Running applications on separate machines quickly becomes an inefficient use of compute and therefore money. In this article, I demonstrate how to create a service which runs on Amazon’s Elastic Container Service with an auto scaling SpotFleet using CloudFormation.

*

Photo by Hello I'm Nik 🎞 on Unsplash*

Before beginning this article, I want to define some AWS lingo. Otherwise, this may all just go over your head:

• EC2 — Elastic Compute Service

• ASG — Auto Scaling Group

• ALB — Application Load Balancer

• SpotFleet — a collection of different machine types of spot instances

• ECS — Elastic container service — Orchestrates containers

• ECR — Elastic Container Registry — Stores Docker images

Cloud Computing enables the spinning up and tearing down of servers as compute or memory required to run an application scales. EC2s on AWS can be grouped together in an auto scaling group (ASG) which can scale on metrics such as CPU load and memory.

Architecture

For this example, I am going to create a Dockerised Python web server and deploy it to an ECS cluster which auto scales the number of containers whilst the ASG of machines it is running on scales too. An ALB is used to create an API which load balances the containers running the service. The machines forming the ASG consist of a single on-demand instance and a SpotFleet comprising two Spot instances of different types of machines.

The service

The service is an asynchronous Python web server running on port 5000 with CORS enabled. Note that the healthcheck endpoint is required for ECS to keep track of the service.

from aiohttp import web
import aiohttp_cors
import json

async def healthcheck(_):
    headers = {
        "Cache-Control": "no-cache, no-store, must-revalidate",
        "Pragma": "no-cache",
        "Expires": "0",
    }
    return web.Response(text=json.dumps("Healthy"), headers=headers, status=200)

async def helloworld(_):
    return web.Response(text="<h1>HELLO WORLD</h1>", content_type='text/html', status=200)


app = web.Application()
cors = aiohttp_cors.setup(app)
app.router.add_get("/healthcheck", healthcheck)
app.router.add_get("/", helloworld)

cors = aiohttp_cors.setup(app, defaults={
    "*": aiohttp_cors.ResourceOptions(
            allow_credentials=True,
            expose_headers="*",
            allow_headers="*",
        )
})

# Configure CORS on all routes.
for route in list(app.router.routes()):
    cors.add(route)

if __name__ == "__main__":
    print("Starting service")
    web.run_app(app, host="0.0.0.0", port=(5000))

Deployment

To deploy this system, I am using the AWS CLI with CloudFormation. It is as simple as:

aws cloudformation create-stack --stack-name service --template-body file://template.yml --capabilities CAPABILITY_NAMED_IAM

The deployment is split into five templates:

• vpc.yml

• load_balancer.yml

• cluster.yml

• machines.yml

• service.yml

Let’s Build! 🔩

VPC

I am building this service inside a VPC described in a previous article, but it’s pretty standard. There are three public and three private (hybrid) subnets.

Load Balancer

The service requires a public facing load balancer which distributes HTTP requests to the machines running the web server.

LoadBalancerSecGroup:
    Type: AWS::EC2::SecurityGroup
    Properties:
        GroupDescription: Load balancer only allow http port traffic
        VpcId: !ImportValue VPCID
        SecurityGroupIngress:
        CidrIp: 0.0.0.0/0
        FromPort: 80
        IpProtocol: TCP
        ToPort: 80
LoadBalancer:
    Type: AWS::ElasticLoadBalancingV2::LoadBalancer
    Properties:
        SecurityGroups:
        - !Ref LoadBalancerSecGroup
        Subnets:
        - !ImportValue PublicSubnetA
        - !ImportValue PublicSubnetB

Cluster

The cluster orchestrates containers running on the machines. If you are unfamiliar with Docker, check out this article. Dockerising the Python web server can be done in few lines:

FROM python:3.7-slim
COPY requirements.txt /app/requirements.txt
WORKDIR /app
RUN pip install -r requirements.txt
COPY src /app
EXPOSE 5000
CMD python app.py

The following template configures an ECS cluster and ECR to store the Docker image of the Python web server.

