My Profile Photo

Stefan Prodan


DevOps Consultant. Passionate about software architecture, domain driven design and DevOps. Loves programming in .NET, JS and Go


Terraforming Docker Swarm Clusters on Scaleway

As Docker Swarm is getting more features and looks more stable I’ve decided to take it for a spin and try out different scenarios on managing swarm and services running on it. Doing these experiments means that I have to easily create, scale up/down and destroy clusters. What better tool for the job than Terraform, and since I already use Scaleway for lab work, I’ve made a Terraform project to automate Docker Swarm clusters. The Scaleway Terraform provider is pretty basic, so if you are new to Terraform this little project will help you get started. Scaleway is one of the cheapest VPS providers out there, that means playing around with Terraform will cost you almost nothing.

Initial setup

Clone the repository and install the dependencies:

$ git clone https://github.com/stefanprodan/scaleway-swarm-terraform.git
$ cd scaleway-swarm-terraform

# requires brew
$ make init

Running make init will install Terraform and jq using Homebrew and will pull the required Terraform modules. If you are on linux, after installing Terraform and jq packages, run terraform init. Note that you’ll need Terraform v0.10 or newer to run this project.

Before running the project you’ll have to create an access token for Terraform to connect to the Scaleway API. Navigate to Scaleway dashboard and select the credentials option from your name dropdown. In the credential page add your public SSH key and create a new token named terraform.

Using this token and your access key, create two environment variables:

$ export SCALEWAY_ORGANIZATION="<ACCESS-KEY>"
$ export SCALEWAY_TOKEN="<ACCESS-TOKEN>" 

Usage

Create a Docker Swarm Cluster with one manager and two workers:

# create a workspace
terraform workspace new dev

# generate plan
terraform plan

# run the plan
terraform apply 

This will do the following:

  • reserves public IPs for each node
  • creates a security group for the manager node allowing SSH and HTTP/S inbound traffic
  • creates a security group for the worker nodes allowing SSH inbound traffic
  • provisions three VC1S servers with Ubuntu 16.04 LTS and Rancher boot script
  • starts the manager node and installs Docker CE using the local SSH agent
  • customizes the Docker daemon systemd config by enabling the experimental features and the metrics endpoint
  • initializes the manager node as Docker Swarm manager and extracts the join tokens
  • starts the worker nodes in parallel and setups Docker CE the same as on the manager node
  • joins the worker nodes in the cluster using the manager node private IP

The naming convention for a swarm node is in <WORKSPACE>-<ROLE>-<INDEX> format, running the project on workspace dev will create 3 nodes: dev-manager-1, dev-worker-1, dev-worker-2. If you don’t create a workspace then you’ll be running on the default one and your nods prefix will be default. You can have multiple workspaces, each with it’s own state, so you can run in parallel different Docker Swarm clusters.

Customizing the cluster specs via terraform variables:

terraform apply \
-var docker_version=17.06.0~ce-0~ubuntu \
-var region=ams1 \
-var manager_instance_type=VC1S \
-var worker_instance_type=VC1S \
-var worker_instance_count=2

You can scale up or down the Docker Swarm Cluster by modifying the worker_instance_count. On scale up, all new nodes will join the current cluster. When you scale down the workers, Terraform will first drain the node and remove it from the swarm before destroying the resources.

After running the Terraform plan you’ll see several output variables like the Swarm tokes, the private and public IPs of each node and the current workspace. You can use the manager public IP variable to connect via SSH and lunch a service within the Swarm.

$ ssh [email protected]$(terraform output swarm_manager_public_ip)

[email protected]:~# docker service create \
    --name nginx -dp 80:80 \
    --replicas 2 \
    --constraint 'node.role == worker' nginx

$ curl $(terraform output swarm_manager_public_ip)

You could also expose the Docker engine remote API and metrics endpoint on the public IP by running:

terraform apply -var docker_api_ip="0.0.0.0"

If you chose to do so, you should allow access to the API only from your IP. You’ll have to add a security group rule for ports 2375 and 9323 to the managers and workers groups.

resource "scaleway_security_group_rule" "docker_api_accept" {
  security_group = "${scaleway_security_group.swarm_managers.id}"

  action    = "accept"
  direction = "inbound"
  ip_range  = "<YOUR-IP-RANGE>"
  protocol  = "TCP"
  port      = 2375
}

Test your settings by calling the API and metrics endpoint from your computer:

$ curl $(terraform output swarm_manager_public_ip):2375/containers/json

$ curl $(terraform output swarm_manager_public_ip):9323/metrics

And last you can tear down the whole infrastructure with terraform destroy -force.

