What to protect.
WHAT NEEDS BACKING UP
Data in volumes Databases, through their own tools Configuration describing the system Secrets, stored securely Images, if they cannot be rebuilt
WHAT DOES NOT
Containers themselves.
WHY
They are recreated from images.
WHAT TO PREFER FOR DATABASES
The database's own dump or backup tool.
WHY NOT A VOLUME SNAPSHOT ALONE
A snapshot of a running database may not be consistent.
WHAT MAKES A SNAPSHOT USABLE
Quiescing the database first, or a tool that coordinates.
WHAT TO BACK UP FOR THE CLUSTER ITSELF
The manifests, which should be in version control anyway The state store, for managed clusters where it is yours
WHY MANIFESTS IN VERSION CONTROL IS THE REAL ANSWER
The cluster can be rebuilt from them.
WHAT THAT IMPLIES
Anything created by hand is not recoverable.
WHAT TO AVOID
Changing the cluster directly.
WHAT TO TEST
Rebuilding into a new cluster from manifests and backups.
WHY
It is the only way to know the backups are complete.
WHAT TO SCHEDULE
That test, periodically.
WHAT TO DOCUMENT
The recovery procedure, step by step.
WHAT TO STORE BACKUPS
Outside the cluster, and outside the same provider account.