Skip to content

The Definitive Guide to Docker Swarm

Updated: at 15:00

In this guide, you will learn everything you need to know about Docker Swarm and how to use it to scale and securely maintain your Docker projects.

With applications needing more and more computing resources and uptime of nearly 100% it becomes very hard to maintain and scale your software without some kind of management system. This is where Docker swarm comes into play. Docker Swarm provides an easy way to scale and maintain your containers and services.

This guide will show you all the important concepts, commands and the structure of the configuration file. It will also give a real-world example of how you can deploy a real-world application at the bottom of the article.

Table of contents

Open Table of contents

Why care about Docker Swarm?

Before we get into the technical details about what a Swarm actually is and where it can help us let’s discuss why someone would use it in the first place.

Load balancing

Swarm has a built-in load balancer that lets you specify how to distribute service and container between your different nodes. You can also expose ports for external load balancing services.

Integrated into the Docker Engine

Swarm is directly integrated into the Docker CLI and doesn’t require any additional orchestration software or other tools to create or manage a swarm.

Scaling

Swarm lets you define the number of tasks you want to run for each service. This number can be changed using a single command which is handled by the swarm manager.

Rolling updates

Swarm lets you apply service updates incrementally which means that it updates a specific amount of replicas at a time and your service will always be up even while updating.

What is a Swarm?

Docker Swarm is a cluster management and orchestration tool that makes it easy to scale and manage your already existing docker services. A swarm consists of multiple Docker hosts that run in the so-called swarm mode and act eighter as managers (managing member relationships) or as workers (run the services). A given Docker host can be a manager, worker or can perform both roles.

When creating a service in a swarm you define the optimal state of your service (number of replicas, ports of the service, network and storage resources, and more). Docker will try to maintain this desired state by restarting/rescheduling unavailable tasks and balancing the load between different nodes.

Nodes

A node is an instance of the Docker engine participating in the swarm. You can run one or multiple nodes on a single device, but production deployments typically include Docker nodes distributed across multiple physical devices.

Docker Swarm Nodes

Src:https://docs.docker.com/engine/swarm/how-swarm-mode-works/nodes/

Managers Nodes

Manager nodes distribute and schedule incoming tasks onto the Worker nodes, maintain the cluster state and perform orchestration and cluster management functions. Manager Nodes can also optionally run services for Worker nodes.

Cluster management tasks include:

There should always be multiple manager nodes in your swarm because of the following reasons:

That is why Docker recommends you implement an odd number of nodes according to your projects availability requirements.

Note: Docker recommends a maximum of seven manager nodes for a swarm.

Worker Nodes

Worker nodes are also instances of the Docker Engine whose sole purpose is to execute containers and services as instructed by the Manager Nodes.

To deploy your application to a swarm, you need at least one manager node. By default, all manager nodes are also workers. To prevent the scheduler from placing tasks on your manager node in a multi-node swarm, you need to set the availability to Drain.

Services

A service is the definition of the tasks to execute on the nodes. It is the primary root of user interaction with the swarm.

When you create a service, you specify which container image to use and which commands to execute inside running containers. You also define other options for the service including:

Here is an example of an HTTP server balancing its load on three replicas:

Swarm services

Src: https://docs.docker.com/engine/swarm/how-swarm-mode-works/services/

As you can see the service has three different tasks and each task invokes exactly one container. A task represents a slot where the scheduler can place a container. Once the container is live, the scheduler recognizes that the task is in a running state.

Task scheduling

A task carries a Docker container and the command that is executed inside the container. It is the atomic scheduling unit of the swarm. Tasks are assigned by the manager node to worker nodes according to the number of replicas set in the service.

When a service is created or updated, the orchestrator realizes the desired state by scheduling tasks. Each task is a slot that the scheduler fills by spawning a container which is the instantiation of a task. Now when one of these containers fails its health check or crashes, the orchestrator creates a new replica task that spawns a new container to replace the failing one.

The diagram below shows you how swarm mode accepts services and schedules tasks for the worker nodes.

