Oven logo

Oven

Amazon EC2 Auto Scaling Construct Library

---

End-of-Support

AWS CDK v1 has reached End-of-Support on 2023-06-01. This package is no longer being updated, and users should migrate to AWS CDK v2.

For more information on how to migrate, see the Migrating to AWS CDK v2 guide.


This module is part of the AWS Cloud Development Kit project.

Auto Scaling Group

An AutoScalingGroup represents a number of instances on which you run your code. You pick the size of the fleet, the instance type and the OS image:

# vpc: ec2.Vpc


autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=ec2.InstanceType.of(ec2.InstanceClass.BURSTABLE2, ec2.InstanceSize.MICRO),
    machine_image=ec2.AmazonLinuxImage()
)

NOTE: AutoScalingGroup has an property called allowAllOutbound (allowing the instances to contact the internet) which is set to true by default. Be sure to set this to false if you don't want your instances to be able to start arbitrary connections. Alternatively, you can specify an existing security group to attach to the instances that are launched, rather than have the group create a new one.

# vpc: ec2.Vpc


my_security_group = ec2.SecurityGroup(self, "SecurityGroup", vpc=vpc)
autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=ec2.InstanceType.of(ec2.InstanceClass.BURSTABLE2, ec2.InstanceSize.MICRO),
    machine_image=ec2.AmazonLinuxImage(),
    security_group=my_security_group
)

Alternatively you can create an AutoScalingGroup from a LaunchTemplate:

# vpc: ec2.Vpc
# launch_template: ec2.LaunchTemplate


autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    launch_template=launch_template
)

To launch a mixture of Spot and on-demand instances, and/or with multiple instance types, you can create an AutoScalingGroup from a MixedInstancesPolicy:

# vpc: ec2.Vpc
# launch_template1: ec2.LaunchTemplate
# launch_template2: ec2.LaunchTemplate


autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    mixed_instances_policy=autoscaling.MixedInstancesPolicy(
        instances_distribution=autoscaling.InstancesDistribution(
            on_demand_percentage_above_base_capacity=50
        ),
        launch_template=launch_template1,
        launch_template_overrides=[autoscaling.LaunchTemplateOverrides(instance_type=ec2.InstanceType("t3.micro")), autoscaling.LaunchTemplateOverrides(instance_type=ec2.InstanceType("t3a.micro")), autoscaling.LaunchTemplateOverrides(instance_type=ec2.InstanceType("t4g.micro"), launch_template=launch_template2)]
    )
)

Machine Images (AMIs)

AMIs control the OS that gets launched when you start your EC2 instance. The EC2 library contains constructs to select the AMI you want to use.

Depending on the type of AMI, you select it a different way.

The latest version of Amazon Linux and Microsoft Windows images are selectable by instantiating one of these classes:

# Pick a Windows edition to use
windows = ec2.WindowsImage(ec2.WindowsVersion.WINDOWS_SERVER_2019_ENGLISH_FULL_BASE)

# Pick the right Amazon Linux edition. All arguments shown are optional
# and will default to these values when omitted.
amzn_linux = ec2.AmazonLinuxImage(
    generation=ec2.AmazonLinuxGeneration.AMAZON_LINUX,
    edition=ec2.AmazonLinuxEdition.STANDARD,
    virtualization=ec2.AmazonLinuxVirt.HVM,
    storage=ec2.AmazonLinuxStorage.GENERAL_PURPOSE
)

# For other custom (Linux) images, instantiate a `GenericLinuxImage` with
# a map giving the AMI to in for each region:

linux = ec2.GenericLinuxImage({
    "us-east-1": "ami-97785bed",
    "eu-west-1": "ami-12345678"
})

NOTE: The Amazon Linux images selected will be cached in your cdk.json, so that your AutoScalingGroups don't automatically change out from under you when you're making unrelated changes. To update to the latest version of Amazon Linux, remove the cache entry from the context section of your cdk.json.

We will add command-line options to make this step easier in the future.

AutoScaling Instance Counts

AutoScalingGroups make it possible to raise and lower the number of instances in the group, in response to (or in advance of) changes in workload.

When you create your AutoScalingGroup, you specify a minCapacity and a maxCapacity. AutoScaling policies that respond to metrics will never go higher or lower than the indicated capacity (but scheduled scaling actions might, see below).

