[djsumdog]

I don't get why k8s is the dominant scheduler. If you have a 3 ~ 6 person platform team that can set one up, or build a secure terraform of CFN codebase to establish an AWS/EKS system, they can be nice. But I've also worked at DCOS/marathon shops where it worked just as well.

The trouble with all these schedulers is they can't go from just one node (where scheduling and processes run on the same node .. and minikube is a hack; not a production system) to 100. You can't just setup a small k8s, and then add a node, and another node, and scale up. You go from a single docker system, to a big managed k8s system.

There needs to be more competition. It's the same deal with the dominance of systemd as the only system layer. Only the small startups seem to be using more lightweight stuff like Nomad, k3s, RancherOS (Rancher is mostly going the managed k8s solution anyway; even though they have their own k3s implementation).

A running k8s system can be okay, but there is a lot of room for improvement (in terms of making it simpler). Both DCOS and k8s seem to waste a lot of resources. Docker could have competed in this space, but everyone complained about all the bugs in Swarm and it never really went anywhere.

I did a writeup on container orchestration systems late last year:

https://penguindreams.org/blog/my-love-hate-relationship-wit...