☁️ Free practice test question from: KCNA (KCNA)

Option B combines two mechanisms to ensure the workload runs exclusively on GPU nodes:

1. The toleration allows the pod to schedule on nodes with the hardware=gpu:NoSchedule taint. This is necessary because without this toleration, the pod would be rejected from these nodes.

2. The nodeAffinity requiring nodes labeled hardware=gpu ensures the pod will only be scheduled on nodes with this label, not on any untainted nodes.

#####

A) (nodeSelector alone) would try to schedule pods only on nodes with the hardware=gpu label, but those nodes have taints that would reject the pods without proper tolerations.

C) (toleration alone) would allow the pods to run on GPU-tainted nodes, but wouldn't prevent them from running on other untainted nodes. The pods could still be scheduled anywhere else in the cluster.

D) (pod affinity) is trying to use the wrong mechanism entirely. Pod affinity targets other pods, not node characteristics. It would attempt to co-locate with existing pods on GPU nodes, but wouldn't guarantee placement on GPU nodes specifically.