Within Kubernetes, pods are used to run the applications. Within the pods, it is specified which resources are required to run the application. The scheduler (kube-schedular) within Kubernetes checks where to run the application depending on the demands and the nodes coupled to the cluster.
Pods themselves have a limited lifespan and are removed when new versions are deployed or, for example, when a node fails, pods can be replaced by pods on the same or another node.