Scheduling Docker containers to Cluster: Swarm or CoreOS?

1
votes

I would like to schedule Docker containers to a small cluster. The jobs that will be running will mostly be one-time-jobs like Rapidminer computations, not any lasting services.

I have tried to read a lot about the available systems and I believe that Kubernetes is not really needed, because of the absence of lasting jobs like web servers, which need to keep running and restarted asap if they fail.

However, I am not sure if my initial idea of using Swarm on simple Ubuntu LTS hosts shouldn't be replaced with using CoreOS. Especially because CoreOS seems to have its own scheduling solution, although it does look more complicated.

Does CoreOS provide benefits that would make it worth using over Ubuntu+Swarm for the given use case?

2
dockercoreosdocker-swarm
if you're running 1 time jobs, SWARM is probably best. Mesos, Kubernetes and such are meant to orchestrate containers on a cluster and make sure they run at all time even when a node fails. Scripting Swarm deploy will probably be best suited for your type of application - MrE

2 Answers

2
votes

If you are looking for better scheduling solutions, you should go with CoreOS (If they are only two choices, otherwise go with Mesos!). If you have your own scheduling algorithm and ready to integrate it with swarm from scratch, then it could be a good experience to choose swarm, as the scheduler there is very basic.

1
votes

We use Mesos. It's very easy to set up, e.g. there are rpms for mesos masters and slaves. There is native support for docker containerization and a clean rest interface.