Help Center/ Relational Database Service_RDS for PostgreSQL/ User Guide/ Instance Modifications/ Configuring Auto Scaling for Read Replicas
Updated on 2026-10-08 GMT+08:00
View PDF
Share

Configuring Auto Scaling for Read Replicas

Scenarios

RDS for PostgreSQL supports auto scaling for read replicas. When the primary instance's CPU usage exceeds the limit for read replica scale-out, RDS will automatically add read replicas. When it falls below the limit for scale-in, read replicas will be automatically removed. Auto scaling for read replicas can handle changes in database load, helping you optimize resource costs and utilization.

Permissions

  • To use this feature, submit a service ticket to request required permissions.
  • Permissions are required for configuring this feature. For details, see Permissions.
  • To use this feature, the RDSAccessProjectResource agency is required and IAM actions (for example, creating and deleting read replicas) must be added. For details, see Permissions.

Constraints

Table 1 Constraints

Phase

Constraints

Before auto scaling
  • Only the primary/standby instance type supports this feature. The primary instance must have at least one read replica.
  • Only the cloud SSD and extreme SSD storage types support this feature.
  • Before using this feature, enable the database proxy service and ensure that there is only one proxy instance, the proxy instance is working properly, and Associate New Nodes is enabled for it.
  • When removing read replicas, the system also closes existing connections that may be still processing. To prevent request failures, you are advised to implement a retry mechanism on the application side.
  • To avoid unnecessary read replica creation and removal, set longer observation and silent periods and make sure there is a significant difference between your scale-out and scale-in thresholds.
  • Auto scaling for read replicas cannot be triggered in any of the following scenarios:
    • A read replica is in the Offloading read workloads to standby node state.
    • The instance is abnormal, stopped, or frozen.
    • If the total number of read replicas (including both manually and automatically created read replicas) has reached the maximum allowed, no more read replicas can be automatically added. During auto scale-in for read replicas, the system will retain at least one of the automatically created read replicas.
    • The instance is a yearly/monthly instance that has unfinished orders or is in arrears.
    • The account quota is insufficient, the account status is abnormal, or the IP addresses of the subnet where the primary instance resides are insufficient.

Auto scaling in progress
  • During scale-out, the system adds read replicas based on the specifications you selected in the auto scaling policy.
  • During scale-in, the system removes only healthy, automatically created, pay-per-use read replicas and keeps at least one of them. Manually created read replicas are not removed.

After auto scaling
  • After auto scaling for read replicas is configured, certain operations are unavailable. To perform any of the following operations, disable auto scaling for read replicas first:
    • Creating a proxy instance
    • Deleting the proxy instance
    • Disabling Associate New Nodes for the proxy instance

Billing

Auto scaling for read replicas is billed based on the cost of the automatically added read replicas. For details, see RDS Pricing Details.

Configuring Auto Scaling for Read Replicas

  1. Log in to the RDS console.
  2. Click in the upper left corner and select a region.
  3. On the Instances page, click the name of the target primary/standby instance to go to the Summary page.
  4. Find Auto Scaling for Read Replicas and click Configure.

    Figure 1 Configuring auto scaling for read replicas

  5. In the displayed dialog box, configure the required parameters.

    Figure 2 Configuring the parameters
    Table 2 Parameters

    Parameter

    Description

    Auto Scale-Out for Read Replicas

    It controls whether to enable auto scale-out for read replicas.

    Observation Period
    • If auto scale-out for read replicas is enabled and the average CPU usage of the primary instance reaches the specified threshold within the observation period, the system automatically adds read replicas. The system then enters a silent period after each successful change.
    • The observation period ranges from 5 minutes to 30 minutes.

    Max. Read Replicas

    It specifies the maximum number of read replicas the system can add. If the total number of manually and automatically created read replicas reaches five, auto scale-out will not be triggered.

    Trigger If Avg. vCPU Usage Exceeds

    If the average vCPU usage of the primary instance exceeds this threshold during the observation period, the system automatically adds read replicas.

    Specifications of New Read Replicas

    Specifications of read replicas the system will automatically add.

    Auto Scale-In for Read Replicas

    Enable it if needed.

    Min. Read Replicas

    Minimum number of automatically created read replicas the system can keep. Manually created read replicas are not included.

    Trigger If Avg. vCPU Usage Drops To

    If auto scale-in for read replicas is enabled and the average CPU usage of the primary instance drops below the specified threshold within the observation period, the system removes read replicas. The system then enters a silent period after each successful change.

    Silent Period

    It is the minimum interval between two changes (triggered automatically or manually). During a silent period, no more auto scaling tasks can be triggered.

    NOTICE:

    The silent period for CPU and memory auto scaling is independent of that for read replica auto scaling. If both features are enabled and neither is in the silent period, the CPU and memory auto scaling task takes precedence. If both are in the silent period, the task whose silent period ends first is executed first.

  6. Click OK.

Checking the Change History

  1. On the Instances page, click the name of the target primary/standby instance to go to the Summary page.
  2. Find Auto Scaling for Read Replicas and click Change History.
  3. In the displayed dialog box, check the change time, change type, status, original specifications, and new specifications of the historical tasks.
Parent Topic: Instance Modifications