The source code for this lab exercise is available on GitHub.

In our previous article Running stateful workloads on Kubernetes with Rook Ceph, we saw how Kubernetes CSI enables us to take volume snapshots on supported storage backends as a first step towards protecting our data on Kubernetes. However, snapshots operate at the infrastructure level so they do not understand how applications operate, manage and structure their data. This implies that snapshots, by nature, are crash-consistent but not application-consistent. For busy stateful workloads such as databases processing many transactions per second, crash-consistency is insufficient for data protection since in-progress transactions are not recorded so restoring from a snapshot may still lead to data loss and leave the application in an inconsistent state.

Kanister provides a robust and flexible solution for defining your own actions for performing application-aware backups on Kubernetes. It is a CNCF sandbox project originally created by the Veeam Kasten team as an integral component of their enterprise-ready Kubernetes data protection platform. It does this by defining blueprints, which serve as templates for application-specific backup and restore logic. The backup administrator or application owner may then instantiate actions defined in these blueprints by creating ActionSets which perform the actual application-specific backup and recovery procedures.