GBase 8a, a core analytical database product developed by GBASE, has long held a leading position in the Chinese database industry, particularly in the analytical database sector. Over more than a decade of evolution and development, GBase 8a has grown alongside customer business needs, achieving large-scale application in industries such as finance and telecommunications. This article delves into the high availability features of GBase 8a from the perspective of meeting key customer demands.

Evolution of GBase 8a with Customer Needs

The evolution of the GBase 8a MPP Cluster product can be categorized into three main stages: column-store database, MPP + column-store database, and Logical Data Warehouse (LDW). The architecture has also undergone three major evolutions to meet the growing demands of customers, including data volume support, computational performance (both loading and querying), and high reliability, availability, and stability.

High Availability Features of GBase 8a

Based on years of customer application practice and iterative development, GBase 8a can meet the primary high availability needs of customers. Currently, it offers three levels of high availability capabilities: cluster-level, cluster node-level, and process-level.

(1) Cluster-Level High Availability

GBase 8a employs two techniques to achieve cluster-level high availability: data synchronization and mirror clustering.

• Data Synchronization: GBase 8a has developed a data synchronization tool that performs block-level incremental synchronization. This approach significantly improves the efficiency of synchronizing large amounts of data compared to traditional logic log-based synchronization. The tool supports incremental data synchronization between two homogeneous GBase 8a clusters, enabling the creation of active-active clusters to meet disaster recovery needs and achieve cluster-level disaster management.

• Mirror Clustering: This technology meets the needs of customers requiring active-active clusters within the same city. It allows users to create a mirror cluster within the federated architecture of GBase 8a. Data written to the primary cluster is synchronized in real-time to the backup cluster, which is transparent to upper-level applications. Additionally, it supports read-write separation, enabling read operations to access the backup cluster while supporting dual-write functionality.

Cluster-level high availability in GBase 8a satisfies disaster recovery, active-active cluster, and read-write separation requirements.

(2) Cluster Node-Level High Availability

GBase 8a clusters consist of three types of nodes: GCluster nodes, GNode nodes, and GCware nodes.

• GCluster Nodes: These form the scheduling cluster, responsible for managing access, authentication, and SQL parsing. Each GCluster node operates independently with a failover mechanism. If a node fails, other nodes take over the ongoing tasks, ensuring data consistency for SQL operations.

• GNode Nodes: These nodes store multiple data replicas. With multiple replicas, GNode nodes provide high availability. For instance, with three replicas, GBase 8a allows for the failure of two GNode nodes while still providing service through the remaining node.

• GCware Nodes: These manage metadata and data consistency using the Raft protocol. As long as the remaining GCware nodes meet the minimum requirement for the Raft protocol, the GCware cluster continues to provide service even if some nodes are unavailable.

(3) Process-Level High Availability

Core processes such as GNode, GCluster, and GCware run on each cluster node. These processes are monitored in real-time and can be promptly recovered if a failure occurs.

Conclusion

The high availability features of GBase 8a effectively meet customer needs, ensuring that it remains a leading product in the database industry, especially in the analytical database sector. GBase 8a provides a mature, stable, user-friendly, and high-performance analytical database, offering robust support for upper-level applications like data warehouses, data marts, business intelligence, and decision support systems.