All Products
Document Center


Last Updated:May 19, 2023

Tair supports the standard, cluster, and read/write splitting architectures. You can select the architecture that suits your business requirements.


The following table lists the supported architectures. You can click the architecture name to view more details.



Standard architecture

  • Master-replica deployment model: An instance consists of a master node and a replica node. The master node handles your daily workloads, and the replica node stays in hot standby mode to ensure high availability. If the master node fails, the high availability (HA) system switches the workloads to the replica node within seconds after the failure occurs. This mechanism ensures the stability of your workloads.

  • Standalone deployment model: An instance contains only a single data node that cannot ensure high availability. Standard standalone instances are suitable for cache-only scenarios that do not require high data reliability. You can scale up the data node to meet high queries per second (QPS) requirements in a cost-efficient manner.

Cluster architecture

  • Distributed deployment model: Each data shard works as a master-replica high-availability node and provides automatic disaster recovery and failover.

  • Standalone deployment model: Each data shard contains only a single data node that cannot ensure high availability. Cluster standalone instances are suitable for cache-only or high QPS scenarios.


The cluster architecture supports the following connection modes:

  • Proxy mode: manages connections to your instance in an intelligent manner to reduce the costs of application development.

  • Direct connection mode: allows a client to bypass proxy nodes and directly access backend shards to reduce the network overhead and service response time. This mode is suitable for business scenarios that are latency-sensitive.

Read/write splitting architecture

Read/write splitting instances use the master-replica architecture to provide high availability. Read replicas are attached to the master node to facilitate data replication and linear scaling of read performance. Read replicas can alleviate performance issues caused by hotkeys. Read/write splitting instances are suitable for business scenarios that feature high read/write ratios. The read/write splitting architecture provides two types of instances:

  • Cloud disk-based instance: Read replicas synchronize data from the master node by using star replication. You can specify up to five read replicas.

  • Local disk-based instance: Read replicas use chained replication. You can specify one, three, or five read replicas.

Documentation applicability

You must understand that architectures and series types such as DRAM-based instances are different concepts. The descriptions and topics listed in the preceding table are applicable to related architectures and series types. For example, you can refer to the standard architecture information for different instance series types that use the architecture. This rule is also applicable to the cluster and read/write splitting architectures.