HA and DR Policies¶
Whether you use DCS as the frontend cache or backend data store, DCS is always ready to ensure data reliability and service availability. The following figure shows the evolution of DCS DR architectures.
Figure 1 DCS DR architecture evolution¶
To meet the reliability requirements of your data and services, you can choose to deploy your DCS instance within a single AZ or across AZs.
Single-AZ HA¶
Single-AZ deployment means deploying an instance within a physical equipment room. DCS provides process/service HA, data persistence, and hot standby DR policies for different types of DCS instances.
Single-node DCS instance: When DCS detects a process fault, a new process is started to ensure service HA.
Figure 2 HA for a single-node DCS instance deployed within an AZ¶
Master/Standby DCS instance: Data is persisted to disk in the master node and incrementally synchronized and persisted to the standby node, achieving hot standby and data persistence.
Figure 3 HA for a master/standby DCS instance deployed within an AZ¶
Cluster DCS instance: Similar to a master/standby instance, data in each shard (instance process) of a cluster instance is synchronized between master and standby nodes and persisted on both nodes.
Figure 4 HA for a cluster DCS instance deployed within an AZ¶
Cross-AZ DR¶
The master and standby nodes of a master/standby or cluster DCS instance can be deployed across AZs (in different equipment rooms). Power supplies and networks of different AZs are physically isolated. When a fault occurs in the AZ where the master node is deployed, the standby node connects to the client and takes over data read and write operations.
Figure 5 Cross-AZ deployment of a master/standby DCS instance¶
Note
This mechanism applies in a similar way to a cluster DCS instance. Each shard (process) is deployed across AZs.
When creating a master/standby or cluster DCS instance, select a standby AZ that is different from the primary AZ.
Backup, configuration modification, and password change functions cannot be used during the fault.