There are three ways to scale your capacity:

  • In response to a metric (also known as step scaling); for example, you might want to scale out if the CPU usage across your cluster starts to rise, and scale in when it drops again.
  • By trying to keep a certain metric around a given value (also known as target tracking scaling); you might want to automatically scale out and in to keep your CPU usage around 50%.
  • On a schedule; you might want to organize your scaling around traffic flows you expect, by scaling out in the morning and scaling in in the evening.

The general pattern of autoscaling will look like this:

# vpc: ec2.Vpc
# instance_type: ec2.InstanceType
# machine_image: ec2.IMachineImage


auto_scaling_group = autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=instance_type,
    machine_image=machine_image,

    min_capacity=5,
    max_capacity=100
)

Step Scaling

This type of scaling scales in and out in deterministics steps that you configure, in response to metric values. For example, your scaling strategy to scale in response to a metric that represents your average worker pool usage might look like this:

 Scaling        -1          (no change)          +1       +3
            │        │                       │        │        │
            ├────────┼───────────────────────┼────────┼────────┤
            │        │                       │        │        │
Worker use  0%      10%                     50%       70%     100%

(Note that this is not necessarily a recommended scaling strategy, but it's a possible one. You will have to determine what thresholds are right for you).

Note that in order to set up this scaling strategy, you will have to emit a metric representing your worker utilization from your instances. After that, you would configure the scaling something like this:

# auto_scaling_group: autoscaling.AutoScalingGroup


worker_utilization_metric = cloudwatch.Metric(
    namespace="MyService",
    metric_name="WorkerUtilization"
)

auto_scaling_group.scale_on_metric("ScaleToCPU",
    metric=worker_utilization_metric,
    scaling_steps=[autoscaling.ScalingInterval(upper=10, change=-1), autoscaling.ScalingInterval(lower=50, change=+1), autoscaling.ScalingInterval(lower=70, change=+3)
    ],

    # Change this to AdjustmentType.PERCENT_CHANGE_IN_CAPACITY to interpret the
    # 'change' numbers before as percentages instead of capacity counts.
    adjustment_type=autoscaling.AdjustmentType.CHANGE_IN_CAPACITY
)

The AutoScaling construct library will create the required CloudWatch alarms and AutoScaling policies for you.

Target Tracking Scaling

This type of scaling scales in and out in order to keep a metric around a value you prefer. There are four types of predefined metrics you can track, or you can choose to track a custom metric. If you do choose to track a custom metric, be aware that the metric has to represent instance utilization in some way (AutoScaling will scale out if the metric is higher than the target, and scale in if the metric is lower than the target).

If you configure multiple target tracking policies, AutoScaling will use the one that yields the highest capacity.

The following example scales to keep the CPU usage of your instances around 50% utilization:

# auto_scaling_group: autoscaling.AutoScalingGroup


auto_scaling_group.scale_on_cpu_utilization("KeepSpareCPU",
    target_utilization_percent=50
)

To scale on average network traffic in and out of your instances:

# auto_scaling_group: autoscaling.AutoScalingGroup


auto_scaling_group.scale_on_incoming_bytes("LimitIngressPerInstance",
    target_bytes_per_second=10 * 1024 * 1024
)
auto_scaling_group.scale_on_outgoing_bytes("LimitEgressPerInstance",
    target_bytes_per_second=10 * 1024 * 1024
)

To scale on the average request count per instance (only works for AutoScalingGroups that have been attached to Application Load Balancers):

# auto_scaling_group: autoscaling.AutoScalingGroup


auto_scaling_group.scale_on_request_count("LimitRPS",
    target_requests_per_second=1000
)

Scheduled Scaling

This type of scaling is used to change capacities based on time. It works by changing minCapacity, maxCapacity and desiredCapacity of the AutoScalingGroup, and so can be used for two purposes:

  • Scale in and out on a schedule by setting the minCapacity high or the maxCapacity low.
  • Still allow the regular scaling actions to do their job, but restrict the range they can scale over (by setting both minCapacity and maxCapacity but changing their range over time).

A schedule is expressed as a cron expression. The Schedule class has a cron method to help build cron expressions.

