedule back on or • Pull your infra changes and, in concert, push a new version of the flow (by calling

flow.register()

again, no code changes needed)

I really appreciate the info. I'll try to go back and read the docs again, but I think we skip this register() step?

with Flow('flow-name', schedule=schedules.IntervalSchedule(interval=timedelta(seconds=300))) as flow:
    t1=func()
    t2=func2(t1)
flow.run()

then we execute the python scipt, the process runs "forever", and kicks the flow off every 300 secs.

Ohh ok, I see now. You're just running straight from Prefect Core, you don't have the database and UI spun up with Prefect Server, am I understanding correctly?

sorry yes, not prefect server, I lead you wrong.

No worries! Give me a few to check if there's a best-practices way to accomplish that with Core-only

AFAICT there isn't a good way to do this builtin to core. You could hack this together with a custom schedule though (probably a

filter

or

clock

, see https://docs.prefect.io/core/concepts/schedules.html#complex-schedules). You'd want the schedule to respond to some external event (perhaps an os signal, perhaps it starts an http server and waits for a ping, perhaps it watches a file on disk for changes, up to you). When notified, it'd filter any newly scheduled events. Something like (pseudocode):

def myfilter(event):
    if shouldnt_schedule_more_runs():
        return False
    return True

schedule = Schedule(..., filters=[myfilter])

Damn, thats brilliant.