Custom Termination Policy w/CDK

Create a custom termination policy using the AWS CDK.

· #cdk, #typescript

One of the features AWS offers for Auto Scaling Groups is the ability to create a Custom Termination Policy, by providing a Lambda Function that will be called when your ASG wants to scale in.

Unfortunately, the CDK still doesn't support custom termination policies (as of June 2022), but you can do it yourself by using an escape hatch and manually adding the appropriate permissions.

import { Stack, StackProps } from 'aws-cdk-lib';
import { Construct } from 'constructs';
import * as iam from 'aws-cdk-lib/aws-iam';
import * as ec2 from 'aws-cdk-lib/aws-ec2';
import * as autoscaling from 'aws-cdk-lib/aws-autoscaling';
import * as lambda from 'aws-cdk-lib/aws-lambda';

export class ExampleStack extends Stack {
constructor(scope: Construct, id: string, props: StackProps) {
super(scope, id, props);

// Create an Auto Scaling Group
const asg = new autoscaling.AutoScalingGroup(this, 'ASG', {
vpc: vpc, /* assume some pre-existing VPC */
instanceType: new ec2.InstanceType('c5.xlarge'),
machineImage: ec2.MachineImage.genericLinux({
'us-east-1': 'ami-0b0ea68c435eb488d',
'us-west-2': 'ami-0688ba7eeeeefe3cd'
minCapacity: 1,
maxCapacity: 25,
vpcSubnets: {
subnetType: ec2.SubnetType.PRIVATE_WITH_NAT

// Create a lambda function
const terminationFunction = new lambda.Function(this, 'TerminationFunction', {
runtime: lambda.Runtime.NODEJS_16_X,
handler: 'AgentScaleInFunction.handler',
code: lambda.Code.fromAsset('dist')

// Allow the function to be invoked by the ASG principal
terminationFunction.addPermission('AllowInvokeByAutoScaling', {
principal: new iam.ArnPrincipal(

// Use escape hatch to get the Cfn object for the ASG
const cfnAsg = asg.node.defaultChild as autoscaling.CfnAutoScalingGroup;

// Assign our lambda function as the termination policy
cfnAsg.terminationPolicies = [terminationFunction.functionArn];

For the termination function itself, that's up to you -- keep in mind that the ASG will only wait 2 seconds for a response from the function, so your decision process can't be too complex (or if it is complex, you'll need to do it elsewhere, perhaps in a separate lambda that tags your instances so you know which ones are safe to terminate).

Here's an example of a simple termination function that just returns the passed instances.

import { AutoScalingScaleInHandler, AutoScalingScaleInEvent, Context } from 'aws-lambda';

export const handler: AutoScalingScaleInHandler = async (
event: AutoScalingScaleInEvent,
context: Context

) => {
const instanceIdsToTerminate = => instance.InstanceId);

// Insert your termination logic here.

return {
InstanceIDs: instanceIdsToTerminate