Auto Scaling

Auto Scaling

The last step proved that ECS and Fargate are now serving your website and it is running as a container.

However, what happens during peak hours? Scaling up/down - would it be a great solution for your workload?

Auto Scaling is a solution to adjust capacity on the fly by adding and removing tasks based on your load.

It scales in and out, preventing you from paying unnecessary resources. In addition, it adjusts your tasks to absorb the current workload.

Load test without Auto scaling

The tool artillery is a powerful solution for load testing.

The steps below will use this tool to emulate 100 new users per second for 600 seconds.

Click here to learn more about Artillery.

Let’s flood our application with hundreds of calls.

Using Cloud9 Terminal, install the artillery tool.

npm install -g artillery 

Ensure that your terminal is pointing to the right folder.

cd ~/environment/awscloudbuilders-containerise-dotnet/

Create the ecs-load.yml file.

nano ecs-load.yml

Paste the configuration below and substitute the YOUR_ALB_URL with the address of your application.

config:
  target: 'http://YOUR_ALB_URL'
  phases:
      - duration: 600
        arrivalRate: 100
  http:
      pool: 10
scenarios:
   - flow:
       - get:
           url: "/"

Save the file by pressing

1. CONTROL + X 
2. Y
3. Enter 

Run the command

artillery run ecs-load.yml

Warning: High CPU usage warning is an expected error.

The output from the command above should be similar to the lines below.

Started phase 0, duration: 300s @ 12:25:16(+0000) 2020-02-18
Report @ 12:25:26(+0000) 2020-02-18
Elapsed time: 10 seconds
Scenarios launched:  4994
Scenarios completed: 2030
Requests completed:  2030
RPS sent: 502.31
Request latency:
    min: 4
    max: 5854.9
    median: 2483.9
    p95: 5572
    p99: 5844.4
Codes:
    200: 2030

Leave the terminal running the artillery and open the Amazon ECS console.

Click on Cluster and then click on awsbuilders-cluster.

Open the service awsbuilders-task-service.

Click on Metrics and open the Cloudwatch link for the metric CPUUtilization.

Adjust the period to 1 minute and setup the graph to display the last 15 minutes.

As highlighted in yellow below, the CPU utilization of your single task is increasing significantly due to the heavy load.

Go back to your Cloud9 Terminal and stop the artillery tool by pressing CONTROL + C.

Setup the Auto Scaling

Open the cluster awsbuilders-cluster and then click on your service.

Click on Update.

Click on Next step.

Once again, click on Next step.

Select the Configure Service Auto Scaling to adjust your service’s desired count and then insert the values below:

  1. Minimum number of tasks: 2
  2. Desired number of tasks: 2
  3. Maximum number of tasks: 10
  4. IAM role for Service Auto Scaling: Create new role

The option The create new role may not be available if your account is new. If that is your case, you don’t need to select the Create new role.

The settings above configure auto scaling to keep at least two tasks running, with a maximum of 10 tasks running at the same time.

Click here to learn more about ECS and Auto Scaling.

Click on Add scaling policy.

Select Target Tracking and insert the values below:

  1. Policy name: awsbuilders-autoscaling
  2. Target value: 5
  3. Scale-out cooldown period: 2
  4. Scale-in cooldown period: 2

The auto scaling policy will keep your metric as close as possible to the target value of 5%. Also, it won’t scale in/out within 2 seconds from the last scaling activity.

Click here to learn more about Target Tracking Scaling Policies.

Click here to learn more about Step Scaling Policies

Click on Save.

Click on Next Step.

Update the service by click on Update Service.

Wait for the changes to apply.

Your service is now provisioning the extra task as per the previous change.

Wait until the second task change the Desired status to Running.

Second load test

Run the artillery command again.

artillery run ecs-load.yml

Watch the cloudwatch metric CPUUtilization for 10 minutes and you should see a similar graph as per the below image.

Basically, auto scaling adds more tasks to reduce the cpu utilization as the workload increases. It also removes the unnecessary tasks once the load is reduced.

Check the running tasks in the Tasks pane of your service.