This repo contains useful things for playing with Docker Swarm Mode.

There is automation here for running a swarm with multiple hosts on a single machine using containers as managers and workers.

There are also several stack files which serve as a starting point for understanding the scheduler and playing with different use-cases.

Run a Swarm in Containers

cd swarm
source start

This exports some environment vars into your shell. If you are already part of a swarm, ignore the error.

docker stack up admin -c ../admin/docker-compose.yaml
# alternatively, `cd ../admin && docker-compose up -d`

In a few moments, you should see a visualization @ localhost:8090 showing your host and any swarm services.

docker-compose up -d worker
# you should see a worker node join the swarm
docker-compose scale worker=4
# 3 more should join in a few seconds

You now have a multi-host swarm that you can play around in. For speed, configure your docker engine to use:

"registry-mirrors": [
  "http://localhost:5000"
]

When you're done, cd swarm && ./kill.

Some Helpful Commands

export DOCKER_HIDE_LEGACY_COMMANDS=true
docker --help
docker node ls
docker node --help
docker stack ls
docker stack --help
docker service ls
docker service --help
docker service inspect <SERVICE> --pretty

# list all node's engine labels:
for nid in `docker node Wls -q`; do docker node inspect $nid -f "{{.Description.Engine.Labels}}"; done

How it Works

The start script:

• ensures a swarm exists
• exports the join tokens as env vars
• exports the {{.Swarm.NodeAddr}} as $HOST_IP

swarm/docker-compose.yaml depends on these env vars. It contains services for managers and workers. These services run docker in docker (dind) as priviledged containers on the host. The entrypoint is overridden to run an inline shell script that:

• starts the container's dockerd with overlay2 and the admin_mirror
• traps interrupts
• waits until docker info succeeds
• joins the host's swarm with the service's $TOKEN

You can classify nodes by adding --label flags to the $opts. These are engine labels which shouldn't be confused with swarm node labels.

Using Stacks

Stack files allow you to use the docker-compose schema to declare a desired state for many related services. These services will achieve their desired state through the swarm scheduler.

This repo contains a few different stack files that demonstrate neat use cases.

You can deploy a stack by running:

docker stack up -c stack_name/docker-compose.yaml stack_name

You can change the file and run the command again to watch swarm remediate the state.

The admin stack is recommended since it runs services that help you understand and speed up what you are doing.

Try the minio stack if you want to play with a self-hosted, distributed, S3 compatible object store.

The routing stack is great for trying traefik, a dynamically configured load-balancer.