When you really need to dial up the scaling to levels of Twitter and Airbnb, you probably need something even more powerful than Swarm or Kubernetes, which is where Mesos and Marathon come into play. Apache Mesos was not actually built with Docker in mind but as a general cluster-management tooling that provides resource management in a consistent way for applications that run on top of it with APIs. You can run anything from scripts, actual applications, and multiple platforms (such as HDFS and Hadoop) with relative ease. For container-based orchestration and scheduling on this platform these days, Marathon is the general go-to here.

Marathon runs as an application (in a very loose sense of the word) on top of Mesos as the container orchestration platform and provides all kind of niceties, such as a great UI (though Kubernetes has one too), metrics, constraints, persistent volumes (experimental at the time of writing this), and many others. As a platform, Mesos and Marathon are probably the most powerful combo for handling clusters in the tens-of-thousands-of-nodes range, but to get everything pieced together, unless you use the pre-packaged DC/OS solution (https://dcos.io/), it is in my experience really, really tricky to get up and running compared to the other two. If you need to cover the range of medium-to-largest of scales with added flexibility in order to run other platforms (such as Chronos) on it too, currently, I would strongly recommend this combo.