Data High Reliability

All data of GeminiDB databases is stored in three copies on the underlying DFV distributed storage.

• If a storage node is faulty, the distributed storage mechanism ensures that reads and writes are routed to other available nodes.
• If a shard on an available storage node becomes faulty, the upper-layer service reports the faulty shard and route reads and writes to other available nodes.
• Faulty shards can be automatically repaired through the gossip protocol. The repaired shard can continue to provide services. If a shard fails to be repaired, a new shard needs to be rebuilt.