Set up PostgreSQL cluster on Kubernetes in 5 minutes

Docker first

The world has become containerised and where before we’d spend hours trying to perfect an environment to then begin installing an application, we can now simply load the container and set some environment variables.

Productionising

However, rarely are these apps ready to go. If we’re going to run them in production we need to think about things like high availability, reliability, secrets and backups. Enter Kubernetes, which allows us to configure multiple instances of an app into a cluster and distribute volumes over multiple nodes.

PostgreSQL

Postgres is a relational database. As part of the data layer, it represents the ‘harder’ sort of deployment: one that contains state that we care about.

Getting started with a cluster

To preserve that state reliably (and be performant) we cluster the app, which means instantiating multiple instances of those containers and getting them to talk to each other.

Install Helm

Download and install Kubernetes Helm, which is sort of like a package manager for Kubernetes:

wget https://get.helm.sh/helm-v3.2.4-linux-amd64.tar.gz
tar -zxvf helm-v3.2.4-linux-amd64.tar.gz

Add the Bitnami repo

Bitnami offer a pre-configured set of Helm charts for multi-node PostgreSQL:

helm repo add bitnami https://charts.bitnami.com/bitnami

Fire up cluster

Create a cluster configuration file (values.yaml), for any settings that we don’t want to be default. I’ve gone with 3 nodes rather than 2.

---
postgresql:
    replicaCount: 3
...

Next we spin up the PostgreSQL cluster, confined to a namespace to make it easier to clean up afterwards:

RELEASE_NAME="hapostgres"
RELEASE_NAMESPACE="${RELEASE_NAME}"
helm install ${RELEASE_NAME} bitnami/postgresql-ha --create-namespace --namespace=${RELEASE_NAMESPACE}

Inspecting the cluster

Helm is an over-the-top package manager, so underneath it we’re still using kubectl, from which the former inherits its settings (typically .kube/config).

kubectl get all -n ${RELEASE_NAMESPACE}

Shutting down

Helm makes this pretty easy:

helm delete ${RELEASE_NAME} --namespace=${RELEASE_NAMESPACE}

but we just need to tell Longhorn (or whatever persistence layer you’re using on your K8S cluster) to clean up the volumes and finally, delete the containing namespace:

kubectl delete pvc --all -n ${RELEASE_NAMESPACE}
kubectl delete namespace ${RELEASE_NAMESPACE}