CoreOS — an operating system for containers

A typical Linux server has hundreds of packages, most of which you do not need. CoreOS takes a radically different approach: a minimal OS designed exclusively for running containers. No package manager; everything lives in containers.

The CoreOS philosophy¶

A read-only OS with a dual-partition system for atomic updates with rollback. etcd for distributed configuration, fleet for primitive orchestration. Configuration via cloud-config — declarative, no manual setup.

#cloud-config
coreos:
  etcd2:
    discovery: https://discovery.etcd.io/abc123
    advertise-client-urls: http://$private_ipv4:2379
  fleet:
    public-ip: $private_ipv4
  units:
    - name: etcd2.service
      command: start
    - name: fleet.service
      command: start

etcd: distributed configuration¶

A Raft-based key-value store for sharing configuration between services. Watchers react to changes in real time — automatic reconfiguration without restarts.

Our experience¶

Debugging without traditional tools is uncomfortable. The ecosystem is young. Fine for early adopters; proceed with caution for enterprise production. We are watching Kubernetes as a more robust orchestration solution.

The future of container-optimised OS¶

CoreOS points the way: minimal, immutable, container-first. Concepts such as atomic updates will be standard in a few years.