ECSCluster:
    Type: 'AWS::ECS::Cluster'
    Properties:
    ClusterName: ECSCluster
    CapacityProviders:
        - FARGATE_SPOT
    DefaultCapacityProviderStrategy:
        - CapacityProvider: FARGATE_SPOT
        Weight: 1
    ClusterSettings:
        - Name: containerInsights
        Value: enabled

ECRRepository: 
    Type: AWS::ECR::Repository

Machines

Scaling up and down with demand requires more than one machine in the autoscaling group of EC2s. Launch Templates allow the configuration of the machines:

• SSM for keyless SSH access and patching

• User Data script joins the machine to the ECS cluster

  LaunchTemplate:
  Type: AWS::EC2::LaunchTemplate
  Metadata:
  AWS::CloudFormation::Init:
  config:
  packages:
  yum:
  amazon-ssm-agent: []
  commands:
  00_amazon_ssm_agent_start:
  command: systemctl start amazon-ssm-agent
  files:
  /etc/cfn/cfn-hup.conf:
  content: |
  [main]
  stack=${AWS::StackId}
  region=${AWS::Region}
  interval=1
  mode: "000400"
  owner: root
  group: root
  /etc/cfn/hooks.d/cfn-auto-reloader.conf:
  content: |
  [cfn-auto-reloader-hook]
  triggers=post.update
  path=Resources.LaunchTemplate.Metadata.AWS::CloudFormation::Init
  action=/opt/aws/bin/cfn-init --verbose --stack=${AWS::StackName} --region=${AWS::Region} --resource=LaunchTemplate
  runas=root
  services:
  sysvinit:
  cfn-hup:
  enabled: true
  ensureRunning: true
  files:
  - "/etc/cfn/cfn-hup.conf"
  - "/etc/cfn/hooks.d/cfn-auto-reloader.conf"
  Properties:
  LaunchTemplateData:
  CreditSpecification:
  CpuCredits: Unlimited
  ImageId: 'ami-09266271a2521d06f' #ecs optimised image for eu-west-1
  InstanceType: t2.micro
  IamInstanceProfile:
  Arn: !GetAtt InstanceProfile.Arn
  Monitoring:
  Enabled: true
  SecurityGroupIds:
  - !Ref EC2SecurityGroup
  UserData:
  Fn::Base64:
  Fn::Sub:
  - |
  #!/bin/bash -xe
  yum update -y
  yum install -y aws-cli aws-cfn-bootstrap
  echo ECS_CLUSTER=${Cluster} >> /etc/ecs/ecs.config
  echo ECS_ENABLE_SPOT_INSTANCE_DRAINING=1 >> /etc/ecs/ecs.config
  # -r needs to be name of the LaunchTemplate resource. In this case: LaunchTemplate
  /opt/aws/bin/cfn-init -v -s ${AWS::StackName} -r LaunchTemplate --region ${AWS::Region}
  - { Cluster: !ImportValue ECSCluster }

The scaling configuration sets the group of EC2s to scale up when total CPU usage exceeds 70%.

AutoScalingPolicy:
    Type: AWS::AutoScaling::ScalingPolicy
    Properties:
    AutoScalingGroupName: !Ref AutoScalingGroup
    PolicyType: TargetTrackingScaling
    TargetTrackingConfiguration:
        PredefinedMetricSpecification:
        PredefinedMetricType: ASGAverageCPUUtilization
        TargetValue: 70

Spot instances are far cheaper than on demand. The recommended way of taking advantage of spot instances is by creating a SpotFleet — a collection of different types of instances which increase the likelihood of maintaining desired capacity.

AutoScalingGroup: 
    Type: AWS::AutoScaling::AutoScalingGroup
    Properties:
    MinSize: "1"
    MaxSize: "2"
    DesiredCapacity: "2"
    HealthCheckGracePeriod: 300 
    MixedInstancesPolicy:
        InstancesDistribution:
        OnDemandBaseCapacity: 1
        OnDemandPercentageAboveBaseCapacity: 0
        SpotAllocationStrategy: capacity-optimized
        LaunchTemplate:
        LaunchTemplateSpecification: 
            LaunchTemplateId: !Ref LaunchTemplate
            Version: !GetAtt LaunchTemplate.LatestVersionNumber
        Overrides:
            - InstanceType: t3.micro
            - InstanceType: t3a.micro
            - InstanceType: t2.micro
    VPCZoneIdentifier:
        - !ImportValue PrivateSubnetA
        - !ImportValue PrivateSubnetB

Service

We configure the load balancer to listen on port 80 for HTTP requests and send them to a Target Group — a reference we can use when defining the service to access traffic.

