Updated on 2022-12-16 GMT+08:00

Cluster DR

Background

GaussDB(DWS) clusters support disaster recovery. Data of the primary cluster is periodically synchronized to a standby cluster in another region or AZ. After considering several typical DR methods in the industry, such as dual-write on the application side and multiple remote copies, as well as the architecture characteristics of GaussDB(DWS) and the replicability of the DR solution, we have chosen the dual-cluster DR.

Technical Principles

Dual-cluster DR architecture:

  • Data synchronization: Node-to-node data synchronization is used between two clusters with the same number of DNs based on the MPP distribution characteristics. The network between the two clusters must be connected mutual trust must be configured.
  • Periodic synchronization: Data is synchronized at a configurable interval.
  • Data content: Cluster data including row-store data, column-store data, library data, and configuration information is synchronized.

Cluster switchover:

There are two types of DR switchover. One is planned switchover, which is used for DR drills. The other is failover, which is performed upon cluster faults.

For a planned switchover, if the RPO is 0, the primary cluster synchronizes data to the standby cluster and becomes the standby cluster. During a failover, the standby cluster immediately becomes the primary cluster. In this case, the RPO is not 0.

Benefits

The dual-cluster HA DR is loosely coupled and is selected by vendors such as Oracle and DB2. The advantages are as follows: (a) The two clusters are independent of each other, and the standby cluster does not affect the primary cluster. (b) The dual-cluster switchover can easily support the software upgrade and application upgrade of the major versions. It helps with the data security protection of large-scale MPP clusters.

For more information, see DR Overview.