Each primary zone of PolarDB clusters for Cluster Edition has multiple compute nodes that include one primary node and one or more read-only nodes.


Only PolarDB Cluster Edition clusters support multiple nodes. Single Node and Archive Database do not support multiple nodes. For more information about the features and differences of each edition, see Product editions.

Multi-node architecture


Each PolarDB cluster contains one primary node and two or more read-only nodes. A PolarDB cluster can contain a maximum of 15 read-only nodes and a minimum of one read-only node. All the nodes in the same cluster have the same specifications.

If a system failure occurs, the multi-node architecture supports automatic failovers from the primary node to the read-only nodes. This ensures high availability of the cluster.

In addition, PolarProxy of a PolarDB cluster can be used to perform read/write splitting on the nodes. For more information, see PolarProxy.

Add or delete read-only nodes

A PolarDB cluster of Cluster Edition can contain up to 15 read-only nodes. Each cluster must have at least one read-only node to ensure high availability. All nodes in a cluster use the same specifications.

You can manually add or delete read-only nodes to modify the cluster performance based on your requirements. For more information, see Add or remove read-only nodes.

  • It takes 5 minutes to add a node. This operation does not affect the database.
  • After you delete a node, transient connections occur on the node. The connections to the other nodes are not affected. We recommend that you delete the node during off-peak hours and make sure that your applications can automatically reconnect to the cluster.
  • We recommend that you use the default cluster endpoint to connect an application to your cluster. This way, the cluster endpoint is automatically adjusted after you add or delete a read-only node. In this case, you do not need to modify the application settings. After you add a read-only node, the new node is automatically identified and service traffic is automatically distributed to the new node for load balancing. After you delete a read-only node, the unavailable node is automatically filtered.