09 January 2017

In my previous post, Deploy on Kubernetes GKE with Terraform, we've seen how to start to use kubernetes but in a very simple way. The next thing we would like to do is persist the configuration, so we don't need to reconfigure our bot each time we start the cluster. This post explain how to do it from the configuration created in the previous one.

Again we'll use Leanmanager bot but everything applies to any other system which needs to store configuration or data in a database. In the case of Leanmanager we are using Boltdb, a pure Go key/value store. Boltdb is great for development but it doesn't support to have more than one process opening the same database file, so it may be problematic if we want to have more than one Docker instance at the same time. Yet it's enough for our purposes and the process is similar for Consul which it's already in the Roadmap.

Create your persistent disks

If we want to persist data, we are going to need a disk, that's common sense. In GCE we can do it very easily:

gcloud compute disks create --size 1GB leanmanager-disk

But again, we want to do it in an automated way with Terraform. Use the following file leanmanager-disk.tf:

variable "disk_name" {
  default = "leanmanager-disk"

resource "google_compute_disk" "default" {
name  = "${var.disk_name}"
  type  = "pd-ssd"
  zone = "${var.region}"
  size  = "1"

If you want to know more about it, visit the Terraform Google_Compute_Disk reference documentation.

Tell the container about the disk

In our previous post, we've launched the bot using kubectl run. This is OK for simple configuration but if we need to have something more complex, it doesn't scale. We can create a pod, a group of one or more containers, using a YAML file like this:

apiVersion: v1
kind: Pod
  name: leanmanager
    name: leanmanager
    - image: antonmry/leanmanager:latest
      name: leanmanager
        - name: LEANMANAGER_TOKEN
        - name: LEANMANAGER_PATHDB
          value: /mnt
          # This name must match the volumes.name below.
        - name: leanmanager-persistent-storage
          mountPath: /mnt
    - name: leanmanager-persistent-storage
        # This disk must already exist.
        pdName: leanmanager-disk
        fsType: ext4

The file is auto-explanatory except the value LEANMANAGER_TOKEN_TEMPLATE. I don't want to hardcode the Token here because the file will be uploaded to Github. Instead of that, I want to use my local environment variable LEANMANAGER_TOKEN but this isn't supported yet in K8s, see Kubernetes equivalent of env-file in Docker.

So I've created a YAML template and in the Terraform file changed the last local-exec to:

  provisioner "local-exec" {
    command = "cp leanmanager-pod-template.yaml leanmanager.tmp.yaml && sed -i -- 's/LEANMANAGER_TOKEN_TEMPLATE/${var.LEANMANAGER_TOKEN}/g' leanmanager.tmp.yaml"

  provisioner "local-exec" {
    command = "kubectl create -f leanmanager.tmp.yaml"

  provisioner "local-exec" {
    command = "rm -f leanmanager.tmp.yaml"

Basically, I'm replacing strings with sed. Other more sophisticate approaches are possible as K8s secrets or Ansible, but this is simple and enough for the task we want to do.

Create the pod and test

Time to create the cluster and the pod:

terraform plan

The bot should connect. Now we can do some changes in the configuration, delete the pod:

kubectl delete pod leanmanager

Create it again:

kubectl create -f leanmanager.yaml

And check the status with the following commands and, once it's in Running state, see if everything has been persisted:

kubectl get pod leanmanager
kubectl logs leanmanager


Persist data in Kubernetes is quite easy, even if you are going to do it automatically.

If you want to check all the files, the full project and the associated PR are in Github.

Not already linked but useful resources