Disaster recovery
What to do if you lose a majority of the nodes or a whole data centre.
A cluster only survives the loss of a minority of replicas per range. To lose more without losing data, you need backups outside the cluster.
How it works
While every range keeps a majority of its replicas, the cluster recovers on its own: it elects new leaders and recreates the missing replicas. If a range loses its majority, it stops accepting writes until its nodes come back. If the nodes do not come back, the only way out is restoring a backup into a new cluster.
What it means for your application
Schedule backups with teseladb admin backup and keep them off the cluster's machines. Test the restore now and then. What was written after the last backup is lost if you have to restore.
Limits
A backup is consistent at its timestamp. Restoring needs an empty cluster.