The correct answer is B: kubelet. The Container Runtime Interface (CRI) defines how Kubernetes interacts with container runtimes in a consistent, pluggable way. The component that speaks CRI is the kubelet, the node agent responsible for running Pods on each node. When the kube-scheduler assigns a Pod to a node, the kubelet reads the PodSpec and makes the runtime calls needed to realize that desired state—pull images, create a Pod sandbox, start containers, stop containers, and retrieve status and logs. Those calls are made via CRI to a CRI-compliant runtime such as containerd or CRI-O.
Why not the others:
kubeadm bootstraps clusters (init/join/upgrade workflows) but does not run containers or speak CRI for workload execution.
kube-apiserver is the control plane API frontend; it stores and serves cluster state and does not directly integrate with runtimes.
kubectl is just a client tool that sends API requests; it is not involved in runtime integration on nodes.
This distinction matters operationally. If the runtime is misconfigured or CRI endpoints are unreachable, kubelet will report errors and Pods can get stuck in ContainerCreating, image pull failures, or runtime errors. Debugging often involves checking kubelet logs and runtime service health, because kubelet is the integration point bridging Kubernetes scheduling/state with actual container execution.
So, the node-level component responsible for CRI integration is the kubelet—option B.
=========
