Which part of Kubernetes manages the automatic restart of failed containers inside a pod?

In Kubernetes, the automatic restart of failed containers within a Pod is managed by kubelet — an agent running on each cluster node.

Kubelet monitors the state of containers in a Pod, and if a container crashes, it attempts to restart it according to the restart policy specified in the Pod manifest (e.g., Always, OnFailure, Never).

Thus, it is kubelet that is responsible for monitoring and recovering containers at the node level.