In an application architecture, a database system is responsible for data storage and query processing. It is crucial for enterprise data security and business continuity. High availability is the primary consideration in database system architecture design. High availability requirements include high service availability and high data reliability. This topic describes the high availability technologies for shared storage disaster recovery services in an OceanBase AI Database cluster. Data reliability technologies are described in the backup and recovery topic.
Disaster recovery
OceanBase AI Database implements a shared storage disaster recovery deployment solution, providing users with high availability capabilities of RPO = 0 and RTO < 8s.
The OceanBase AI Database shared storage architecture ensures data integrity through object storage and log integrity through LogService, thereby providing the following high availability solutions:
Multi-replica high availability LogService solution based on the Paxos consensus protocol
This solution provides multi-replica high availability disaster recovery capabilities based on the Paxos consensus protocol. Read/write (RW) nodes directly write logs to multiple replicas of LogService through the Paxos protocol for persistence. Read-only (RO) nodes read logs from the nearest LogService replica and locally replay hot data. When a minority of replicas are unavailable (one replica is allowed to be unavailable in a three-replica cluster), the read and write operations of compute nodes are not affected, and services can continue normally.
High availability read/write (RW) node solution based on two-replica compute nodes
This solution relies on two read-only replicas to provide disaster recovery capabilities. When the read/write node becomes unavailable, the database automatically performs a disaster recovery switchover to restore service, ensuring no data loss (RPO = 0) and a fault recovery time within 8 seconds (RTO < 8s).
High availability solution based on single-replica deployment
This solution applies to the single-replica deployment mode. When only a single compute node is deployed, the single-replica startup mechanism can be used to provide disaster recovery capabilities. When the compute node becomes unavailable, a new compute node can be started to restore data and metadata from object storage and incremental data from LogService, thereby restoring full service capabilities. This solution ensures no data loss (RPO = 0) and a fault recovery time in minutes (RTO is in minutes).
Cross-AZ disaster recovery solution based on single-replica deployment
This solution applies to scenarios where AZ-level failures need to be prevented under the single-replica deployment mode. The deployment configuration is a single AZ for compute nodes and cross-AZ deployment for LogService. When the AZ where the compute node resides becomes unavailable, system tenants recover via REPLACE TENANT, and user tenants recover by changing locality, ensuring no data loss (RPO = 0).