Dedicating OCP nodes for specific workloads

You can use the kubectl taint command to dedicate OCP cluster nodes for specific workloads. You can dedicate GPU nodes for CML workloads, and NVME nodes for CDW workloads.

Run the following command to get a list of all of the cluster nodes:

kubectl get nodes

Run the following command to list information about a specific cluster node:

kubectl describe node <node_name>

In the returned output, look for the Taints field.

Dedicate a GPU node for CML workloads

Run the following command to dedicate a GPU node for CML workloads:
```
kubectl taint nodes <node_name> nvidia.com/gpu=true:NoSchedule
```
No other workload pods will be allowed to run on the tainted node.
Run the following command to confirm that the taint has been successfully applied:
```
kubectl describe node <node_name>
```
In the returned output, look for the Taints field.

To remove the taint, run the following command:
```
kubectl taint nodes <node_name> nvidia.com/gpu=true:NoSchedule-
```
This command returns:
```
node/<node_name> untainted
```

Dedicate a SSD node for CDW workloads

Run the following command to dedicate a GPU node for CDW workloads:
```
kubectl taint nodes <node_name> ssd/nvme=true:NoSchedule
```
No other workload pods will be allowed to run on the tainted node.
Run the following command to confirm that the taint has been successfully applied:
```
kubectl describe node <node_name>
```
In the returned output, look for the Taints field.

To remove the taint, run the following command:
```
kubectl taint nodes <node_name> ssd/nvme=true:NoSchedule-
```
This command returns:
```
node/<node_name> untainted
```

Additional Notes

note

To taint the node of an existing cluster which already has CML and CDW workspaces running, you must also run the following commands:

kubectl drain <node_name> --ignore-daemonsets --delete-emptydir-data --timeout=600s
kubectl uncordon <node_name>