Docker Swarm Scheduler

Src: https://docs.docker.com/engine/swarm/how-swarm-mode-works/services/

Replicated and global services

There are two different ways you can deploy a service, replicated and global.

Replicated services specify the number of identical tasks (replicas) you want to run. These replicas will then be split up on the different worker nodes and each serves the same content.

A global service is a service that runs one task on every node you have in your swarm and doesn’t need a pre-specified number of tasks. Global services are usually used for monitor agents or any other type of container that you want to run on every node.

Here is a visual representation of a three-service replica and a global service.

Global and replicated services

Src: https://docs.docker.com/engine/swarm/how-swarm-mode-works/services/

Getting started

Now that you know the key concepts of Docker swarm we can continue by learning the basics about creating and managing a cluster.

Installation

Swarm can be run on nearly any operating system and is very easy to install so let’s get into it.

Windows and Mac:

Compose is included in the Windows and Mac Desktop installation and doesn’t have to be installed separately. The installation instructions can be found here:

Linux:

If you are using a physical Linux machine or cloud hosting service as a host, simply follow the installation instructions provided by Docker.

Creating a Swarm

The first step after installing Docker on your machine is creating a swarm. For that, we need to run the following command.

docker swarm init --advertise-addr <MANAGER-IP>

The MANAGER-IP is the IP that the Swarm node manager will use to advertise the Swarm Cluster Service (If you are using Docker Desktop for Mac or Docker Desktop for Windows to test single-node swarm, simply run docker swarm init with no arguments).

Add nodes to swarm

With the swarm cluster created, we can now add a new node worker using the docker command provided by the output of the command above.

docker swarm join --token SWMTKN-1-41r5smr3kgfx780781xxgbenin2dp7qikfh9eketc0wrhrkzsn-8lbew6gpgxwd5fkn52l7s6fof 192.168.65.3:2377

The node will join the swarm as a worker node. If you want to give it manager privileges you either need to promote it or use another invite token.

The invite token can be displayed using the following command:

docker swarm join-token manager

Viewing the current nodes

The status of the current node in your swarm can be verified using the node ls command.

docker node ls

From the output of the command, you will see that your node is active and ready to use.

Promoting or demoting a node

You can promote or demote a node to the manager or worker role. This is useful if a special node is unavailable and you need to replace it.

# Promote node to manager
docker node promote docker-node

# Demote node to worker
docker node promote docker-node

You can also change roles using the update command.

docker node update --role manager docker-node

Leaving the swarm

A node can leave the swarm on their own or be removed by a manager node

# Leaving the swarm
docker swarm leave

#Removing a node from the swarm
docker node rm worker1

Deploy a service

After creating a swarm and adding your node to it you can proceed by running a service on it.

docker service create --replicas 4 --name hellogoogle alpine ping google.com

Here we create a service using the service create command and the following options:

After that you can use the service ls command to list all running services:

docker service ls

Output:
ID                  NAME                MODE                REPLICAS            IMAGE               PORTS
uebs1viyktap        hellogoogle         replicated          4/4                 alpine:latest

Scale a service

Now that you have a service running on your swarm you can scale the number of containers and services.

docker service scale <SERVICE-ID>=<NUMBER-OF-TASKS>
# For our example
docker service scale hellogoogle=10

This command will scale the number of replicas to 10.

Inspect a service

You can get the details of your service using the inspect command.

docker service inspect SERVICE_NAME

The output can also be customized by using extra tags like the —pretty tag to make the output more readable.

Deleting a service

Services can be removed using the rm command.

docker service rm hellogoogle

Updating a service

Docker swarm also lets you perform a rolling update on your running services. Rolling updates have the following advantages:

Basic updates:

Updates are performed using the docker service update command:

# Create a service
docker service create --name nginx --replicas 3 --update-delay 10s nginx:mainline

# Updating the image
docker service update --image nginx:stable nginx

Update-order:

Defines if the new container starts before the old one is killed or otherwise.

docker service update -d --update-order start-first nginx

Parallelism:

This flag will tell Swarm how many tasks it will update in parallel.

docker service update --update-parallelism 3 nginx

Defining a rollback

Docker swarm allows you to automatically rollback to previous versions of your services.

docker service rollback nginx

They can also automatically be executed when an error occurs in the update process.

docker service update --detach=false --update-failure-action rollback nginx

For more information visit the official documentation.

