Data distribution to nodes
In Hyper-Converged Database (HCD), the cluster manages the total amount of data as a ring. HCD divides the ring into ranges equal to the number of nodes, and each node handles one or more data ranges. Before a node can join the ring, you must assign it a token. The token value determines the node’s position in the ring and its range of data. HCD partitions table data across the nodes based on the row key.
To determine the node where the first replica of a row will live, HCD locates the node with a token value greater than that of the row key. Each node handles the region of the ring between the node itself and its predecessor, excluding the predecessor’s token value. When you sort the nodes in token order, the last node becomes the predecessor of the first node.
For example, consider a simple four-node cluster where all of the row keys managed by the cluster are numbers in the range of 0 to 100.
HCD assigns each node a token representing a point in this range.
In this simple example, the token values are 0, 25, 50, and 75.
The first node with token 0 handles the wrapping range from 76 to 0.
The node with the lowest token also accepts row keys less than the lowest token and more than the highest token.
HCD provides several types of partitioners for your cluster.
HCD assigns an initial_token value to each node so that each node handles roughly an equal amount of data.
