This repository contains my preferred method of standing-up a GKE Kubernetes cluster, with Rancher 2.x as an orchestration layer, and a cluster-wide Traefik ingress behind a GCP Layer 4 load balancer.

Important note: This is a method appropriate for Rancher 2.x < 2.1. The supported installation to a target cluster is now via Helm. This repository will be updated accordingly, when my time allows.

This may not represent the best practice for your organization or business requirements and no warranty is made regarding suitability. Use at your own risk and with due diligence as to the contents of the configuration included.

The goal is to install Rancher 2.x on a GKE cluster, configure Traefik to handle all the cluster's HTTP/HTTPS termination (including for Rancher itself) and take advantage of GKE's hosted K8s master while Rancher stores its configuration in the cluster's etcd store.

1. Create a GKE cluster in the GCP web console or CLI. Optionally enable network policy (see note below, however) and disable HTTP load balancing, as we will be doing our own via Traefik. I also suggest disabling basic auth and client certificate issuance.
2. Download and install the GCP gcloud utility and kubectl, if you don't already have them.
3. Fetch credentials for your cluster:

   gcloud container clusters get-credentials NAME [--region=REGION | --zone=ZONE, -z ZONE] [GCLOUD_WIDE_FLAG …]

4. Deploy the required cluster role bindings in cluster-admin.yml. This will require your GCP account to either have container service admin permissions (preferred/easier), OR use admin credentials.

   • Give your account admin permissions:

     kubectl create clusterrolebinding owner-cluster-admin-binding \
         --clusterrole cluster-admin \
         --user <account>
     kubectl apply -f k8s/cluster-admin.yml

   • Use admin credentials:

     gcloud container clusters describe <cluster name> --zone <zone> | grep password
     kubectl config set-credentials <credentialname> --username=admin --password=<password from above>
     kubectl config set-context <context name> --user=<credentialname> --cluster=<full GKE cluster name>
     kubectl --context=<context name> apply -f k8s/cluster-admin.yml

     Where <full GKE cluster name> is the name key of the cluster entry in your ~/.kube/config file, in the rough format of gke_PROJECTNAME_ZONE|REGION_CLUSTERNAME.

     If you used basic auth credentials to create role bindings, you may wish to set "Basic Authentication" to Disabled in your cluster settings, at this point.

5. Create the etcd operator (supervisor) and some basic cluster configuration, including the L4 load balancer: kubectl apply -f k8s/cluster.yml. If you wish to use a previously-reserved static IP, specify loadBalancerIP in the service spec.
   1. If you did not use a previously-reserved IP, retrieve the L4 load balancer IP: kubectl -n cluster-services get service. It will be listed as the EXTERNAL-IP. Optionally reserve it as a static IP in your VPC network configuration in the cloud console, to prevent inadvertent loss.
6. Create the etcd cluster for Traefik: kubectl apply -f k8s/etcd.yml. You may want to consider an etcd cluster backup or persistence strategy, which is out of scope, here.
7. Store the traefik config (replacing your hostmaster address for the default in the ACME config): kubectl apply -f k8s/storeconfig.yml.
8. Set a DNS record for rancher, on the load balancer IP. Place this hostname for CATTLE_SERVER_URL in k8s/cattle.yml.
9. Start Traefik: kubectl apply -f k8s/traefik-controller.yml.
10. Start Rancher: kubectl apply -f k8s/cattle.yml.
11. Log in to Rancher (which should now have a Let's Encrypt TLS certificate) and set your admin password.
12. Validate in the Google Cloud console that all your workloads (including system objects) are in an OK status.
13. If you are wishing to use Rancher's network policy features, there is a ticket relating to Rancher's support for Calico in Rancher >= 2.0.8. Prior to this version, it was possible to trick Rancher into creating project-scoped network policy rules on Calico, but this was changed and at the moment, network policies are only supported on RKE- created clusters, and if you opt in.
14. If you wish to use Google's cloud SQL with a proxy container, consult k8s/mysql.yml and this documentation. You will need to create a JSON snippet of credentials and store them in a secret.

Contributions are welcome in the form of pull requests.

Copyright 2018 Brad Jones LLC. MIT license.