The following example scales the fleet out in the morning, going back to natural scaling (all the way down to 1 instance if necessary) at night:

# auto_scaling_group: autoscaling.AutoScalingGroup


auto_scaling_group.scale_on_schedule("PrescaleInTheMorning",
    schedule=autoscaling.Schedule.cron(hour="8", minute="0"),
    min_capacity=20
)

auto_scaling_group.scale_on_schedule("AllowDownscalingAtNight",
    schedule=autoscaling.Schedule.cron(hour="20", minute="0"),
    min_capacity=1
)

Configuring Instances using CloudFormation Init

It is possible to use the CloudFormation Init mechanism to configure the instances in the AutoScalingGroup. You can write files to it, run commands, start services, etc. See the documentation of AWS::CloudFormation::Init and the documentation of CDK's aws-ec2 library for more information.

When you specify a CloudFormation Init configuration for an AutoScalingGroup:

  • you must also specify signals to configure how long CloudFormation should wait for the instances to successfully configure themselves.
  • you should also specify an updatePolicy to configure how instances should be updated when the AutoScalingGroup is updated (for example, when the AMI is updated). If you don't specify an update policy, a rolling update is chosen by default.

Here's an example of using CloudFormation Init to write a file to the instance hosts on startup:

# vpc: ec2.Vpc
# instance_type: ec2.InstanceType
# machine_image: ec2.IMachineImage


autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=instance_type,
    machine_image=machine_image,

    # ...

    init=ec2.CloudFormationInit.from_elements(
        ec2.InitFile.from_string("/etc/my_instance", "This got written during instance startup")),
    signals=autoscaling.Signals.wait_for_all(
        timeout=Duration.minutes(10)
    )
)

Signals

In normal operation, CloudFormation will send a Create or Update command to an AutoScalingGroup and proceed with the rest of the deployment without waiting for the instances in the AutoScalingGroup.

Configure signals to tell CloudFormation to wait for a specific number of instances in the AutoScalingGroup to have been started (or failed to start) before moving on. An instance is supposed to execute the cfn-signal program as part of its startup to indicate whether it was started successfully or not.

If you use CloudFormation Init support (described in the previous section), the appropriate call to cfn-signal is automatically added to the AutoScalingGroup's UserData. If you don't use the signals directly, you are responsible for adding such a call yourself.

The following type of Signals are available:

  • Signals.waitForAll([options]): wait for all of desiredCapacity amount of instances to have started (recommended).
  • Signals.waitForMinCapacity([options]): wait for a minCapacity amount of instances to have started (use this if waiting for all instances takes too long and you are happy with a minimum count of healthy hosts).
  • Signals.waitForCount(count, [options]): wait for a specific amount of instances to have started.

There are two options you can configure:

  • timeout: maximum time a host startup is allowed to take. If a host does not report success within this time, it is considered a failure. Default is 5 minutes.
  • minSuccessPercentage: percentage of hosts that needs to be healthy in order for the update to succeed. If you set this value lower than 100, some percentage of hosts may report failure, while still considering the deployment a success. Default is 100%.

Update Policy

The update policy describes what should happen to running instances when the definition of the AutoScalingGroup is changed. For example, if you add a command to the UserData of an AutoScalingGroup, do the existing instances get replaced with new instances that have executed the new UserData? Or do the "old" instances just keep on running?

It is recommended to always use an update policy, otherwise the current state of your instances also depends the previous state of your instances, rather than just on your source code. This degrades the reproducibility of your deployments.

The following update policies are available:

  • UpdatePolicy.none(): leave existing instances alone (not recommended).
  • UpdatePolicy.rollingUpdate([options]): progressively replace the existing instances with new instances, in small batches. At any point in time, roughly the same amount of total instances will be running. If the deployment needs to be rolled back, the fresh instances will be replaced with the "old" configuration again.
  • UpdatePolicy.replacingUpdate([options]): build a completely fresh copy of the new AutoScalingGroup next to the old one. Once the AutoScalingGroup has been successfully created (and the instances started, if signals is configured on the AutoScalingGroup), the old AutoScalingGroup is deleted. If the deployment needs to be rolled back, the new AutoScalingGroup is deleted and the old one is left unchanged.

Allowing Connections

See the documentation of the @aws-cdk/aws-ec2 package for more information about allowing connections between resources backed by instances.

