We have a stateless architecture so the task is generally just to push X containers onto Y machines. We use Ansible to distribute the containers onto he right

We have a stateless architecture so the task is generally just to push X containers onto Y machines. We use Ansible to distribute the containers onto he right servers. Because it's HTTP/REST, most of our service discovery comes from pointing at a known set of EC2 load balancers, but we talking about etcd or Zookeeper for config management and service discovery. There is more to do here and it's the missing piece of the puzzle, but you can get quite far without a Fleet/SERF type abstraction layer.