TargetGroup:
    Type: AWS::ElasticLoadBalancingV2::TargetGroup
    Properties:
    Port: 5000
    Protocol: HTTP
    VpcId: !ImportValue VPCID
    HealthCheckIntervalSeconds: 60
    HealthCheckTimeoutSeconds: 5
    UnhealthyThresholdCount: 5
    HealthCheckPath: /healthcheck
    TargetGroupAttributes:
        - Key: deregistration_delay.timeout_seconds
        Value: 2

LoadBalancerListener:
    Type: AWS::ElasticLoadBalancingV2::Listener
    Properties:
    DefaultActions:
        - TargetGroupArn: !Ref TargetGroup
        Type: forward
    LoadBalancerArn: !ImportValue LoadBalancerArn
    Port: 80
    Protocol: HTTP

ListenerRule:
    Type: AWS::ElasticLoadBalancingV2::ListenerRule
    Properties:
    Actions:
        - TargetGroupArn: !Ref TargetGroup
        Type: forward
    Conditions:
        - Field: path-pattern
        Values:
            - '*'
    ListenerArn: !Ref LoadBalancerListener
    Priority: 1

ECS runs the Task Definition as a persistent service using the web server image in ECR.

TaskDefinition:
    Type: AWS::ECS::TaskDefinition
    Properties:
    Family: AppTaskDefinition
    TaskRoleArn: !GetAtt TaskRole.Arn
    ExecutionRoleArn: !ImportValue ExecutionRoleArn
    Memory: 0.5Gb
    Cpu: 256
    ContainerDefinitions:
        - Name: ServiceContainer
        PortMappings:
            - ContainerPort: 5000
        Essential: true
        Image: <your_image_tag>
        LogConfiguration:
            LogDriver: awslogs
            Options:
            awslogs-group: !Ref LogGroup
            awslogs-region: !Ref 'AWS::Region'
            awslogs-stream-prefix: ecs

Service:
    Type: AWS::ECS::Service
    DependsOn:
    - ListenerRule
    Properties:
    Cluster: !ImportValue ECSCluster
    LaunchType: EC2
    DesiredCount: 2
    LoadBalancers:
        - ContainerName: ServiceContainer
        ContainerPort: 5000
        TargetGroupArn: !Ref TargetGroup
    DeploymentConfiguration:
        MinimumHealthyPercent: 100
        MaximumPercent: 300
    HealthCheckGracePeriodSeconds: 30
    TaskDefinition: !Ref TaskDefinition

Configuring an auto-scaling policy on the containers works in much the same way as the EC2 machines. This is because they have defined memory and CPU so can be scaled based on those metrics, too:

AutoScalingTarget:
    Type: AWS::ApplicationAutoScaling::ScalableTarget
    Properties:
    MaxCapacity: 3
    MinCapacity: 2
    ResourceId: !Join ["/", [service, !ImportValue ECSCluster, !GetAtt Service.Name]]
    RoleARN: !ImportValue ECSServiceAutoScalingRoleArn
    ScalableDimension: ecs:service:DesiredCount
    ServiceNamespace: ecs

AutoScalingPolicy:
    Type: AWS::ApplicationAutoScaling::ScalingPolicy
    Properties:
    PolicyName: ServiceAutoScalingPolicy
    PolicyType: TargetTrackingScaling
    ScalingTargetId: !Ref AutoScalingTarget
    TargetTrackingScalingPolicyConfiguration:
        PredefinedMetricSpecification:
        PredefinedMetricType: ECSServiceAverageCPUUtilization
        ScaleInCooldown: 10
        ScaleOutCooldown: 10
        TargetValue: 70

Usage

After a successful deployment, it is possible to access the DNS name of the ALB in the EC2 section of the AWS console, which should look something like:

loadb-LoadB-R7RVQD09YC9O-1401336014.eu-west-1.elb.amazonaws.com

Now it is possible to make a request to this URL and get a response:

$ curl loadb-LoadB-R7RVQD09YC9O-1401336014.eu-west-1.elb.amazonaws.com'
<h1>HELLO WORLD!</h1>

Use a Domain Name

AWS provides an ugly Load Balancer address such as:

loadb-LoadB-R7RVQD09YC9O-1401336014.eu-west-1.elb.amazonaws.com

But it’s quite simple to use a custom domain using AWS. Firstly, transfer your DNS management to Route 53 and then create a new record set aliased to the load balancer.

Thanks For Reading

I hope you have enjoyed this article. If you like the style, check out T3chFlicks.org for more tech focused educational content (YouTube, Instagram, Facebook, Twitter).

Resources:

• https://aws.amazon.com/blogs/containers/deep-dive-on-amazon-ecs-cluster-auto-scaling/

• https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/aws-properties-autoscaling-autoscalinggroup-mixedinstancespolicy.html

• https://docs.aws.amazon.com/autoscaling/ec2/userguide/asg-purchase-options.html