Tutorial: Configure auto scaling to handle a heavy workload - Application Auto Scaling
PrerequisitesStep 1: Register your scalable targetStep 2: Set up scheduled actions according to your requirementsStep 3: Add a target tracking scaling policyStep 4: Next stepsStep 5: Clean up

Tutorial: Configure auto scaling to handle a heavy workload

Important

Before you explore this tutorial, we recommend that you first review the following introductory tutorial: Tutorial: Get started with scheduled scaling using the AWS CLI.

In this tutorial, you learn how to scale out and in based on time windows when your application will have a heavier than normal workload. This is helpful when you have an application that can suddenly have a large number of visitors on a regular schedule or on a seasonal basis.

You can use a target tracking scaling policy together with scheduled scaling to handle the extra load. Scheduled scaling automatically initiates changes to your MinCapacity and MaxCapacity on your behalf, based on a schedule that you specify. When a target tracking scaling policy is active on the resource, it can scale dynamically based on current resource utilization, within the new minimum and maximum capacity range.

After completing this tutorial, you’ll know how to:

  • Use scheduled scaling to add extra capacity to meet a heavy load before it arrives, and then remove the extra capacity when it's no longer required.

  • Use a target tracking scaling policy to scale your application based on current resource utilization.

Prerequisites

This tutorial assumes that you have already done the following:

Note

While completing this tutorial, there are two steps in which you set your resource's minimum and maximum capacity values to 0 to reset the current capacity to 0. Depending on which resource you're using with Application Auto Scaling, you might be unable to reset the current capacity to 0 during these steps. To help you address the issue, a message in the output will indicate that minimum capacity cannot be less than the value specified and will provide the minimum capacity value that the AWS resource can accept.

Step 1: Register your scalable target

Start by registering your resource as a scalable target with Application Auto Scaling. A scalable target is a resource that Application Auto Scaling can scale out and scale in.

To register your scalable target with Application Auto Scaling

  • Use the following register-scalable-target command to register a new scalable target. Set the --min-capacity and --max-capacity values to 0 to reset the current capacity to 0.

    Replace the sample text for --service-namespace with the namespace of the AWS service you're using with Application Auto Scaling, --scalable-dimension with the scalable dimension associated with the resource you're registering, and --resource-id with an identifier for the resource. These values vary based on which resource is used and how the resource ID is constructed. See the topics in the AWS services that you can use with Application Auto Scaling section for more information. These topics include example commands that show you how to register scalable targets with Application Auto Scaling.

    Linux, macOS, or Unix

    aws application-autoscaling register-scalable-target \
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier \
  --min-capacity 0 --max-capacity 0

    Windows

    aws application-autoscaling register-scalable-target --service-namespace namespace --scalable-dimension dimension --resource-id identifier --min-capacity 0 --max-capacity 0

    If successful, this command returns the ARN of the scalable target.

    {
    "ScalableTargetARN": "arn:aws:application-autoscaling:region:account-id:scalable-target/1234abcd56ab78cd901ef1234567890ab123"
}

Step 2: Set up scheduled actions according to your requirements

You can use the put-scheduled-action command to create scheduled actions that are configured to meet your business needs. In this tutorial, we focus on a configuration that stops consuming resources outside of working hours by reducing capacity to 0.

To create a scheduled action that scales out in the morning

  1. To scale out the scalable target, use the following put-scheduled-action command. Include the --schedule parameter with a recurring schedule, in UTC, using a cron expression.

    On the specified schedule (every day at 9:00 AM UTC), Application Auto Scaling updates the MinCapacity and MaxCapacity values to the desired range of 1-5 capacity units.

    Linux, macOS, or Unix

    aws application-autoscaling put-scheduled-action \
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier \
  --scheduled-action-name my-first-scheduled-action \
  --schedule "cron(0 9 * * ? *)" \
  --scalable-target-action MinCapacity=1,MaxCapacity=5

    Windows

    aws application-autoscaling put-scheduled-action --service-namespace namespace --scalable-dimension dimension --resource-id identifier --scheduled-action-name my-first-scheduled-action --schedule "cron(0 9 * * ? *)" --scalable-target-action MinCapacity=1,MaxCapacity=5

    This command does not return any output if it is successful.

