Data Persistence

Although not mandatory, it is recommended to have a clear separation between compute workload and storage when setting up a Kubernetes cluster.

In particular, this approach makes it much easier to adopt a greenfield cluster upgrade strategy.

In order to achieve this, clusters need to be totally stateless and achieve data persistence outside of the cluster (using external RDBMS, NoSQL databases, or object storage).

We thus recommend to avoid the use of storageClasses, which allow self provisioning of Persistent Volumes (PVs), as it breaks the blue/green infrastructure paradigm that the DevOps Stack supports.