Docker Swarm: clustering, services, and tasks

Swarm is Docker’s built-in cluster manager. It groups several Docker hosts so you can treat them as one resource.

Two major components

1. Enterprise-grade secure cluster

   • Manage one or more Docker nodes as a cluster.
   • Encrypted distributed cluster store.
   • Encrypted networks.
   • Secure join tokens.

2. Orchestration engine for microservices

   • API for deploying and managing microservices.
   • Declarative manifest files.
   • Rolling updates, rollbacks, and scaling.

Nodes

A swarm consists of one or more Docker nodes. Each node is either a manager or a worker.

Managers:

• Manage cluster state.
• Dispatch tasks to workers.

Workers:

• Accept and execute tasks.

State is stored in etcd.

Security

Swarm uses Transport Layer Security (TLS) for:

• encrypted communication;
• node authentication;
• role authorization.

Services and tasks

The atomic scheduling unit is a swarm service. A service adds to a plain container:

• scaling;
• rolling updates;
• rollback.

A container wrapped in a service is a task or replica.

A service is an application running in the swarm. It has a replica count — how many copies of the task to run.

A task is essentially one container scheduled on a swarm node.

Networking

In Swarm mode, Docker creates:

• an ingress overlay network (default);
• docker_gwbridge to connect node daemons.

See Networking for CNM and driver details.

Related topics

• Orchestration — Swarm vs Kubernetes vs Mesos
• Service discovery — external discovery tools
• WordPress as a stack — example docker stack deploy