Skip to content

Managing databases in Kubernetes

More or less after the deprecation of the bitnami charts, I was faced with a lot of problems with my databases. To avoid a similar incident in the future, I prefer to start deploying my own databases in the cluster and for achieving it that I prefer to have an operator that will make things easier.

cloudnative-pg

Since most of my deployments were postgresl DBs I will go with a operator for the PostgreSQL. This operator seems to be the most mature and has a lot of contributors so it shoud be future proof.

helm repo add cnpg https://cloudnative-pg.github.io/charts
helm repo update
kubectl create namespace databases
helm install cnpg cnpg/cloudnative-pg -f operator_values.yml --version v0.28.0 --namespace databases

kubectl apply -f snapshots.yml

We will be also using a special db storage class for the databases based on the suggestions of the operator:

kubectl apply -f db_storage.yml

After the operator is deployed we can create databses. We provide 2 examples. One for highly available cluster:

kubectl create namespace test-database

kubectl create secret generic db-user-pass --namespace test-database \
    --from-literal=username=app \
    --from-literal=password=$(head -c 512 /dev/urandom | LC_CTYPE=C tr -cd 'a-zA-Z0-9' | head -c 64)

helm install database cnpg/cluster -f ha_db_values.yml --version v0.3.1 --namespace test-database

The high availability one, does not replicate the data on the level of longhorn, but each database has their own non duplicate volume and they sync the data between the instances of the databases

The "normal" not highly available databases:

kubectl create namespace test-database

kubectl create secret basic-auth db-user-pass --namespace test-database \
    --from-literal=username=app \
    --from-literal=password=$(head -c 512 /dev/urandom | LC_CTYPE=C tr -cd 'a-zA-Z0-9' | head -c 64)

helm install database cnpg/cluster -f non_ha_db_values.yml --version v0.3.1 --namespace test-database

The non high availability example just has one instance of the database and uses longhorn for replication of the data.

Backup for databases

Currently it is not possible to add automated backups via the helm charts. But the cloudnative-pg supports backups via snapshots. So to work around it we will do it manually.

We will have to change the definition of the Cluster manually. For the examples above we have to execute:

KUBE_EDITOR="nano" kubectl -n test-database edit cluster database-cluster

and add in the definition of the cluster: 

  backup:
    volumeSnapshot:
       className: default-snapshot-class

To trigger immediately a backup we can use the backup.yaml. In other database we have to change name: database-cluster to the right name.

kubectl plugin

Instructions from the official site: 

curl -sSfL \
  https://github.com/cloudnative-pg/cloudnative-pg/raw/main/hack/install-cnpg-plugin.sh | \
  sudo sh -s -- -b /usr/local/bin

Resources

  • https://medium.com/@camphul/cloudnative-pg-in-the-homelab-with-longhorn-b08c40b85384
  • https://cloudnative-pg.io/documentation/current/kubectl-plugin/