1,669 reads

Creating a Instance Scheduler using AWS CDK

by Yi AiAugust 11th, 2020

Too Long; Didn't Read

The AWS CDK is a software development framework to define cloud infrastructure as code and provision it through CloudFormation. The CDK integrates fully with AWS services and allows developers to use high-level construct to define infrastructure in code. In this article we will build a CDK version of AWS EC2 Instance Scheduler solution that enables us to easily configure custom start and stop schedules for our Amazon EC2 and Amazon RDS instances. At the end of this article, we have a pipeline to deploy a serverless solution that starts and stops EC2 instances based on a schedule.

Companies Mentioned

Coins Mentioned

featured image - Creating a Instance Scheduler using AWS CDK

In this article, we will build a CDK version of AWS EC2 Instance Scheduler solution that enables us to easily configure custom start and stop schedules for our Amazon EC2 and Amazon RDS instances.

The Architecture

At the end of this article, we have a pipeline to deploy a serverless solution that start and stop Amazon EC2 instances in an autoscalling group and Amazon RDS instances based on a schedule.

Here’s the architecture diagram:

Through the article we will be creating the following resources:

Deploy an AWS Step Function with two parallel tasks.
Create an SNS topic to send notifications.
Create a Cloudeatch event rule which will trigger Step Function based on a schedule.
Create a CI/CD pipeline with CodeBuild and CodePipeline.

Prerequisites

To deploy the CDK application, there are a few prerequisites that need to be met:

Setup an AWS account.
Install the latest aws-cli.
Install AWS CDK CLI.
Deploy a multi-AZ wordpress website with RDS (optional).

Before you begin

First, create an AWS CDK project by entering the following commands at the command line.

$mkdir cdk-sample 
$cd cdk-sample cdk init --language=javascript

Next, install CDK modules, we will use below modules in our project.

$npm install @aws-cdk/core @aws-cdk/aws-codebuild @aws-cdk/aws-codepipeline @aws-cdk/aws-codepipeline-actions @aws-cdk/aws-events @aws-cdk/aws-events-targets @aws-cdk/aws-iam @aws-cdk/aws-lambda @aws-cdk/aws-sns @aws-cdk/aws-ssm @aws-cdk/aws-stepfunctions @aws-cdk/aws-stepfunctions-tasks

We need to add a stage parameter because we want to deploy our stack to multiple stages (dev and production).

Let’s add

DEP_ENV

bin/cdk-sample.js

#!/usr/bin/env node

const cdk = require('@aws-cdk/core');
const BaseStackConstruct = require('../lib/base-stack');

const deployEnv = process.env.DEPLOY_ENV || 'dev';
const app = new cdk.App();
new BaseStackConstruct(app, `CDKSample`, {
  env: { region: 'ap-southeast-2' },
  stage: deployEnv,
});

Define the base stack class

Now let’s define our base stack class. In the base stack class, we’ll add the code to instantiate three separate stacks: SNS stack, StepFunction stack and CodePipeline stack.

Make the code look like the following.

const { Construct } = require('@aws-cdk/core');
const SNSAlertStac = require('./sns-stack');
const StepfunctionStack = require('./lambda-stack');
const CodePipelineStack = require('./codepipeline-stack');
const lambda = require('@aws-cdk/aws-lambda');

module.exports = class BaseStackConstruct extends Construct {
  constructor(scope, id, props) {
    super(scope, id, props);

    //  stack for SNS
    const sNSAlertStac = new SNSAlertStac(scope, `SNSAlert-${props.stage}`);

    // Deploy step functions from codepipeline
    if (!process.env.MANUAL_DEPLOY) {
      const cfnParametersCode = lambda.Code.fromCfnParameters();

      new StepfunctionStack(scope, `Stepfunction-${props.stage}`, {
        snsTopic: sNSAlertStac.topic,
        lambdaCode: cfnParametersCode,
      });

      new CodePipelineStack(scope, `CodepipelienStack-${props.stage}`, {
        lambdaCode: cfnParametersCode,
        stage: props.stage,
      });
    } else {
      //deploy step function from local env
      new StepfunctionStack(scope, `Stepfunction-${props.stage}`, {
        snsTopic: sNSAlertStac.topic,
      });
    }
  }
};

We can set optional environment variable MANUAL_DEPLOY to true if we want to deploy only step function locally.