Drain a node

Drain availability prevents a node from receiving new tasks from the swarm. It also stops all tasks running on the node and launches replica tasks on the other available node with an active availability.

docker node update --availability drain worker1

Inspect the node:

You can inspect the node to see its availability:

docker node inspect --pretty worker1

Update it back to active:

After you have made the required changes on your node and are ready to enable it again.

docker node update --availability active worker1

Routing mesh

The routing mesh is a cluster-wide transport-layer (L4) load balancer that routes all incoming requests in the swarm to published ports of an available container on a node. It allows all the swarm nodes to accept connections on the services published ports.

Routing mesh

Src: https://success.docker.com/article/ucp-service-discovery-swarm

You can publish a port using the —publish flag or the short version -p:

docker service create -p 80:80 --name nginx --replicas 3 nginx:mainline

More information about how you can bypass the routing mesh and configure an external load balancer can be found on the official documentation.

Service logs

Logging is a very important topic for containerized applications and is also handled in Docker swarm.

Get the logs of a service:

Getting the logs of a service is very similar to getting the logs of a single container.

docker service logs hellogoogle

Following the logs:

You can also get a real-time view of your logs using the —follow flag.

docker service logs --follow hellogoogle

You can also use a custom logging driver or customize the output you get by the docker service logs command using more flags.

Manage sensitive data using secrets

A secret is a piece of data that should not be transmitted over a network or stored unencrypted. Docker provides a service to centrally manage such data and securely transmit it to only those containers that need access to it.

Docker secrets

Src: https://www.docker.com/blog/docker-secrets-management/

Secrets can be created, inspected and removed via the command line using the following commands.

Managing a secret

We can create a secret using the secret create command which has two parameters: the name of the secret and the secret itself.

echo "Secret" | docker secret create my_secret -

After creating the secret you can inspect it or display all secrets that are available on your machine.

# List all secrets
docker secret ls

# Inspect a specific secret
docker inspect SECRET_NAME

Lastly, you can remove a secret using the secret rm command.

docker secret rm my_secret

Passing a secret to a service

Secrets can be added to services on creating or while the containers are running using the following commands.

The --secret tag can be used to add a secret while creating a service.

docker service  create --name="nginx" --secret="my_secret" nginx:latest

Adding and removing secrets from running services can be done using the --secret-add and --secret-rm tags on the service update command.

docker service update --secret-rm="my_secret" nginx

Check out the docker secrets documentation for more information.

Limiting resources

Limiting the resources your services can access is a vital part of a container orchestration tool. Swarm makes this easy by providing tags that can be added to your service commands.

Here is an example of how you could use these tags:

docker service  create --name=nginx --limit-cpu 0.1 --limit-memory 1G nginx:latest

You can check the limitations of your service using the inspect command.

docker service inspect --pretty nginx

More information about limiting resources of services can be found in the documentation.

Labels

Adding labels to your nodes and services can be really helpful when having a big cluster and makes tasks like filtering easier.

Docker labels

Src: https://docs.docker.com/v17.12/datacenter/ucp/2.2/guides/admin/configure/add-labels-to-cluster-nodes/

The —label-add tag can be used to add a new label to an already existing node.

docker node update --label-add <key>=<value> <node-id>

Labels can also be added to services and containers which I will not go into in this article but you find more information in the official documentation or the docker service command docs.

Docker Stack files

Docker Stack is an extension of the already existing docker-compose file which lets you define deployment options for your swarm configuration like the number of replicas or resource limitations of your service.

If you have no experience with docker-compose yet I would recommend looking into this article first.

Creating a stack

As said before docker stack is an extension of the docker-compose file and just lets you define some extra attributes for your swarm deployment. These attributes can be defined using the deploy key in a compose file.

