Flyte enables production-grade orchestration for machine learning workflows and data processing created to accelerate local workflows to production.

Flyte

Is there a way to make a cleanup task execute on termination from within the UI?

Screenshot 2024-10-04 at 11.02.39 AM.png

Perhaps failure node: <https://docs.flyte.org/en/latest/user_guide/development_lifecycle/failure_node.html>

the failure node is that `fn0` , it's getting skipped when i terminate from the UI

image.png

I made a node fail by raising an exception in it, and it runs the cleanup as expected

Ahh, so it doesn’t execute the cleanup task when killed via the UI based `Abort` :thinking_face:

yep, but we have things we need to teardown on `Abort` too!

<@U07655DJTDM> is this similar to what's described by <https://github.com/flyteorg/flyte/issues/2810|this issue>?

I think this one is closer: <https://github.com/flyteorg/flyte/issues/3854>

I think the first one covers our use case better

We need to cleanup some compute resources on termination. I think agents might also work but tbh the agents interface has a more complicated deploy/infra considerations so we've shied away from it

&gt; the agents interface has a more complicated deploy/infra considerations so we've shied away from it
I hear you, and your feedback is super appreciated.
If there's any way we can help you, it'd be amazing

honestly I think if you could just deploy an agent ephemerally as if it were a task (which is basically the proposal in the issue you linked) I'm sure we would use that feature

cc <@U077P82HG8H> <@U07JCU5EBTN> <@U07JLUBCNFN> This would make it a lot easier to control some of our spark costs