V1RayCluster

polyaxon._flow.run.ray.ray.V1RayCluster()

Ray clusters are used to run Ray applications on Kubernetes.

Ray Ray is an open-source unified compute framework that makes it easy to scale AI and Python workloads, from reinforcement learning to deep learning to tuning, and model serving.

• Args:
  • kind: str, should be equal raycluster
  • entrypoint: str, optional
  • runtime_env: Dict, optional
  • metadata: int, Dict, optional
  • ray_version: str, optional
  • head: V1RayReplica, optional
  • workers: Dict[str, V1RayReplica], optional
  • enable_in_tree_autoscaling: bool, optional
  • autoscaler_options: V1RayAutoscalerOptions, optional

YAML usage

run:
  kind: raycluster
  entrypoint:
  runtimeEnv:
  metadata:
  rayVersion:
  head:
  workers:
  enableInTreeAutoscaling:
  autoscalerOptions:

Python usage

from polyaxon.schemas import V1Environment, V1Init, V1RayCluster, V1RayReplica
ray_cluster = V1RayCluster(
    connections=["connection-name1"],
    volumes=[k8s.V1Volume(...)],
    ray_version="2.5.0",
    head=V1RayReplica(...),
    worker=V1RayReplica(...),
    enable_in_tree_autoscaling=True,
)

Fields

kind

The kind signals to the CLI, client, and other tools that this component’s runtime is a cluster.

If you are using the python client to create the runtime, this field is not required and is set by default.

run:
  kind: raycluster

entrypoint

The entrypoint command for this cluster.

run:
  kind: raycluster
  entrypoint: python train.py

runtimeEnv

The runtime environment for this cluster.

run:
  kind: raycluster
  runtimeEnv:
    pip: ["requests==2.26.0", "pendulum==2.1.2"]
    env_vars: {"counter_name": "test_counter"}

rayVersion

The version of Ray the application uses.

run:
  kind: raycluster
  rayVersion: 2.5.0
  ...

head

Ray head replica specification

run:
  kind: raycluster
  head:
    rayStartParams:
      dashboard-host: '0.0.0.0'
    container:
    ...
  ...

workers

List of worker replica specifications

run:
  kind: raycluster
  workers:
    small-group:
      replicas: 1
      minReplicas: 1
      maxReplicas: 5
    ...
  ...

enableInTreeAutoscaling

Enable the KubeRay in-tree autoscaler for automatic worker scaling. When enabled, the Ray autoscaler will automatically scale workers based on workload demands.

Note: The idleTimeoutSeconds for the autoscaler is derived from termination.culling.timeout.

run:
  kind: raycluster
  enableInTreeAutoscaling: true
  workers:
    gpu-workers:
      minReplicas: 0
      maxReplicas: 10
  ...

autoscalerOptions

Optional configuration for the Ray autoscaler behavior.

run:
  kind: raycluster
  enableInTreeAutoscaling: true
  autoscalerOptions:
    upscalingMode: Default
    imagePullPolicy: IfNotPresent
  ...