Code walkthrough

The first step in creating the Swarm cluster is to define the Scaleway provider, base image and boot script. You’ll have to use Racher boot script since Scaleway’s custom Xenial kernel is missing IPVS_NFCT and IPVS_RR.

provider "scaleway" {
  region = "${var.region}"
}

data "scaleway_bootscript" "rancher" {
  architecture = "x86_64"
  name_filter  = "rancher"
}

data "scaleway_image" "xenial" {
  architecture = "x86_64"
  name         = "Ubuntu Xenial"
}

As you can see, the Scaleway region is loaded from the region variables. This allows you to change the region when running Terraform apply. Besides the region there are more variables defined, all located in the variables.tf file.

variable "region" {
  default = "ams1"
}

variable "manager_instance_type" {
  default = "VC1S"
}

variable "worker_instance_type" {
  default = "VC1S"
}

variable "worker_instance_count" {
  default = 2
}

variable "docker_version" {
  default = "17.06.0~ce-0~ubuntu"
}

variable "docker_api_ip" {
  default = "127.0.0.1"
}

In order to create the Swarm manager, first you need to reserve a public IP for it. Terraform provisioners will use this IP to run commands on the manager node via SSH.

resource "scaleway_ip" "swarm_manager_ip" {
  count = 1
}

Having the public IP resource ready, the Scalewaly image and boot script you can define a server that will act as the Swarm manager.

resource "scaleway_server" "swarm_manager" {
  count          = 1
  name           = "${terraform.workspace}-manager-${count.index + 1}"
  image          = "${data.scaleway_image.xenial.id}"
  type           = "${var.manager_instance_type}"
  bootscript     = "${data.scaleway_bootscript.rancher.id}"
  security_group = "${scaleway_security_group.swarm_managers.id}"
  public_ip      = "${element(scaleway_ip.swarm_manager_ip.*.ip, count.index)}"

  connection {
    type = "ssh"
    user = "root"
  }

  provisioner "remote-exec" {
    inline = [
      "mkdir -p /etc/systemd/system/docker.service.d",
    ]
  }

  provisioner "file" {
    content     = "${data.template_file.docker_conf.rendered}"
    destination = "/etc/systemd/system/docker.service.d/docker.conf"
  }

  provisioner "file" {
    source      = "scripts/install-docker-ce.sh"
    destination = "/tmp/install-docker-ce.sh"
  }

  provisioner "remote-exec" {
    inline = [
      "chmod +x /tmp/install-docker-ce.sh",
      "/tmp/install-docker-ce.sh ${var.docker_version}",
      "docker swarm init --advertise-addr ${self.private_ip}",
    ]
  }
}

After Terraform creates the server based on the configuration block, it will connect via SSH to execute a series of commands. First it will create the directory for Docker engine systemd conf file, then it will render the conf file using the docker_conf template and it will copy this file on the manager server.

[Service]
ExecStart=
ExecStart=/usr/bin/dockerd -H fd:// \
  -H tcp://${ip}:2375 \
  --storage-driver=overlay2 \
  --dns 8.8.4.4 --dns 8.8.8.8 \
  --log-driver json-file \
  --log-opt max-size=50m --log-opt max-file=10 \
  --experimental=true \
  --metrics-addr ${ip}:9323

I’m using a template, so that at runtime you can change the IP variable in order to expose the Docker remote API and metrics endpoint on the internet. The default value of the IP is 127.0.0.1 so the API is only accessible from localhost. The template resource is defined as:

data "template_file" "docker_conf" {
  template = "${file("conf/docker.tpl")}"

  vars {
    ip = "${var.docker_api_ip}"
  }
}

Next the Docker CE script is copied on the server and executed. The script receives the Docker CE version to install, this way you’ll not end up with different Docker versions if you decide to scale up the cluster at a later time. After Docker is installed this node is initialized as the Swarm manager.

Before you can start creating the Swarm worker nodes you’ll need a way to store the join token generated by the manager. Storing data not exposed by the Terraform providers can be done using an external data source. The external data source protocol receives a JSON object and expects another JSON as output.

#!/usr/bin/env bash
set -e

eval "$(jq -r '@sh "HOST=\(.host)"')"

MANAGER=$(ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
    [email protected]$HOST docker swarm join-token manager -q)

WORKER=$(ssh -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
    [email protected]$HOST docker swarm join-token worker -q)

jq -n --arg manager "$MANAGER" --arg worker "$WORKER" \
    '{"manager":$manager,"worker":$worker}'

The above script extracts the host argument from the input JSON, connects to the manager node and returns the two tokens as a JSON object. Using this script you can define an external data source like this:

data "external" "swarm_tokens" {
  program = ["./scripts/fetch-tokens.sh"]

  query = {
    host = "${scaleway_ip.swarm_manager_ip.0.ip}"
  }

  depends_on = ["scaleway_server.swarm_manager"]
}

You need to mark this data source as dependent on the existence of the manager node or Terraform will execute it right after the manager’s public IP has been reserved and before the actual server has been created resulting in a SSH connection error.

