Cluster

Understanding the SableDB Cluster

SableDB organizes its data into a distributed cluster using a system of slots and shards.

Slots: Data Distribution

Every key in SableDB is assigned a unique number between 0 and 16,383, determining its slot. Each node in the cluster is responsible for a specific range of these slots. If a client requests a key from a node that doesn't own its corresponding slot, the node returns a MOVED error, directing the client to the correct location.

Shards: High Availability

A SableDB cluster is built from one or more shards. Each shard consists of a primary node and can optionally include multiple replicas for redundancy. Ideally, slots are evenly distributed among these shards, with each managing 16384 / K slots in a cluster of K shards. This architecture can theoretically support up to 16,000 primary nodes, accommodating substantial workloads.

Centralized Cluster Management

SableDB employs a central database (itself a SableDB instance) to manage cluster communication and maintain its overall state. Each node periodically registers and updates its information (like owned slots and public address) in this central database, and also pulls information about other nodes from it.

Flexible Slot Migration

SableDB provides commands for transferring slot ownership between shards. For example, to move slot 1234 from Node_1 to Node_2, you'd simply connect to Node_1 via sabledb-cli and execute SLOT SENDTO <NODE_ID> 1234. The command returns OK upon successful completion.