KubeRay CLI provides the ability to manage kuberay resources (ray clusters, compute templates etc) through command line interface.
!!! note
The KubeRay CLI is an optional interface backed by the KubeRay API server.
It provides a layer of simplified configuration for KubeRay resources.
The KubeRay CLI is community-managed and is not officially endorsed by the
Ray maintainers. At this time, the only officially supported methods for
managing KubeRay resources are
- Direct management of KubeRay custom resources via kubectl, kustomize, and Kubernetes language clients.
- Helm charts.
KubeRay CLI maintainer contacts (GitHub handles):
@Jeffwan @scarlet25151
Please check release page and download the binaries.
- Kuberay operator needs to be running.
- Kuberay apiserver needs to be running and accessible.
- Kuberay CLI uses Cobra framework for the CLI application.
- Kuberay CLI depends on kuberay apiserver to manage these resources by sending grpc requests to the kuberay apiserver.
You can build kuberay binary following this way.
cd kuberay/cli
go build -o kuberay -a main.go
- Default kuberay apiserver endpoint:
127.0.0.1:8887. - If kuberay apiserver is not run locally, this must be set in order to manage ray clusters and ray compute templates.
./kuberay config get endpoint
./kuberay config reset endpoint
./kuberay config set endpoint <kuberay apiserver endpoint>
Usage:
kuberay cluster create [flags]
Flags:
--environment string environment of the cluster (valid values: DEV, TESTING, STAGING, PRODUCTION) (default "DEV")
--head-compute-template string compute template name for ray head
--head-image string ray head image
--head-service-type string ray head service type (ClusterIP, NodePort, LoadBalancer) (default "ClusterIP", which creates a headless ClusterIP service)
--name string name of the cluster
-n, --namespace string kubernetes namespace where the cluster will be
--user string SSO username of ray cluster creator
--version string version of the ray cluster (default "1.9.0")
--worker-compute-template string compute template name of worker in the first worker group
--worker-group-name string first worker group name
--worker-image string image of worker in the first worker group
--worker-replicas uint32 pod replicas of workers in the first worker group (default 1)
Known Limitation: Currently only one worker compute template is supported during creation.
./kuberay cluster get -n <namespace> <cluster name>
./kuberay cluster -n <namespace> list
./kuberay cluster delete -n <namespace> <cluster name>
Usage:
kuberay template compute create [flags]
Flags:
--cpu uint32 ray pod CPU (default 1)
--gpu uint32 ray head GPU
--gpu-accelerator string GPU Accelerator type
--memory uint32 ray pod memory in GB (default 1)
--name string name of the compute template
-n, --namespace string kubernetes namespace where the compute template will be stored
./kuberay template compute get -n <namespace> <compute template name>
./kuberay template compute list -n <namespace>
./kuberay template compute delete -n <namespace> <compute template name>
Configure the endpoints
kubectl port-forward svc/kuberay-apiserver-service 8887:8887 -n ray-system
./kuberay config set endpoint 127.0.0.1:8887
Create compute templates
./kuberay template compute create -n <namespace> --cpu 2 --memory 4 --name "worker-template"
./kuberay template compute create -n <namespace> --cpu 1 --memory 2 --name "head-template"
List compute templates created
./kuberay template compute list
Create the cluster
./kuberay cluster create -n <namespace> --name test-cluster --user jiaxin.shan \
--head-compute-template head-template \
--head-image rayproject/ray:1.9.2 \
--worker-group-name small-wg \
--worker-compute-template worker-template \
--worker-image rayproject/ray:1.9.2
List the clusters
./kuberay cluster list