$export MANUAL_DEPLOY=true && cdk deploy Stepfunction-dev

Define SNS stack

We’ll add the code (lib/sns-stack.js) to create an SNS topic and subscribe an email to the created topic.

const { Stack } = require('@aws-cdk/core');
const sns = require('@aws-cdk/aws-sns');
const subs = require('@aws-cdk/aws-sns-subscriptions');

const EMAIL_SUBSCRIPTION = '[email protected]';
module.exports = class SNSAlertStac extends Stack {
  constructor(scope, id, props) {
    super(scope, id, props);
    this.topic = new sns.Topic(this, 'AlertTopic', {
      displayName: 'Step function execution failed',
    });
    this.topic.addSubscription(new subs.EmailSubscription(EMAIL_SUBSCRIPTION));
  }
};

Define Step function and Lambdas

Now we’ll expand our

lib/lambda-stack.js

file and add Lambda functions and Step function.

In this example we will create two lambda functions, updateScalingGroupFn to update auto scaling group, and updateDBClusterFn to start/stop RDS instances.

const { Stack, ScopedAws, Duration } = require('@aws-cdk/core');
const { Function, Runtime, Code } = require('@aws-cdk/aws-lambda');
const { PolicyStatement } = require('@aws-cdk/aws-iam');
const { SfnStateMachine } = require('@aws-cdk/aws-events-targets');
const { Rule, Schedule } = require('@aws-cdk/aws-events');
const ssm = require('@aws-cdk/aws-ssm');
const sfn = require('@aws-cdk/aws-stepfunctions');
const tasks = require('@aws-cdk/aws-stepfunctions-tasks');
const { join } = require('path');

module.exports = class StepfunctionStack extends Stack {
  constructor(scope, id, props) {
    super(scope, id, props);

    // retriving autoscaling group name and rds db cluster name from SSM
    const autoScalingGroupName = ssm.StringParameter.valueForStringParameter(
      this,
      '/cdksample/autoscalling/name'
    );
    const dBInstanceName = ssm.StringParameter.valueForStringParameter(
      this,
      '/cdksample/dbcluster/name'
    );

    const { accountId, region } = new ScopedAws(this);

    //creating lambda functions
    const updateScalingGroupFn = new Function(this, 'updateScalingGroupFn', {
      runtime: Runtime.NODEJS_12_X,
      handler: 'lambda_update_asg/index.handler',
      code: props.lambdaCode || Code.fromAsset(join(__dirname, '../src')),
      environment: {
        autoScalingGroupName,
      },
    });

    const updateDBClusterFn = new Function(this, 'updateDBCluster', {
      runtime: Runtime.NODEJS_12_X,
      handler: 'lambda_update_db_cluster/index.handler',
      code: props.lambdaCode || Code.fromAsset(join(__dirname, '../src')),
      environment: {
        dBInstanceName,
      },
    });

    //IAM policies for updateScalingGroupFn lambda
    const statementUpdateASLGroup = new PolicyStatement();
    statementUpdateASLGroup.addActions('autoscaling:UpdateAutoScalingGroup');
    statementUpdateASLGroup.addResources(
      `arn:aws:autoscaling:${region}:${accountId}:autoScalingGroup:*:autoScalingGroupName/${autoScalingGroupName}`
    );
    updateScalingGroupFn.addToRolePolicy(statementUpdateASLGroup);

    const statementDescribeASLGroup = new PolicyStatement();
    statementDescribeASLGroup.addActions(
      'autoscaling:DescribeAutoScalingGroups'
    );
    statementDescribeASLGroup.addResources('*');
    updateScalingGroupFn.addToRolePolicy(statementDescribeASLGroup);

