Hi there! I’ve written a task (it happens to be a function task, but it doesn’t have to be - could be a class that inherits from Task) that runs a nomad job. It polls for state and throws any exceptions up the stack. If the job fails, prefect is made aware of it and it works fine. I’m trying to understand how to go the other direction. If I cancel a flow run - how can I tell my Nomad job to stop? Is the thing to use here a

state_handler

? It’s sort of surprising to me that, for example, the kubernetes

RunNamespaceJob

doesn’t seem to include any logic around this (https://github.com/PrefectHQ/prefect/blob/master/src/prefect/tasks/kubernetes/job.py#L580).

This is a good question, and I think we don’t currently have support for it. I think it falls under thsi . It’s on the roadmap though

so I think this is a bit different - in this case K8s or Dask are being used as executors

Ah ok I see what you mean. So if the Flow is cancelled, the current shutdown is aggressive and that’s the spirit of the issue above. If the Flow fails though, you are right that a state handler will be able to help you, but for the

RunNamespaceJob

, I have seen people have difficulty with passing the

pod

details or

job

details out to that they can shut it down in the state handler. I think if it fails, you are right that the state handler is a good approach if you can get the necessary info in there.

Ok cool - I’m going to read a bit more on the

state_handler

bit then. Seems to me that the

Task

metaclass should have a method stub like:

def do_this_on_cancel():
    ...

def do_this_on_fail():
    ...

But I’m no developer…

The state handler supports that with:

def mystatehandler(task,old_state,new_state):
    if new_state.is_failed():
       # logic