While k8s does not in and of itself solve some of the issues pointed out above, it does centralize a lot of these problems. These centralized problems are then often easily addressed by the cloud providers actually running the k8 cluster. GKE automatically sends log data and metrics through fluentd to their Stackdriver platform including error alerting. If something even prints something that looks like a stack trace to stdout or stderr, Stackdriver sends alerts and creates an issue for us.

To the first degree, it is. The way people use containers is typically one binary starting up, which equates to one process in Linux. Sure, docker et al have overhead, but that's constant overhead, and much of the hard work is done in cgroups.

You just might be wrong here, friend.