Docker Swarm Classic (https://github.com/docker-archive/classicswarm) is dead. Docker Swarm Mode is alive, and I know some people use it, but it's very niche compared to k8s.

As someone who interacts with k8s regularly, I often feel like there is a place for a simpler k8s alternative. But looking at history I see the attempts like Swarm fail.

What do you think played the decisive role in the k8s victory? Features, openness, extensibility, something else? And do you think if there is a demand for a simper k8s alternative or is it destined to repeat k8s complexity to be useful? If there is such a demand, where did Swarm fail in particular?

I don't think Swarm Mode failed. We are using it in production for 1000s of Hosts very reliably, it's a great piece of technology! :-)

K8s simply has all the hype and no one got fired for running k8s. I was thinking the same as you, k8s is quite complex, has a lot of options and moving parts which we don't need.

It’s just one anecdote but I did run a Docker Swarm Classic cluster and a Kubernetes cluster for years on about 10 bare metal PCs in my worklab.

I initially preferred Docker Swarm for its simplicity, and most of the nodes were in the Swarm cluster.

But I quickly noticed ask issue: Docker Swarm had networking issues. I remember that I wasn’t alone experiencing network issues and disliking debugging with iptables.

Meanwhile, Kubernetes with the flannel CNI was super reliable. I actually ditched Docker Swarm for Rancher 1. It wasn’t based on Kubernetes at the time. It had some JVM based networking thing that worked but used a lot of CPU. We also had some Mesos testing.

Maybe the new Docker Swarm works, but it’s too late. Kubernetes won.

Kubernetes is more complex but tools like kompose makes the transition easier.