libswarm is a lightweight, standardized abstraction for federating and orchestrating distributed systems. It can discover and communicate with all machines in your system, using a variety of backend adaptors, and expose them via a single, straightforward API.
This is very useful if your system spans multiple locations on the Internet (your home network, office lan, colocated rack, cloud provider, etc.) and uses heterogeneous tools for deployment, clustering and service discovery (manually configured IPs on your laptop, zeroconf at the office, ssh tunnels + custom DNS on the rack, Mesos, Consul or Etcd/Fleet on your cloud provider, etc.).
libswarm federates all these elements under a unified namespace and lets you orchestrate them using a lightweight and flexible API.
Because the libswarm API is dead-simple to learn and built on standard components, there is a growing ecosystem of tools already compatible with it, which you can use to manage your swarm out of the box:
-
swarmctlis a simple command-line utility which lets you easily compose any combination of backends, and orchestrate them straight from the terminal. -
(coming soon) Shipyard offers a powerful web interface to manage any swarm-compatible distributed system straight from the browser.
-
(coming soon) Swarmlog is a logging tool which consumes all log streams from your swarm and forwards them to a variety of logging backends.
-
(coming soon) Deis, the popular Docker-based application platform, supports libswarm natively, and can deploy applications to it out-of-the-box.
-
(coming soon) Drone.io, the Docker-based continuous integration platform, can spawn build and test workers on any swarm.
-
(coming soon) Fig, the popular development tool, supports libswarm out of the box.
-
(coming soon) Docker, the open-source container engine, has announced that they will adopt libswarm as a built-in, and recommend it as the preferred way to orchestrate and manage a Docker-based cluster.
Libswarm ships with a simple daemon which can control all machines in your distributed system using a variety of backend adaptors, and exposes it on a single, unified endpoint.
Currently swarmd uses the standard Docker API as its frontend, which means any tool which speaks Docker can control swarmd transparently: dokku, flynn, deis, docker-ui, shipyard, fleet, mesos... and of course the Docker client itself.
Note: in the future swarmd will expose the Docker remote API as "just another backend", and expose its own native API as a frontend.
Usage example:
./swarmd tcp://localhost:4242 &
docker -H tcp://localhost:4242 info
You can listen on several frontend addresses, using any address format supported by Docker. For example:
./swarmd tcp://localhost:4242 tcp://0.0.0.0:1234 unix:///var/run/docker.sock
Libswarm supports the following backends:
The debug backend simply catches all commands and prints them on the terminal for inspection.
It returns StatusOK for all commands, and never sends additional data. Therefore, docker
clients which expect data will fail to parse some of the responses.
Usage example:
./swarmd --backend debug tcp://localhost:4242
The simulator backend simulates a docker daemon with fake in-memory data. The state of the simulator persists across commands, so it's useful to analyse the side-effects of commands, or for mocking and testing.
Currently the simulator only implements the containers command. It can be passed
arguments at load-time, and will use these arguments as a list of containers.
For example:
./swarmd --backend 'simulator container1 container2 container3' tcp://localhost:4242 &
docker -H tcp://localhost:4242 ps
In this example the docker client should report 3 containers: container1, container2 and container3.
The forward backend connects to a remote Docker API endpoint. It then forwards all commands it receives to that remote endpoint, and forwards all responses back to the frontend.
For example:
./swarmd --backend 'simulator myapp' unix://a.sock &
./swarmd --backend 'forward unix://a.sock' unix://b.sock
docker -H unix://b.sock ps
This last command should report 1 container: myapp.
Create a simple my-backend.go:
func MyBackend() engine.Installer {
return &myBackend{}
}
func (f *computeEngineForward) Install(eng *engine.Engine) error {
eng.Register("mybackend", func(job *engine.Job) engine.Status {
job.Eng.Register("containers", func(job *engine.Job) engine.Status {
log.Printf("%#v", *job)
return engine.StatusOK
})
return engine.StatusOK
})
return nil
}
Then edit backends.go, and add your backend:
...
MyBackend().Install(back)
...
Solomon Hykes
Code and documentation copyright 2013-2014 Docker, inc. Code released under the Apache 2.0 license. Docs released under Creative commons.