Add Nodepool to IBM Cloud Openshift Cluster Procedure

The following procedure adds a new nodepool named to pipelines-l4 an existing cluster using the ibmcloud command line interface (CLI) using the gx3.16x80.l4 VPC flavor, which has the following specifications:

• 16 cores
• 80GB memory
• 32Gbps network speed
• 1 L4 GPU

For additional flavors, see VPC flavors. Modify as needed for the organizations requirements.

Install Software Requirements

The following software or runtimes are required for Wallaroo 2025.1. Most are automatically available through the supported cloud providers.

Create the Openshift IBM Cloud CUDA Nodepool

The following steps creates the nodepool with CUDA hardware using the ibmcloud CLI.

1. Retrieve the following information:

   • VPC_ID: The ID of the IBM Cloud® Virtual Private Cloud. Retrieved with ibmcloud oc vpcs. For example:
     • VPC_ID="r006-cac4bfbe-d04d-481a-a099-ba243ea64afd"
   • SUBNET_ID_1: The subnet ID used. This is retrieved with the following command:
     • ibmcloud oc subnets --provider vpc-gen2 --vpc-id <your-vpc-id> --zone <your-zone-name>

2. Set the environmental variables. Modify as needed to match the target cluster. The variable WORKER_POOL_NAME is used to determine the new nodepool. This example uses pipelines-l4.

   # Set the environmental variables
   CLUSTER_NAME="samplecluster"
   ZONE_1="us-south-1"
   SUBNET_ID_1="0717-6f46918e-2107-48ae-b023-eb053601697b"
   L4_GPU_FLAVOR="gx3.16x80.l4"
   # the name of the new node.
   WORKER_POOL_NAME="pipelines-l4"
   # the name of the acceleration label used
   ACCELERATION_LABEL="l4"
   

   The following command is used to retrieve zone details, including the subnet id.

   ibmcloud oc subnets --provider vpc-gen2 --vpc-id <your-vpc-id> --zone <your-zone-name>
   

3. Create the nodepool and add it to the target cluster with the standard and the custom label.

   # Create the nodepool and add it to the cluster.
   ibmcloud oc worker-pool create vpc-gen2 --cluster "$CLUSTER_NAME" --name "$WORKER_POOL_NAME" --flavor "$L4_GPU_FLAVOR" --size-per-zone 1 \
       --label wallaroo.ai/node-purpose=pipelines \
       --label wallaroo.ai/accelerator=$ACCELERATION_LABEL
   

4. Add the nodepool to the zone and subnet.

   # add to the zones and subnet
   ibmcloud oc zone add vpc-gen2 --zone "$ZONE_1" --subnet-id "$SUBNET_ID_1" --cluster "$CLUSTER_NAME" --worker-pool "$WORKER_POOL_NAME"
   

5. Wait for the workers for the nodepool to be ready.

   # wait for the nodepool to be ready
   echo "Waiting for nodes in '$WORKER_POOL_NAME' pool to be ready before tainting..."
   kubectl wait --for=condition=Ready node -l ibm-cloud.kubernetes.io/worker-pool-name=$WORKER_POOL_NAME --timeout=15m
   

6. Add the standard taints. If non-standard taints are used, modify as needed.

   # apply the taints
   echo "Tainting '$WORKER_POOL_NAME' pool nodes..."
   kubectl taint nodes -l ibm-cloud.kubernetes.io/worker-pool-name=$WORKER_POOL_NAME wallaroo.ai/pipelines=true:NoSchedule --overwrite
   kubectl taint nodes -l ibm-cloud.kubernetes.io/worker-pool-name=$WORKER_POOL_NAME nvidia.com/gpu=$ACCELERATION_LABEL:NoSchedule --overwrite
   

   IMPORTANT NOTE: Verify that the label is communicated to developers for model deployment. Labels are required for deploying models in Wallaroo with GPUs enabled. For more details, see Deployment Configuration with the Wallaroo SDK: GPU Support