Trying to plan out a deployment for an application and am wondering if it makes sense to have multiple pods in a container vs putting them in separate pods. I expect one of the containers to potentially be operating near its allocated memory limit. My understanding is that this presents the risk of this container getting OOMKilled. If that's the case, would it restart the entire pod (so the other container in the pod is restarted as well) or will it only restart the OOMKilled container?
No, only the specific container.
For the whole pod to be recreated there needs to be a change in the Pod's ownerObject (tipically a Replicaset) or a scheduling decision by kube-scheduler.