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

Flyte

Hi, we need some clarification on Flyte Agents. We recently wrote a custom Flyte Agent to create a RayJob on our GKE cluster.

Following the examples, we added a custom task (inheriting from `PythonFunctionTask`) that maps to the custom agent and passes the parameters to the agent via `get_custom` method.

We haven't yet deployed the Agent Service, but we're confused that won't the Flytepropeller create a container for the custom task inheriting from `PythonFunctionTask` ? Do we even need the custom task?

I think you should be able to reuse existing ray task, if you don’t want to use the ray backend plugin. <https://github.com/flyteorg/flytekit/blob/3b7cb3c1de75f626d6fdceb64b54a95ea9f98e6d/plugins/flytekit-ray/flytekitplugins/ray/task.py#L54>

but you need to update the propeller config map to something like
```default-for-task-types:
  ray: agent-service```
<https://github.com/flyteorg/flyte/blob/5d792c09bef73c12df46c48fe197418254093f0c/charts/flyte/values.yaml#L534>

any reason why you need to create a custom agent to run a ray job? just curious.