Skip to content

Install a Cluster

Install NVIDIA Dependencies


You must have Cluster Administrator rights to install these dependencies.

Before installing Run:ai, you must install NVIDIA software on your OpenShift cluster to enable GPUs. NVIDIA has provided detailed documentation. Follow the instructions to install the two operators Node Feature Discovery and NVIDIA GPU Operator from the OpenShift web console.

When done, verify that the GPU Operator is installed by running:

oc get pods -n nvidia-gpu-operator

(the GPU Operator namespace may differ in different operator versions).

Create OpenShift Projects

Run:ai cluster installation uses several namespaces (or projects in OpenShift terminology). The installation will automatically create the namespaces, but if your organization requires manual creation of namespaces, you must create them before installing:

oc new-project runai
oc new-project runai-reservation
oc new-project runai-scale-adjust

The last namespace (runai-scale-adjust) is only required if the cluster is a cloud cluster and is configured for auto-scaling.

Monitoring Pre-check

Not required

Run:ai uses the OpenShift monitoring stack. As such, it requires creating or changing the OpenShift monitoring configuration. Check if a configmap already exists:

oc get configmap cluster-monitoring-config -n openshift-monitoring

If it does,

  1. To the cluster values file, add the flag createOpenshiftMonitoringConfig as described under Cluster Installation below.
  2. Post-installation, edit the configmap by running: oc edit configmap cluster-monitoring-config -n openshift-monitoring. Add the following:

apiVersion: v1
kind: ConfigMap
  name: cluster-monitoring-config
  namespace: openshift-monitoring
  config.yaml: |
      scrapeInterval: "10s"
      evaluationInterval: "10s"
        clusterId: <CLUSTER_ID>
        prometheus: ""
        prometheus_replica: ""
For <CLUSTER_ID> use the Cluster UUID field as shown in the Run:ai user interface under the Clusters area.

Cluster Installation

Perform the cluster installation instructions explained here. When creating a new cluster, select the OpenShift target platform.


The cluster wizard shows extra commands which are unique to OpenShift. Remember to run them all.

Optional configuration

Make the following changes to the configuration file you have downloaded:

Key Change Description
createOpenshiftMonitoringConfig false see Monitoring Pre-check above.
runai-operator.config.project-controller.createNamespaces true Set to false if unwilling to provide Run:ai the ability to create namespaces, or would want to create namespaces manually rather than use the Run:ai convention of runai-<PROJECT-NAME>. When set to false, will require an additional manual step when creating new Run:ai Projects.
runai-operator.config.mps-server.enabled Default is false Allow the use of NVIDIA MPS. MPS is useful with Inference workloads. Requires extra permissions
runai-operator.config.runai-container-toolkit.enabled Default is true Controls the usage of Fractions. Requires extra permissions
runai-operator.config.runaiBackend.password Default password already set password. Need to change only if you have changed the password here The address of the default Prometheus Service If you installed your own custom Prometheus Service, change to its' address


Follow the instructions on the Cluster Wizard


To install a specific version, add --version <version> to the install command. You can find available versions by running helm search repo -l runai-cluster.

oc label ns runai
oc -n openshift-ingress-operator patch ingresscontroller/default --patch '{"spec":{"routeAdmission":{"namespaceOwnership":"InterNamespaceAllowed"}}}' --type=merge

helm install runai-cluster -n runai  \ 
  runai-cluster-<version>.tgz -f runai-<cluster-name>.yaml  


Use the --dry-run flag to gain an understanding of what is being installed before the actual installation. For more details see understanding cluster access roles.

Connect Run:ai to GPU Operator

Not required

Locate the name of the GPU operator namespace and run:

kubectl patch RunaiConfig runai -n runai -p '{"spec": {"global": {"nvidiaDcgmExporter": {"namespace": "INSERT_NAMESPACE_HERE"}}}}' --type="merge"

(Optional) Prometheus Adapter for Inference

The Prometheus adapter is required if you are using Inference workloads and require a custom metric for autoscaling. The following additional steps are required for it to work:

  1. Copy prometheus-adapter-prometheus-config and serving-certs-ca-bundle ConfigMaps from openshift-monitoring namespace to the monitoring namespace
    kubectl get cm prometheus-adapter-prometheus-config --namespace=openshift-monitoring -o yaml \
      | sed 's/namespace: openshift-monitoring/namespace: monitoring/' \
      | kubectl create -f -
    kubectl get cm serving-certs-ca-bundle --namespace=openshift-monitoring -o yaml \
      | sed 's/namespace: openshift-monitoring/namespace: monitoring/' \
      | kubectl create -f -
  2. Allow Prometheus Adapter serviceaccount to create a SecurityContext with RunAsUser 10001:
    oc adm policy add-scc-to-user anyuid system:serviceaccount:monitoring:runai-cluster-prometheus-adapter

Next Steps

Continue to create Run:ai Projects.

Last update: 2023-03-26
Created: 2023-03-26