klondike

klondike

klondike will deploy and manage a Kubernetes cluster using AWS CloudFormation and Ansible. klondike can send metrics to Datadog, aggregate logs via Logstash, and securely manage secrets using Ansible Vault.

This project is opinionated in how it deploys Kubernetes and the related services. All feedback and code contributions are welcome, but the scope of this project is only intended to reflect one method of deployment. klondike will not grow to encompass other deployment methodologies.

klondike is licensed under Apache 2.0. See LICENSE for more details.

Overview

klondike deploys three types of hosts: controllers, workers and a bastion. The controllers operate the Kubernetes control plane, while the workers run pods scheduled to them via the Kubernetes API.

The bastion is an administrative host that facilitates the management of the klondike cluster. An operator uses the bastion for initial cluster deployment, as well as on-going cluster management tasks.

Quickstart

This quickstart guide is intended to get a cluster spun up in as few steps as possible. For more thorough information, visit the klondike documentation.

The creation of a new cluster depends on having installed a few tools locally. Start off by getting these installed, then continue on with cluster bootstrapping.

Install Tools

These tools are used to generate the configuration for a new klondike cluster:

• cfssl
• awscli
• ansible 2.0+

You must have locally-configured credentials for your AWS account working with the aws CLI tool. Ensure you've chosen the appropriate profile before deploying anything!

export AWS_PROFILE=<YOUR-PROFILE-NAME>

Install cfssl

For OS X users, this is available via homebrew:

brew install cfssl

On other operating systems, cfssl must be installed manually. Follow the instructions on GitHub.

Install awscli & netaddr

These tools are installed using pip:

pip install awscli netaddr

Install ansible

For OS X users, this is available via homebrew:

brew install ansible

Linux users can rely on packaging. For example, on Fedora 23:

dnf install ansible

Bootstrap & Deploy

1. Check out the project and cd into the directory:

   git clone [email protected]:planetlabs/klondike.git
   cd klondike/
   

2. Choose a name for your cluster, and set it in an environment variable:

   export CLUSTER=<YOUR-CLUSTER-NAME>
   

3. Create a workspace for your cluster in the git checkout directory and copy the sample config file:

   mkdir -p clusters/$CLUSTER/
   cp contrib/main.yml.sample clusters/$CLUSTER/main.yml
   

4. Update your cluster config file, clusters/$CLUSTER/main.yml, with all unset fields. It is encouraged to use the default values, when available.

5. If the S3 bucket you chose for your cluster does not yet exist, create it now:

   aws s3 mb s3://<YOUR-BUCKET>
   

6. From the root of the git checkout directory, run the configure.yml playbook:

   # Note: If you run into errors here, check your ansible version.
   ansible-playbook -e cluster=$CLUSTER configure.yml
   

7. Upload the created cluster configuration to the S3 bucket you chose for your cluster (see main.yml):

   ./contrib/s3-config push $CLUSTER <YOUR-BUCKET>
   

8. Create the CloudFormation stack and wait for creation to complete. This can take a while!

   ./clusters/$CLUSTER/create-stack.sh
   

9. Track the progress of your cluster using contrib/cluster-info.py. Continue to the next step once your stack reports CREATE_COMPLETE:

   python contrib/cluster-info.py $CLUSTER
   

10. SSH to your bastion host as ubuntu with the cluster's deploy key. The IP of your bastion is available via the contrib/cluster-info.py script:

    ssh -i clusters/$CLUSTER/id_rsa ubuntu@<YOUR-BASTION-IP>
    

11. Finish deploying your infrastructure:

    cd /home/ubuntu/klondike/
    ansible-playbook -e cluster=$CLUSTER site.yml
    

12. Finally, deploy the cluster-level components onto Kubernetes. This may take a few attempts as we must wait for the Kubernetes API to become available:

    ansible-playbook -e cluster=$CLUSTER cluster.yml
    

Note: some environments require associating this VPC with the hosted zone in order to resolve the API hostname.

Now your cluster is ready to go. Validate everything is working by querying the cluster for all nodes:

kubectl get nodes