Running DC/OS on Google Compute Engine

The recommended way for deploying a OSS DC/OS cluster on GCE is using Terraform.

Terraform

Disclaimer: Please note this is a community driven project and not officially supported by Mesosphere directly.

Prerequisites

Install Google SDK

Run this command to authenticate to the Google Provider. This will bring down your keys locally on the machine for terraform to use.

$ gcloud auth login
$ gcloud auth application-default login

Configure your GCP SSH Keys

Set the private key that you will be using to your ssh-agent and set public key in terraform. This will allow you to log into to the cluster after DC/OS is deployed and also helps Terraform setup your cluster at deployment time.

$ ssh-add ~/.ssh/your_private_key.pem
$ cat desired_cluster_profile.tfvars
gcp_ssh_pub_key_file = "INSERT_PUBLIC_KEY_PATH_HERE"
...

Configure a Pre-existing Google Project

Currently terraform-dcos assumes a project already exist in GCP to start deploying your resources against. This repo soon will have support for terraform to create projects on behalf of the user via this document. For the time being a user will have to create this project before time or leverage an existing project.

$ cat desired_cluster_profile.tfvars
gcp_project = "massive-bliss-781"
...

Example Terraform Deployments

Quick Start

We’ve provided all the sensible defaults that you would want to play around with DC/OS. Just run this command to deploy a multi-master setup in the cloud. Three agents will be deployed for you. Two private agents, one public agent.

  • There is no git clone of this repo required. Terraform does this for you under the hood.
terraform init -from-module github.com/dcos/terraform-dcos//gcp
terraform apply -var gcp_project="your_existing_project"

Custom terraform-dcos variables

The default variables are tracked in the variables.tf file. Since this file can be overwritten during updates when you may run terraform get --update when you want to fetch new releases of DC/OS to upgrade too, its best to use the desired_cluster_profile.tfvars and set your custom terraform and DC/OS flags there. This way you can keep track of a single file that you can use manage the lifecycle of your cluster.

For a list of supported operating systems for this repo, see the ones that DC/OS recommends here. You can find the list that Terraform supports here.

To apply the configuration file, you can use this command below.

terraform apply -var-file desired_cluster_profile.tfvars

Advance YAML Configuration

We have designed this project to be flexible. Here are the example working variables that allows very deep customization by using a single tfvars file.

For advance users with stringent requirements, here are the DC/OS flags examples where you can simply paste your YAML configuration in your desired_cluster_profile.tfvars. The alternative to YAML is to convert it to JSON.

$ cat desired_cluster_profile.tfvars
dcos_version = "1.10.2"
os = "centos_7.3"
num_of_masters = "3"
num_of_private_agents = "2"
num_of_public_agents = "1"
expiration = "6h"
dcos_security = "permissive"
dcos_cluster_docker_credentials_enabled =  "true"
dcos_cluster_docker_credentials_write_to_etc = "true"
dcos_cluster_docker_credentials_dcos_owned = "false"
dcos_cluster_docker_registry_url = "https://index.docker.io"
dcos_overlay_network = <<EOF
# YAML
    vtep_subnet: 44.128.0.0/20
    vtep_mac_oui: 70:B3:D5:00:00:00
    overlays:
      - name: dcos
        subnet: 12.0.0.0/8
        prefix: 26
EOF
dcos_rexray_config = <<EOF
# YAML
  rexray:
    loglevel: warn
    modules:
      default-admin:
        host: tcp://127.0.0.1:61003
    storageDrivers:
    - ec2
    volume:
      unmount:
        ignoreusedcount: true
EOF
dcos_cluster_docker_credentials = <<EOF
# YAML
  auths:
    'https://index.docker.io/v1/':
      auth: Ze9ja2VyY3licmljSmVFOEJrcTY2eTV1WHhnSkVuVndjVEE=
EOF
gcp_ssh_pub_key_file = "INSERT_PUBLIC_KEY_PATH_HERE"

Note: The YAML comment is required for the DC/OS specific YAML settings.

Upgrading DC/OS

You can upgrade your DC/OS cluster with a single command. This terraform script was built to perform installs and upgrades from the inception of this project. With the upgrade procedures below, you can also have finer control on how masters or agents upgrade at a given time. This will give you the ability to change the parallelism of master or agent upgrades.

DC/OS Upgrades

Rolling Upgrade

Supported upgraded by dcos.io
Masters Sequentially, Agents Parellel:
terraform apply -var-file desired_cluster_profile.tfvars -var state=upgrade -target null_resource.bootstrap -target null_resource.master -parallelism=1
terraform apply -var-file desired_cluster_profile.tfvars -var state=upgrade
All Roles Simultaniously
Not supported by dcos.io but it works without dcos_skip_checks enabled.
terraform apply -var-file desired_cluster_profile.tfvars -var state=upgrade

Maintenance

If you would like to add or remove (private) agents or public agents from your cluster, you can do so by telling terraform your desired state and it will make sure it gets you there. For example, if I have 2 private agents and 1 public agent in my -var-file I can always override that flag by specifying the -var flag. It has higher priority than the -var-file.

Adding Agents

terraform apply \
-var-file desired_cluster_profile.tfvars \
-var num_of_private_agents=5 \
-var num_of_public_agents=3

Removing Agents

terraform apply \
-var-file desired_cluster_profile.tfvars \
-var num_of_private_agents=1 \
-var num_of_public_agents=1

Important: Always remember to save your desired state in your desired_cluster_profile.tfvars

Redeploy an existing Master

If you want to redeploy a problematic master (ie. storage filled up, not responsive, etc), you can tell terraform to redeploy during the next cycle.

NOTE: This only applies to DC/OS clusters that have set their dcos_master_discovery to master_http_loadbalancer and not static.

Master Node

Taint Master Node

terraform taint google_compute_instance.master.0 # The number represents the agent in the list

Redeploy Master Node

terraform apply -var-file desired_cluster_profile.tfvars

Redeploy an existing Agent

If you want to redeploy a problematic agent (i.e., storage filled up, not responsive, etc.), you can tell terraform to redeploy during the next cycle.

Private Agents

Taint Private Agent

terraform taint google_compute_instance.agent.0 # The number represents the agent in the list

Redeploy Agent

terraform apply -var-file desired_cluster_profile.tfvars

Public Agents

Taint Private Agent

terraform taint google_compute_instance.public-agent.0 # The number represents the agent in the list

Redeploy Agent

terraform apply -var-file desired_cluster_profile.tfvars

Experimental

Adding GPU Private Agents

Coming soon!

Destroy Cluster

You can shutdown/destroy all resources from your environment by running this command below:

terraform destroy -var-file desired_cluster_profile.tfvars