  2. To confirm that your scheduled action exists, use the following describe-scheduled-actions command.

    Linux, macOS, or Unix

    aws application-autoscaling describe-scheduled-actions \
  --service-namespace namespace \ 
  --query 'ScheduledActions[?ResourceId==`identifier`]'

    Windows

    aws application-autoscaling describe-scheduled-actions --service-namespace namespace --query "ScheduledActions[?ResourceId==`identifier`]"

    The following is example output.

    [
    {
        "ScheduledActionName": "my-first-scheduled-action",
        "ScheduledActionARN": "arn",
        "Schedule": "cron(0 9 * * ? *)",
        "ScalableTargetAction": {
            "MinCapacity": 1,
            "MaxCapacity": 5
        },
        ...
    }
]
To create a scheduled action that scales in at night

  1. Repeat the preceding procedure to create another scheduled action that Application Auto Scaling uses to scale in at the end of the day.

    On the specified schedule (every day at 8:00 PM UTC), Application Auto Scaling updates the target's MinCapacity and MaxCapacity to 0, as instructed by the following put-scheduled-action command.

    Linux, macOS, or Unix

    aws application-autoscaling put-scheduled-action \
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier \
  --scheduled-action-name my-second-scheduled-action \
  --schedule "cron(0 20 * * ? *)" \
  --scalable-target-action MinCapacity=0,MaxCapacity=0

    Windows

    aws application-autoscaling put-scheduled-action --service-namespace namespace --scalable-dimension dimension --resource-id identifier --scheduled-action-name my-second-scheduled-action --schedule "cron(0 20 * * ? *)" --scalable-target-action MinCapacity=0,MaxCapacity=0

  2. To confirm that your scheduled action exists, use the following describe-scheduled-actions command.

    Linux, macOS, or Unix

    aws application-autoscaling describe-scheduled-actions \
  --service-namespace namespace \ 
  --query 'ScheduledActions[?ResourceId==`identifier`]'

    Windows

    aws application-autoscaling describe-scheduled-actions --service-namespace namespace --query "ScheduledActions[?ResourceId==`identifier`]"

    The following is example output.

    [
    {
        "ScheduledActionName": "my-first-scheduled-action",
        "ScheduledActionARN": "arn",
        "Schedule": "cron(0 9 * * ? *)",
        "ScalableTargetAction": {
            "MinCapacity": 1,
            "MaxCapacity": 5
        },
        ...
    },
    {
        "ScheduledActionName": "my-second-scheduled-action",
        "ScheduledActionARN": "arn",
        "Schedule": "cron(0 20 * * ? *)",
        "ScalableTargetAction": {
            "MinCapacity": 0,
            "MaxCapacity": 0
        },
        ...
    }
]

Step 3: Add a target tracking scaling policy

Now that you have the basic schedule in place, add a target tracking scaling policy to scale based on current resource utilization.

With target tracking, Application Auto Scaling compares the target value in the policy to the current value of the specified metric. When they are unequal for a period of time, Application Auto Scaling adds or removes capacity to maintain steady performance. As the load on your application and the metric value increases, Application Auto Scaling adds capacity as fast as it can without going above MaxCapacity. When Application Auto Scaling removes capacity because the load is minimal, it does so without going below MinCapacity. By adjusting the capacity based on usage, you only pay for what your application needs.

If the metric has insufficient data because your application does not have any load, Application Auto Scaling does not add or remove capacity. In other words, Application Auto Scaling prioritizes availability in situations where not enough information is available.

You can add multiple scaling policies, but make sure you do not add conflicting step scaling policies, which might cause undesirable behavior. For example, if the step scaling policy initiates a scale-in activity before the target tracking policy is ready to scale in, the scale-in activity will not be blocked. After the scale-in activity completes, the target tracking policy could instruct Application Auto Scaling to scale out again.

To create a target tracking scaling policy

  1. Use the following put-scaling-policy command to create the policy.

    The metrics that are most frequently used for target tracking are predefined, and you can use them without supplying the full metric specification from CloudWatch. For more information about the available predefined metrics, see Target tracking scaling policies.

