Help Center/ ModelArts/ Model Inference/ Inference Deployment (New Version)/ Managing Real-Time Services/ Scaling a Real-Time Service Deployment
Updated on 2026-09-08 GMT+08:00
View PDF
Share

Scaling a Real-Time Service Deployment

Once a real-time service is running, its resource needs might shift as services evolve. ModelArts allows you to deploy and scale these services dynamically. You can modify instance counts or resource settings to match current demand.

ModelArts provides manual scaling and auto scaling to meet different user requirements. Only the number of instances of a single deployment can be changed.

  • Manual scaling allows you to manually change the number of instances of a single deployment.
  • Auto scaling allows you to configure scaling rules to adjust instance count. This helps you use your resources more efficiently. Auto scaling allows you to configure scaling policies to add instances when the traffic is high, and reduce them when the traffic is low. This helps you use your resources more efficiently. Currently, only deployments on dedicated resource pool support auto scaling.
    Table 1 Comparison between manual and auto scaling

    Feature

    Manual Scaling

    Auto Scaling

    Configuration Method

    Manual

    Automatic

    Operation

    Modify the number of instances

    Set scaling rules

    Execution

    Executed after manual configuration

    Triggered by schedule or metrics

    Impact of Scaling Failure

    The number of instances reverts to the previous value.

    The number of instances changes to a specific value.

Constraints

  • To set scaling rules, the service must be in the Stopped, Running, or Alarm state.
  • Real-time services allow up to 10 periodic-based scaling rules. You can add only one scaling rule per metric.

Manual Scaling

  1. Log in to the ModelArts console and choose Model Inference > Real-Time Inference.
  2. Click the target service name. On the service details page that is displayed, switch to the Deploy tab.
  3. Locate the target deployment card and choose ... > Scale.
  4. Click Manual Scaling. In the displayed dialog box, set the number of deployment replicas after scaling. The minimum value is 1. Click OK.

    Check the instance list to view their status after manual scaling.

    To delete a specified instance, click Delete in the Operation column in the deployment replica list. Choose how you want to delete it, enter DELETE, and click OK. Currently, Delete and re-create is supported.

    Deleting a replica reduces the total count permanently. Exercise caution when performing this operation.

Auto Scaling

  1. Log in to the ModelArts console and choose Model Inference > Real-Time Inference.
  2. Click the target service name. On the service details page that is displayed, switch to the Deploy tab.
  3. Locate the target deployment card and choose ... > Scale.
  4. In the Auto Scaling Rule area, click Configure Scaling Rule. In the displayed dialog box, set automatic scaling rules.

    Real-time services allow up to 10 periodic-based scaling rules. You can add only one scaling rule per metric.

    A periodic-based scaling rule cannot be added repeatedly. For details, see Table 2.

    Table 2 Scaling rule parameters

    Scaling Rule Type

    Description

    Periodic

    Automatically adjusts the number of nodes in the pool within a set timeframe, improving resource efficiency and lowering expenses.

    • Trigger Time: Specify a time as required. This indicates the local time where the nodes are deployed.
    • Target Instances: Target number of nodes in the node pool after auto scaling.

  5. Click OK.

    Once you add a rule, activate it in the Scaling Rules section. When enabled, the system adjusts your service deployment size automatically at the set time. Check the scaling history at the bottom of the page.

    Click Delete in the Operation column to delete the created scaling rule.

Parent Topic: Managing Real-Time Services