Does kubernetes actually work as a Mesos framework? I thought that was the point of marathon.
Have you ever used Mesos? It is the only way k8s could run ontop of Mesos. Also:
https://github.com/kubernetes-incubator/kube-mesos-framework
Also, you could maybe compare Marathon to Kubernetes Dashboard + the Kubernetes apiserver maybe, but comparing the two directly is not remotely close.
With Mesos + Marathon, you still need a load balancer for both North <---> South AND East <---> West cluster traffice. This is builtin to Kubernetes via Service and Ingress objects and their associated controllers. There is also Federation, and so much more.
Mesos is just a resource scheduler, and a very good two level one at that, built to prove the theory of two level schedulers working well with interactive and batch workloads concurrently. Kubernetes is the entire package and includes a default scheduler, but allows you to plug in your own.
Disclaimer: I spent a full year working on Mesos and went to MesosCon twice. After speaking in person with Kelsey Hightower at Monitorama, he convinced me to switch to Kubernetes on purely technical reasons. The stuff that you have to DIY on Mesos is built sensibly and pluggably in Kubernetes. From a cluster operator standpoint frankly, Kubernetes is leagues ahead.
Lots of fortune 500s and enterprise customers find the lack if sensible RBAC and real to the core auditing support in Mesos to be seriously lacking.
At this point kubernetes' won in the container area, but that doesn't mean mesos is no longer useful. Running containers was just one of things it could do, its main purpose is to manage hardware.
As for your original question, I don't see any other way mesos would run kubernetes than as a framework. That is the whole point, otherwise there would be no point of using mesos.
BTW there already was kubernetes framework, but it seemd to be abandoned since k8s 1.3 or 1.2. I'm glad it was resurrected.
This seems like an overstatement. From where I’m standing, it seems like there’s plenty of room for many solutions to thrive in this space.
I use and like Kubernetes in local development at my org, and we’re probably going to be testing production workloads on Kubernetes soon. The learning curve is quite steep and almost entirely front-loaded. There will be a significant portion of companies for which a different solution is more desirable, based on that fact alone.