I know the ray plugin typically spins up an an eph...
# ray-integration
p
I know the ray plugin typically spins up an an ephemeral ray cluster to run a ray job (for isolation purposes). Does the flyte-ray plugin support running ray jobs on an already-running static ray cluster? If so, how would one produce that functionality? Thank you!
k
you could use ray.init in the python task to submit the ray task to the existing cluster. like
Copy code
@task
def t1():
    ray.init(<ip_address>)
    ray.remote(...)
p
I notice that you're calling Ray directly inside the task, is there a reason you're not using the actual flytekitplugins-ray plugin (https://pypi.org/project/flytekitplugins-ray/)?
k
The flytekitplugins-ray plugin is used to create ray clusters in flyte.
p
Oh, I thought I could also connect to an already-running cluster via an
address
. In the latest plugin source code (
flytekitplugins/ray/task.py
) I see this `dataclass`:
Copy code
@dataclass
class RayJobConfig:
    worker_node_config: typing.List[WorkerNodeConfig]
    head_node_config: typing.Optional[HeadNodeConfig] = None
    runtime_env: typing.Optional[dict] = None
    address: typing.Optional[str] = None
I would expect this address to work to send a job to an existing Ray Cluster deployed on Kubernetes AWS:
address="<ray://ray-cluster-kuberay-head-svc.ray-cluster-namespace.svc.cluster.local:10001>"
It seems like regardless of weather the
address
field is present or not, the plugin spins up a new ephemeral Ray cluster. I did get it working via your example without the plugin @Kevin Su but I was just curious if there was a way to do this via the plugin. Thanks for your help!
138 Views