ODP requires high availability to ensure service continuity. User requests first reach ODP; if ODP fails, user requests cannot be served normally. ODP also needs to handle database cluster node failures and perform corresponding fault tolerance processing.
ODP is different from an OceanBase AI Database cluster. ODP does not have persistent state; all the data information that its operation relies on comes from accessing the database. Therefore, an ODP failure does not result in data loss. ODP also provides services as a cluster composed of multiple nodes. Which ODP node executes a user request is determined by the user's F5 or other load balancing components. In addition, if an ODP node fails, the load balancing component automatically removes it from service, ensuring that subsequent requests are not sent to the failed node.
During the ODP running process, ODP monitors the database cluster status in real time. On one hand, ODP obtains the health status of each database node and the location of each partition through system tables in real time. On the other hand, ODP detects the service status of database nodes through network connections. Once ODP detects an exception, it marks the corresponding node as faulty and switches the services to another node.