Replication and high availability

Replication factor

Each volume has a replication factor, chosen in the StorageClass with the strategy parameter:

Strategy	Copies	Tolerates	Typical use
RF1	1	No node failures	Development, caches, temporary data
RF2	2	1 node failure	Workloads with moderate requirements
RF3	3	2 node failures	Production (default)

Replication is synchronous: a write is only acknowledged to the application once every available replica has persisted it.

Failover

The application pod always accesses the volume through the primary. If the primary fails:

1. The secondary replicas detect the missing heartbeats.
2. They request quorum votes from the witness and the other replicas.
3. With a majority, a healthy replica is promoted to primary, and the NVMe-oF client reconnects through the new path.
4. When the failed node comes back, its replica is resynchronized automatically.

The system applies epoch-based fencing: a former primary that reappears after a network partition cannot accept writes.