    //IAM policies for updateDBCluster lambda
    const statementDescribeDBCluster = new PolicyStatement();
    statementDescribeDBCluster.addActions('rds:DescribeDBInstances');
    statementDescribeDBCluster.addResources('*');
    updateDBClusterFn.addToRolePolicy(statementDescribeDBCluster);

    const statementToggleDBCluster = new PolicyStatement();
    statementToggleDBCluster.addActions([
      'rds:StartDBInstance',
      'rds:StopDBInstance',
    ]);
    statementToggleDBCluster.addResources('*');
    updateDBClusterFn.addToRolePolicy(statementToggleDBCluster);
  }
};

Next, we will continue by adding our step function definitions. Add the following code to

lib/lambda-stack.js

const updateScalingGroupTask = new tasks.LambdaInvoke(
      this,
      'Update asg task',
      {
        lambdaFunction: updateScalingGroupFn,
      }
    );
    const updateDBClusterTask = new tasks.LambdaInvoke(
      this,
      'StopStart db cluster task',
      {
        lambdaFunction: updateDBClusterFn,
      }
    );

    const sendFailureNotification = new tasks.SnsPublish(
      this,
      'Publish alert notification',
      {
        topic: props.snsTopic,
        message: sfn.TaskInput.fromDataAt('$.error'),
      }
    );

    const stepChain = new sfn.Parallel(
      this,
      'Stop and Start EC2 Instances and RDS in parallel'
    )
      .branch(updateScalingGroupTask)
      .branch(updateDBClusterTask)
      .addCatch(sendFailureNotification);

    const toggleAWSServices = new sfn.StateMachine(this, 'StateMachine', {
      definition: stepChain,
      timeout: Duration.minutes(5),
    });

As we can see, for each Lambda there is a corresponding taskEvent. We only want to parallelize the Lamba workload (one task for the ec2 autoscaling group and another task for RDS instances). Hence we will add a chain to our workflow by calling the

 branch()

function.

To send a notification to Amazon SNS topic if the Lambda fails, we can add an error handling chain to the workflow by calling

addCatch()

function.

Finally, let’s define a CloudWatch Events rule, CloudWatch is triggering execution of state machine workflow every day 7am and 6pm (UTC).

Add the following code to

lib/lambda-stack.js

new Rule(this, 'Rule', {
  schedule: Schedule.expression('cron(0 7,18 * * ? *)'),
  targets: [new SfnStateMachine(toggleAWSServices)],
});

Define pipeline stack

We define our final stack, codepiepeline-stack. It has a source Action targeting the Github repository, a build Action that builds previously defined stacks, and finally a deploy Action that uses AWS CloudFormation. It takes the Cloudformation template (cdk.out/*.template.json) generated by the AWS CDK build action and passes it to AWS CloudFormation for deployment.

Create lib/codepipeline-stack.js and put the following code in it.

const cdk = require('@aws-cdk/core');
const codebuild = require('@aws-cdk/aws-codebuild');
const codepipeline = require('@aws-cdk/aws-codepipeline');
const codepipeline_actions = require('@aws-cdk/aws-codepipeline-actions');
const ssm = require('@aws-cdk/aws-ssm');
const packageJson = require('../package.json');

