All Products
Search
Document Center

PolarDB:Cluster node performance comparison before and after hot replica is enabled

Last Updated:Oct 19, 2023

This topic describes the performance differences before and after hot replica is enabled.

Comparison overview

The following table lists the performance differences between hot replica nodes for which the failover with hot replica feature is enabled and read-only nodes.

Item

Hot replica node

Read-only node

Service capabilities

Read-only services are provided. Prefetching consumes a small amount of memory.

Read-only services are provided.

Switchover (active O&M, such as minor version upgrade and cluster configuration changes)

The number of transactions per second (TPS) drops to zero for about 5 seconds.

The number of TPS drops to zero for more than 10 seconds.

Failover (disaster recovery, such as recovering the primary node)

The number of TPS drops to zero for about 5 seconds.

Service interruption lasts about 60 seconds.

The preceding table indicates that:

  • In active O&M scenarios, read-only nodes for which the failover with hot replica feature is enabled have shorter service interruption time and enable always-on connections and transactions.

  • In disaster recovery scenarios, read-only nodes with hot replica enabled have shorter service interruption time. In addition, services and transactions are not interrupted, and the number of errors reported by clients is significantly reduced.

Detailed information

The following section provides detailed information about failover speed before and after hot replica is enabled:

  • Switchover (active O&M, such as minor version upgrade and cluster configuration changes)

    • Normal read-only nodes (when hot replica and transaction status preservation are disabled): over 10 seconds before TPS drops to zero普通ro-主动运维

    • Hot replica nodes (when hot replica and transaction status preservation are enabled): 5 seconds before TPS drops to zero热备-主动运维

  • Failover (disaster recovery, such as recovering the primary node)

    • Normal read-only nodes (when hot replica and transaction status preservation are disabled): 60 seconds of service interruption普通RO-容灾

    • Hot replica nodes (when hot replica and transaction status preservation are enabled): 5 seconds before TPS drops to zero热备RO-容灾