Database High Availability

GeminiDB uses the decoupled compute and storage architecture. Multiple DB instances in a cluster share the underlying distributed storage. The faults that may occur include compute node fault and storage node fault.

Figure 1 Diagram

• Thanks to the shared storage, data restoration is not required when a compute node becomes faulty. Once a node is faulty, other nodes that are running properly can directly take over services in seconds, and this process is unnoticeable to customers' applications due to the persistent connections provided by the upper-layer proxy.
• There are three data copies in the underlying storage. A single point of failure will not cause the entire system to be unavailable.
  Figure 2 Underlying storage diagram