Max Instance Lifetime

To enable the max instance lifetime support, specify maxInstanceLifetime property for the AutoscalingGroup resource. The value must be between 7 and 365 days(inclusive). To clear a previously set value, leave this property undefined.

Instance Monitoring

To disable detailed instance monitoring, specify instanceMonitoring property for the AutoscalingGroup resource as Monitoring.BASIC. Otherwise detailed monitoring will be enabled.

Monitoring Group Metrics

Group metrics are used to monitor group level properties; they describe the group rather than any of its instances (e.g GroupMaxSize, the group maximum size). To enable group metrics monitoring, use the groupMetrics property. All group metrics are reported in a granularity of 1 minute at no additional charge.

See EC2 docs for a list of all available group metrics.

To enable group metrics monitoring using the groupMetrics property:

# vpc: ec2.Vpc
# instance_type: ec2.InstanceType
# machine_image: ec2.IMachineImage


# Enable monitoring of all group metrics
autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=instance_type,
    machine_image=machine_image,

    # ...

    group_metrics=[autoscaling.GroupMetrics.all()]
)

# Enable monitoring for a subset of group metrics
autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=instance_type,
    machine_image=machine_image,

    # ...

    group_metrics=[autoscaling.GroupMetrics(autoscaling.GroupMetric.MIN_SIZE, autoscaling.GroupMetric.MAX_SIZE)]
)

Termination policies

Auto Scaling uses termination policies to determine which instances it terminates first during scale-in events. You can specify one or more termination policies with the terminationPolicies property:

# vpc: ec2.Vpc
# instance_type: ec2.InstanceType
# machine_image: ec2.IMachineImage


autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=instance_type,
    machine_image=machine_image,

    # ...

    termination_policies=[autoscaling.TerminationPolicy.OLDEST_INSTANCE, autoscaling.TerminationPolicy.DEFAULT
    ]
)

Protecting new instances from being terminated on scale-in

By default, Auto Scaling can terminate an instance at any time after launch when scaling in an Auto Scaling Group, subject to the group's termination policy.

However, you may wish to protect newly-launched instances from being scaled in if they are going to run critical applications that should not be prematurely terminated. EC2 Capacity Providers for Amazon ECS requires this attribute be set to true.

# vpc: ec2.Vpc
# instance_type: ec2.InstanceType
# machine_image: ec2.IMachineImage


autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=instance_type,
    machine_image=machine_image,

    # ...

    new_instances_protected_from_scale_in=True
)

Configuring Instance Metadata Service (IMDS)

Toggling IMDSv1

You can configure EC2 Instance Metadata Service options to either allow both IMDSv1 and IMDSv2 or enforce IMDSv2 when interacting with the IMDS.

To do this for a single AutoScalingGroup, you can use set the requireImdsv2 property. The example below demonstrates IMDSv2 being required on a single AutoScalingGroup:

# vpc: ec2.Vpc
# instance_type: ec2.InstanceType
# machine_image: ec2.IMachineImage


autoscaling.AutoScalingGroup(self, "ASG",
    vpc=vpc,
    instance_type=instance_type,
    machine_image=machine_image,

    # ...

    require_imdsv2=True
)

You can also use AutoScalingGroupRequireImdsv2Aspect to apply the operation to multiple AutoScalingGroups. The example below demonstrates the AutoScalingGroupRequireImdsv2Aspect being used to require IMDSv2 for all AutoScalingGroups in a stack:

aspect = autoscaling.AutoScalingGroupRequireImdsv2Aspect()

Aspects.of(self).add(aspect)

Warm Pool

Auto Scaling offers a warm pool which gives an ability to decrease latency for applications that have exceptionally long boot times. You can create a warm pool with default parameters as below:

# auto_scaling_group: autoscaling.AutoScalingGroup


auto_scaling_group.add_warm_pool()

You can also customize a warm pool by configuring parameters:

# auto_scaling_group: autoscaling.AutoScalingGroup


auto_scaling_group.add_warm_pool(
    min_size=1,
    reuse_on_scale_in=True
)

Future work

  • CloudWatch Events (impossible to add currently as the AutoScalingGroup ARN is necessary to make this rule and this cannot be accessed from CloudFormation).