The high availability of OBKV-HBase relies on the high availability of the entire link, namely, the high availability of OBServer nodes, ODP, and the client on the link. OBKV-HBase inherits the high availability capability of OceanBase AI Database on the server side and supports both direct connection mode and proxy mode in deployment.
High availability of OBServer nodes
OceanBase AI Database provides a variety of technologies for ensuring service high availability, including intra-cluster multi-replica disaster recovery, arbitration-based disaster recovery, and inter-cluster physical standby database disaster recovery. For more information, see High availability in OceanBase AI Database.
High availability of the link
High availability in proxy mode
OceanBase Database Proxy (ODP) is a dedicated proxy server for OceanBase AI Database. The high availability features of ODP cover the following two aspects:
- High availability of ODP services ODP services can be recovered promptly after a failure occurs.
- High availability of OceanBase AI Database
For example, ODP can detect and identify faulty OBServer nodes and add them to the blocklist. It can also shield applications from database faults through proper routing. When an OBServer node recovers, ODP removes it from the blocklist to fully utilize the resources of each server.
High availability in direct connection mode
In direct connection mode of OBKV-HBase, client requests are directly delivered to the corresponding OBServer node without being forwarded by ODP. In a distributed system, message addressing is the key to ensuring that requests are delivered to the correct processing node. A common solution is to cache and query the routing table before delivering messages.
In proxy mode, ODP is deployed as an independent component, and clients do not need to worry about routing. Messages are delivered to ODP based on load balancing algorithms or component deployment rules, and ODP performs routing based on the routing table. The benefit is that ODP has little dependency on the client: as long as the message protocol is followed, ODP can perform routing. The drawback is that ODP is deployed separately and usually does not reside on the same physical node as the client and database server, which introduces message forwarding and network overhead.
In direct connection mode, routing capabilities are integrated into the client, and message routing is completed within the client process. Messages are delivered directly to the corresponding database processing node, avoiding message forwarding and network overhead. However, because routing is implemented within the client, this mode strongly depends on the client.
The OBKV-HBase client also has high availability capabilities similar to those of ODP:
- The OBKV-HBase client caches routing information. When routing information changes due to scaling, failures, or load balancing of OBServer nodes, the OBKV-HBase client refreshes the routing table and applies the message retry mechanism to mask these changes on the database server side from applications, thereby ensuring service continuity.
- The URL of the Config Server is configured on the OBKV-HBase client so that the client can pull the Rslist from the Config Server. The client caches the list and then pulls routing information from the nodes in the list. When the Rslist changes, the OBServer node pushes the change to the Config Server.