Not really. Kubernetes is stupidly unstable. When I was testing it a while back, the master processes kept dying for no good reason and leaving half-on processes with no networking.

And then you get to restart everything. Yay.

Whereas if Kubernetes is on something else (OpenStack/Mesos) you at least have control to restart Kubernetes when it shits its brick.

The "a while back" qualifier here is important. Both Mesos and Kubernetes are relatively new projects in a rapidly moving space. FWIW we are running Kubernetes 1.4x with multiple redundant masters in production with no problems. We are also running Mesos/Marathon internally. These projects did have major issues early on but now they are becoming much more robust.

For control we are using CoreOS with kube-aws. It actually works quite nicely as we can scale and upgrade running clusters. The only issue with it right now is upgrading and managing etcd.

I think the whole issue with OpenStack is the overambitious goal to be a one stop from bare metal servers and switches to applications. Kubernetes with Flannel or Calico can take care of the software level networking and provisioning, while there are a lot of other ways to provision VMs for running Mesos or Kubernetes depending on where you are deploying.

Kubernetes and the push for it is religious. Get out of here with that. Not all of us have been industry < 10 years

I've been in "the industry" professionally since ~2005, and with Linux since 1998 or so. I have a business need for kubernetes and there is nothing at all religious about it. It is simply that we have lots of new applications that need to be spun up and deployed. They might need to scale up or scale down, and we have gobs and gobs of on-premise (read 0 cloud) new hardware. "API Driven" infrastructure is the best way to achieve a better story for the various ops teams, and so we're building out some kubernetes clusters globally for them.

In other news: http://i0.kym-cdn.com/photos/images/facebook/001/044/247/297...

A better 'story'? This is terrible jargon as is the entire storied approach. There are dozens of ways to achieve the technical outcome you aim at and docker is one. Docker is the less toil|overhead, less flexible|performant, vendor lock-in, learn a new toolset approach. Save your indignation for someone who doesn't know better.

Because running docker (which is the least stable part of my entire infra) on thousands of machines is a good idea when there isn't a sensible way to do orchestration? Totally some startup knows better than google about container orchestration. The same google that literally added control groups to the Linux kernel which sort of invented linux containers.

Spare the drama dude. Just because something doesn't work for you doesn't mean it doesn't work very well for others.