Temporarily turn off scaling policies - Amazon SageMaker AI

Temporarily turn off scaling policies

After you configure auto scaling, you have the following options if you need to investigate an issue without interference from scaling policies (dynamic scaling):

  • Temporarily suspend and then resume scaling activities by calling the register-scalable-target CLI command or RegisterScalableTarget API action, specifying a Boolean value for both DynamicScalingInSuspended and DynamicScalingOutSuspended.

    The following example shows how to suspend scaling policies for a variant named my-variant, running on the my-endpoint endpoint.

    aws application-autoscaling register-scalable-target \
  --service-namespace sagemaker \
  --resource-id endpoint/my-endpoint/variant/my-variant \
  --scalable-dimension sagemaker:variant:DesiredInstanceCount \
  --suspended-state '{"DynamicScalingInSuspended":true,"DynamicScalingOutSuspended":true}'

  • Prevent specific target tracking scaling policies from scaling in your variant by disabling the policy's scale-in portion. This method prevents the scaling policy from deleting instances, while still allowing it to create them as needed.

    Temporarily disable and then enable scale-in activities by editing the policy using the put-scaling-policy CLI command or the PutScalingPolicy API action, specifying a Boolean value for DisableScaleIn.

    The following is an example of a target tracking configuration for a scaling policy that will scale out but not scale in. 

    {
    "TargetValue": 70.0,
    "PredefinedMetricSpecification":
    {
        "PredefinedMetricType": "SageMakerVariantInvocationsPerInstance"
    },
    "DisableScaleIn": true
}