Every cluster needs to collect metrics. They are the basis of any alerting system we might want to employ. Without the information about the current and the past state of a cluster, we would not be able to react to problems when they occur nor would we be able to prevent them from happening in the first place. Actually, that is not entirely accurate. We could do all those things, but not in a way that is efficient and scalable.
A good analogy is blindness. Being blind does not mean that we cannot feel our way through an environment. Similarly, we are not helpless without a way to collect and query metrics. We can SSH into each of the nodes and check the system manually. We can start by fiddling with top, mem, df, and other commands. We can check the status of the containers with the docker stats command. We can go from one container...