Google Kubernetes Engine (GKE) with a Canary Deployment
This example provisions a Google Kubernetes Engine (GKE) cluster, using infrastructure-as-code, and then deploys a Kubernetes Deployment into it, to test that the cluster is working. This demonstrates that you can manage both the Kubernetes objects themselves, in addition to underlying cloud infrastructure, using a single configuration language (in this case, Python), tool, and workflow.
Prerequisites
Ensure you have Python 3 and the Pulumi CLI.
We will be deploying to Google Cloud Platform (GCP), so you will need an account. If you don’t have an account, sign up for free here. In either case, follow the instructions here to connect Pulumi to your GCP account.
This example assumes that you have GCP’s gcloud
CLI on your path. This is installed as part of the
GCP SDK.
Running the Example
After cloning this repo, cd
into it and run these commands. A GKE Kubernetes cluster will appear!
Create a new stack, which is an isolated deployment target for this example:
$ pulumi stack init dev
Set the required configuration variables for this program:
$ pulumi config set gcp:project [your-gcp-project-here] $ pulumi config set gcp:zone us-west1-a # any valid GCP zone here $ pulumi config set password --secret [your-cluster-password-here] $ pulumi config set master_version #any valid master version
By default, your cluster will have 3 nodes of type
n1-standard-1
. This is configurable, however; for instance if we’d like to choose 5 nodes of typen1-standard-2
instead, we can run these commands:$ pulumi config set node_count 5 $ pulumi config set node_machine_type n1-standard-2
This shows how stacks can be configurable in useful ways. You can even change these after provisioning.
Deploy everything with the
pulumi up
command. This provisions all the GCP resources necessary, including your GKE cluster itself, and then deploys a Kubernetes Deployment running nginx, all in a single gesture:$ pulumi up
This will show you a preview, ask for confirmation, and then chug away at provisioning your cluster:
Updating stack 'gcp-ts-gke-dev' Performing changes: Type Name Plan + pulumi:pulumi:Stack gcp-py-dev create + ├─ gcp:container:Cluster gke-cluster create + ├─ pulumi:providers:kubernetes gkeK8s create + └─ kubernetes:apps:Deployment canary create + └─ kubernetes:core:Service ingress create ---outputs:--- kubeConfig: "apiVersion: v1\n..." info: 5 changes updated: + 5 resources created Update duration: 2m07.424737735s
After about two minutes, your cluster will be ready, and its config will be printed.
From here, you may take this config and use it either in your
~/.kube/config
file, or just by saving it locally and plugging it into theKUBECONFIG
envvar. All of your usualgcloud
commands will work too, of course.For instance:
$ pulumi stack output kubeconfig --show-secrets > kubeconfig.yaml $ KUBECONFIG=./kubeconfig.yaml kubectl get po NAME READY STATUS RESTARTS AGE canary-n7wfhtrp-fdbfd897b-lrm58 1/1 Running 0 58s
At this point, you have a running cluster. Feel free to modify your program, and run
pulumi up
to redeploy changes. The Pulumi CLI automatically detects what has changed and makes the minimal edits necessary to accomplish these changes. This could be altering the existing chart, adding new GCP or Kubernetes resources, or anything, really.Once you are done, you can destroy all of the resources, and the stack:
$ pulumi destroy $ pulumi stack rm