    Before you run this command, make sure that your predefined metric expects the target value. For example, to scale out when CPU reaches 50% utilization, specify a target value of 50.0. Or, to scale out Lambda provisioned concurrency when usage reaches 70% utilization, specify a target value of 0.7. For information about target values for a particular resource, refer to the documentation that is provided by the service about how to configure target tracking. For more information, see AWS services that you can use with Application Auto Scaling.

    Linux, macOS, or Unix

    aws application-autoscaling put-scaling-policy \
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier \
  --policy-name my-scaling-policy --policy-type TargetTrackingScaling \
  --target-tracking-scaling-policy-configuration '{ "TargetValue": 50.0, "PredefinedMetricSpecification": { "PredefinedMetricType": "predefinedmetric" }}'

    Windows

    aws application-autoscaling put-scaling-policy --service-namespace namespace --scalable-dimension dimension --resource-id identifier --policy-name my-scaling-policy --policy-type TargetTrackingScaling --target-tracking-scaling-policy-configuration "{ \"TargetValue\": 50.0, \"PredefinedMetricSpecification\": { \"PredefinedMetricType\": \"predefinedmetric\" }}"

    If successful, this command returns the ARNs and names of the two CloudWatch alarms that were created on your behalf.

  2. To confirm that your scheduled action exists, use the following describe-scaling-policies command.

    Linux, macOS, or Unix

    aws application-autoscaling describe-scaling-policies --service-namespace namespace \
  --query 'ScalingPolicies[?ResourceId==`identifier`]'

    Windows

    aws application-autoscaling describe-scaling-policies --service-namespace namespace --query "ScalingPolicies[?ResourceId==`identifier`]"

    The following is example output.

    [
    {
        "PolicyARN": "arn",
        "TargetTrackingScalingPolicyConfiguration": {
            "PredefinedMetricSpecification": {
                "PredefinedMetricType": "predefinedmetric"
            },
            "TargetValue": 50.0
        },
        "PolicyName": "my-scaling-policy",
        "PolicyType": "TargetTrackingScaling",
        "Alarms": [],
        ...
    }
]

Step 4: Next steps

When a scaling activity occurs, you see a record of it in the output of the scaling activities for the scalable target, for example: 

Successfully set desired count to 1. Change successfully fulfilled by ecs.

To monitor your scaling activities with Application Auto Scaling, you can use the following describe-scaling-activities command.

Linux, macOS, or Unix

aws application-autoscaling describe-scaling-activities 
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier

Windows

aws application-autoscaling describe-scaling-activities --service-namespace namespace --scalable-dimension dimension --resource-id identifier

Step 5: Clean up

To prevent your account from accruing charges for resources created while actively scaling, you can clean up the associated scaling configuration as follows.

Deleting the scaling configuration does not delete the underlying AWS resource. It also does not return it to its original capacity. You can use the console of the service where you created the resource to delete it or adjust its capacity.

To delete the scheduled actions

The following delete-scheduled-action command deletes a specified scheduled action. You can skip this step if you want to keep the scheduled actions that you created.

Linux, macOS, or Unix

aws application-autoscaling delete-scheduled-action \
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier \
  --scheduled-action-name my-second-scheduled-action

Windows

aws application-autoscaling delete-scheduled-action --service-namespace namespace --scalable-dimension dimension --resource-id identifier --scheduled-action-name my-second-scheduled-action
To delete the scaling policy

The following delete-scaling-policy command deletes a specified target tracking scaling policy. You can skip this step if you want to keep the scaling policy that you created.

Linux, macOS, or Unix

aws application-autoscaling delete-scaling-policy \
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier \
  --policy-name my-scaling-policy

Windows

aws application-autoscaling delete-scaling-policy --service-namespace namespace --scalable-dimension dimension --resource-id identifier --policy-name my-scaling-policy
To deregister the scalable target

Use the following deregister-scalable-target command to deregister the scalable target. If you have any scaling policies that you created or any scheduled actions that have not yet been deleted, they are deleted by this command. You can skip this step if you want to keep the scalable target registered for future use.

Linux, macOS, or Unix

aws application-autoscaling deregister-scalable-target \
  --service-namespace namespace \
  --scalable-dimension dimension \
  --resource-id identifier

Windows

aws application-autoscaling deregister-scalable-target --service-namespace namespace --scalable-dimension dimension --resource-id identifier