Important attributes include:

Here is an example of a stack deployment:

version: "3.7"
services:
  nginx:
    image: nginx:latest
    ports: ["8080:80"]
    deploy:
      replicas: 6
      update_config:
        parallelism: 2
        delay: 10s
      restart_policy:
        condition: on-failure
      resources:
        limits:
          cpus: "0.1"
          memory: 1G

More information about the available options for the deploy keyword can be found here.

Managing a stack

Stacks can be executed by running the docker stack deploy command and providing your stack file.

docker stack deploy -c docker-compose.yml nginxtest

You can also list all your stacks or the services of a specific stack using the ls and ps command.

# List all stacks
docker stack ls

# List the services of a specific stack
docker stack ps STACK_NAME

Removing a stack is similar to removing a service and can be done using the rm command.

docker stack rm STACK_NAME

Example

Now that we have gone through the theory of Swarm let’s see some of the magic we just talked about in action. For that, we are going to deploy a Nestjs GraphQL application which already includes a docker-compose file, so we can focus on the swarm configuration.

The project contains a local image that has to be stored in a registry before the file can be run as a swarm. If you don’t know what a registry is or why we need one I would recommend reading this article on the docker documentation.

Let’s start by cloning the repository with the finished boilerplate from Github.

git clone https://github.com/TannerGabriel/nestjs-graphql-boilerplate.git

This should give you the following folder structure:

Nestjs GraphQL Boilerplate folder structure

Nestjs GraphQL Boilerplate folder structure

As said above the project already contains a docker-compose.yml file that should look like this.

version: '3'

services:
  nodejs:
    build:
      context: ./
      dockerfile: Dockerfile
    restart: always
    environment:
      - DATABASE_HOST=mongo
      - PORT=3000
    ports:
      - '3000:3000'
    depends_on: [mongo]
  mongo:
    image: mongo
    ports:
      - '27017:27017'
    volumes:
      - mongo_data:/data/db

volumes:
  mongo_data: {}

Now we will need to make a few changes to the file, so we can upload the custom Node.js image to a registry (We will set up a local registry for testing purposes).

Pushing the image to the registry

First, we will need to add an image tag to the Node.js services and provide the registry where it should be saved (In our case localhost:5000).

services:
  nodejs:
    image: 127.0.0.1:5000/nodejs

Now we will create the local registry using the following command:

docker service create --name registry -p 5000:5000 registry:2

After the registry is running we can continue with pushing the local image to the registry using the push command.

docker-compose push

Adding the swarm config

With the image in place, we can go ahead and add the swarm configuration to the docker-compose file.

version: '3'

services:
  nodejs:
    image: 127.0.0.1:5000/nodejs
    build:
      context: ./
      dockerfile: Dockerfile
    restart: always
    environment:
      - DATABASE_HOST=mongo
      - PORT=3000
    ports:
      - '3000:3000'
    depends_on: [mongo]
    deploy:
      replicas: 6
      update_config:
        parallelism: 2
        delay: 10s
      restart_policy:
        condition: on-failure
      resources:
        limits:
          cpus: "0.1"
          memory: 1G
  mongo:
    image: mongo
    ports:
      - '27017:27017'
    volumes:
      - mongo_data:/data/db

volumes:
  mongo_data: {}

All right, let’s understand what’s going on here by walking through the code:

Deploying the service

That is it! We have finished our Docker files and can now move on to running the application. This is done using the following command:

docker stack deploy --compose-file docker-compose.yml stackdemo

As indicated by the terminal output, your services are now running and you can check it by either writing docker stack ls in the command line or visiting localhost:3000/graphql. You should see something similar to this:

Nestjs GraphQL playground

Source

Conclusion

You made it all the way until the end! I hope that this article helped you understand the Docker Swarm and how you can use it to improve your development and deployment workflow as a developer.

If you have found this useful, please consider recommending and sharing it with other fellow developers. If you have any questions or feedback, let me know in the comments down below.