class CodePipelineStack extends cdk.Stack {
  constructor(scope, id, props) {
    super(scope, id, props);
    const sourceOutput = new codepipeline.Artifact();
    const cdkBuildOutput = new codepipeline.Artifact('CdkBuildOutput');
    const lambdaBuildOutput = new codepipeline.Artifact('LambdaBuildOutput');
    const branch = props.stage === 'dev' ? 'dev' : 'master';
    new codepipeline.Pipeline(this, 'Pipeline', {
      pipelineName: `DeployScheduleService-${props.stage}`,
      stages: [
        {
          stageName: 'Source',
          actions: [
            new codepipeline_actions.GitHubSourceAction({
              actionName: 'Code',
              output: sourceOutput,
              oauthToken: ssm.StringParameter.valueForStringParameter(
                this,
                '/github/token'
              ),
              owner: 'yai333',
              branch,
              repo: 'cdkexample',
            }),
          ],
        },
        {
          stageName: 'Build',
          actions: [
            new codepipeline_actions.CodeBuildAction({
              actionName: 'Build_CDK_LAMBDA',
              project: new codebuild.PipelineProject(this, 'Build', {
                buildSpec: codebuild.BuildSpec.fromObject({
                  version: +`${
                    packageJson.version.split('.')[0]
                  }.${+packageJson.version.split('.').slice(-2).join('')}`,
                  phases: {
                    install: {
                      commands: 'npm install',
                    },
                    build: {
                      commands: `export DEPLOY_ENV=${props.stage} && npm run cdk synth`,
                    },
                    post_build: {
                      commands: ['cd src && npm install', 'ls src -la'],
                    },
                  },
                  // save the generated files in the correct artifacts
                  artifacts: {
                    'secondary-artifacts': {
                      CdkBuildOutput: {
                        'base-directory': 'cdk.out',
                        files: ['**/*'],
                      },
                      LambdaBuildOutput: {
                        'base-directory': 'src',
                        files: ['**/*'],
                      },
                    },
                  },
                }),
              }),
              input: sourceOutput,
              outputs: [cdkBuildOutput, lambdaBuildOutput],
            }),
          ],
        },
        {
          stageName: 'Deploy',
          actions: [
            new codepipeline_actions.CloudFormationCreateUpdateStackAction({
              actionName: 'Deploy_SNS_Stack',
              templatePath: cdkBuildOutput.atPath(
                `SNSAlert-${props.stage}.template.json`
              ),
              stackName: `SNSStack-${props.stage}`,
              adminPermissions: true,
            }),
            new codepipeline_actions.CloudFormationCreateUpdateStackAction({
              actionName: 'Deploy_Lambda_Stack',
              templatePath: cdkBuildOutput.atPath(
                `Stepfunction-${props.stage}.template.json`
              ),
              stackName: `StepfunctionStack-${props.stage}`,
              adminPermissions: true,
              parameterOverrides: {
                ...props.lambdaCode.assign(lambdaBuildOutput.s3Location),
              },
              extraInputs: [lambdaBuildOutput],
              runOrder: 2,
            }),
          ],
        },
      ],
    });
  }
}

module.exports = CodePipelineStack;

Next, create dev branch and check the code into Git then push it to Github repo.

$git branch dev
$git checkout dev
$git add .
$git commit -m "xxxx"
$git push

Deploying the pipeline

Now we can deploy the pipeline with multiple Stages.

Deploy pipeline to dev stage, source action targets the Github repository dev branch.

$export DEPLOY_ENV=dev && cdk deploy CodepipelienStack-dev

Deploy pipeline to the production stage, source action targets the Github repository master branch.

$export DEPLOY_ENV=production && cdk deploy CodepipelienStack-production

After the deployment finishes, we should have a three-stage pipeline that looks like the following.

Once all stacks have deployed, we can explore it in the AWS Console and give it a try. Navigate to the Step Function in the console and click “Start execution”.

We should see it passes. Let’s check EC2 autoscaling group’s DesiredCapacity and RDS instance’s status.

$aws rds describe-db-instances
"DBInstances": [
{
"DBInstanceIdentifier": "cxxxxxx",
"DBInstanceClass": "db.t2.small",
"Engine": "mysql",
"DBInstanceStatus": "stopped"
...

$aws rds describe-db-instances
"AutoScalingGroups": [
{
"AutoScalingGroupName": "cdk-sample-WebServerGroup-xxxxxx",
"MinSize": 0,
"MaxSize": 0,
"DesiredCapacity": 0,
...

Finally, let’s check the cloudwatch event rule. We should see the cloudwatch event rule looks like the following.

$aws events list-rules
{
"Rules": [
{
"Name": "StepfunctionStack-dev-Rulexxxxxxx",
"Arn": "arn:aws:events:ap-southeast-2:xxxxxxx:rule/StepfunctionStack-dev-Rulexxxxxxx-xxxxxxx",
"State": "ENABLED",
"ScheduleExpression": "cron(0 7,18 * * ? *)",
"EventBusName": "default"
}
  ]
}