Horizontal scaling increases a cluster’s capacity by adding nodes while retaining existing nodes. ScyllaDB redistributes data to new members, unlike vertical scaling, which adds resources to one larger server. section 1.2.2
Scale up a ScyllaDB cluster carefully:
- Start additional nodes with the desired configuration. section 9.4.3
- Add one node at a time, reducing cluster load while each node joins. section 9.4.3
- Wait until the new node is UN in nodetool status before adding another. section 9.4.3
- Balance nodes across racks, then run nodetool cleanup on every node. section 9.4.3
Adding capacity
Suppose an application needs more capacity. Add a configured node, wait for it to become UN, repeat across balanced racks, and finish with nodetool cleanup. Cleanup removes data transferred to the new node, saves disk space, and prevents stale data from being resurrected if token ownership changes later. section 9.4.3
This expansion reflects ScyllaDB’s distributed design: adding nodes improves capacity and supports fault tolerance, while the system favors eventual consistency and availability rather than relational ACID guarantees. section 1.4