Having the join token, you can proceed with the Swarm workers provision. The worker server configuration is similar to the manager with some additions. After installing Docker CE, using the worker token and the manager’s private IP these servers will join the Swarm as worker nodes.

resource "scaleway_ip" "swarm_worker_ip" {
  count = "${var.worker_instance_count}"
}

resource "scaleway_server" "swarm_worker" {
  count          = "${var.worker_instance_count}"
  name           = "${terraform.workspace}-worker-${count.index + 1}"
  image          = "${data.scaleway_image.xenial.id}"
  type           = "${var.worker_instance_type}"
  bootscript     = "${data.scaleway_bootscript.rancher.id}"
  security_group = "${scaleway_security_group.swarm_workers.id}"
  public_ip      = "${element(scaleway_ip.swarm_worker_ip.*.ip, count.index)}"

  # same provisiones as the manager

  provisioner "remote-exec" {
    inline = [
      "chmod +x /tmp/install-docker-ce.sh",
      "/tmp/install-docker-ce.sh ${var.docker_version}",
      "docker swarm join --token ${data.external.swarm_tokens.result.worker} ${scaleway_server.swarm_manager.0.private_ip}:2377",
    ]
  }

  # drain and remove the node on destroy
}

If the number of worker nodes is decreased, you need to make sure they also get removed from the Swarm. Terraform lets you react to destroy actions, so you can create a provisioner that will connect to the manager node and issue a remove command. You can also drain the node and make him leave the cluster before it gets destroyed.

  provisioner "remote-exec" {
    when = "destroy"

    inline = [
      "docker node update --availability drain ${self.name}",
    ]

    on_failure = "continue"

    connection {
      type = "ssh"
      user = "root"
      host = "${scaleway_ip.swarm_manager_ip.0.ip}"
    }
  }

  provisioner "remote-exec" {
    when = "destroy"

    inline = [
      "docker swarm leave",
    ]

    on_failure = "continue"
  }

  provisioner "remote-exec" {
    when = "destroy"

    inline = [
      "docker node rm --force ${self.name}",
    ]

    on_failure = "continue"

    connection {
      type = "ssh"
      user = "root"
      host = "${scaleway_ip.swarm_manager_ip.0.ip}"
    }
  }

If you want to restrict external access you can define group access security rules for the manager and the workers in security-groups.tf file. Please note that Scaleway doesn’t block access by default, so you need to make rules for each opened port on your server. If you need more Swarm managers you can promote any worker node as manager, just make sure you always have an odd number of manager or you’ll run into split brain scenarios if a network split happens.

Cluster monitoring with Weave Cloud Scope

With Weave Cloud Scope you can see your Docker hosts, containers and services in real-time. You can view metrics, tags and metadata of the running processes, containers or hosts. It’s the idea tool to visualize your Docker Swarm clusters and troubleshoot problems that may arise. Scope offers remote access to the Swarm’s nods and containers making it easy to diagnose issues in real-time.

If you don’t have a Weave Cloud account, you can apply for a free trial on Weaveworks website.

In order to deploy a Docker Swarm cluster integrated with Weave Cloud you’ll need to use the weave branch:

$ git checkout -b weave

Login into Wave Cloud and create a new instance, copy the service token and add the following environment variable:

$ export TF_VAR_weave_cloud_token="<SERVICE-TOKEN>"

Now you can run the Terraform project:

terraform workspace new weave

terraform apply \
-var docker_version=17.06.0~ce-0~ubuntu \
-var region=ams1 \
-var manager_instance_type=VC1S \
-var worker_instance_type=VC1S \
-var worker_instance_count=2 \
-var docker_api_ip=0.0.0.0 \
-var docker_api_ip_allow=86.124.244.168

Like before this will setup a three nodes Docker Swarm cluster and will deploy the Scope service on each node.

Let’s take a look at the provisioner code for the manager node:

  provisioner "remote-exec" {
    inline = [
      "chmod +x /tmp/install-docker-ce.sh",
      "/tmp/install-docker-ce.sh ${var.docker_version}",
      "docker swarm init --advertise-addr ${self.private_ip}",
      "curl -sSL git.io/scope -o /usr/local/bin/scope",
      "chmod a+x /usr/local/bin/scope",
      "WEAVESCOPE_DOCKER_ARGS='--restart unless-stopped' scope launch --service-token=${var.weave_cloud_token}",
      "iptables -A INPUT -p tcp --dport 2375 -s ${var.docker_api_ip_allow} -j ACCEPT",
      "iptables -A INPUT -p tcp --dport 2375 -j DROP"
    ]
  }

After installing Docker CE and initializing the Swarm manager, it deploys the Scope container using the Weave Cloud service token. It also grants access to the Docker remote API only to your IP.

Using the remote API you can deploy services from your machine:

$ export DOCKER_HOST=$(terraform output swarm_manager_public_ip)

$ docker service create \
    --name nginx -dp 80:80 \
    --replicas 6 \
    --constraint 'node.role == worker' nginx

$ curl $(terraform output swarm_manager_public_ip)

Now if you login into Weave Cloud, in the explore section you’ll be able to see the Docker Swarm cluster nodes and running containers.

Flow

That’s it! Due to Terraform awesomeness and Docker Swarm simplicity, automating cluster operations achieved with 200 code lines.

If you have any questions or suggestions, please leave a comment here or on GitHub at stefanprodan/scaleway-swarm-terraform.

comments powered by Disqus