- 1. Tutorial overview
- 2. Configuring the Service Auto Scaling
- 3. Generating Load
- 4. Scaling your environment
At this moment, you should be able to run your application inside a container and deploy it on ECS as a service, but as you probably noticed, we didn't create any rule or used any tool to automatically scale the number of containers running in our cluster.
When working with Amazon ECS, we can also take advantage of Auto Scaling features, that will allow us to add more containers of a specific task definition according to some metrics that are already available. In this part of our workshop, we will create some Auto Scaling rules with our service, in order to keep our application available even during access peaks.
When we talk about scaling on ECS, we need to keep in mind that there are two things to be scaled during an event or an access peak: the number of tasks of a given service and the number of instances within your ECS cluster. In this workshop we are going to cover the auto scaling configuration for the tasks only. If you wish to learn more about EC2 Auto Scaling, you can get more information in this link.
Let's get started.
In the AWS Management Console, go to ECS. On the left side menu, click in Clusters and click in the cluster containers-workshop-ecs-cluster. Find the Services tab, select the service containers-workshop-ecs-service and click in Update:
Now, in the Configure service screen, just click in Next step. Click in Next step again under Configure network screen. You will reach the Set Auto Scaling screen. Here is where we are going to make the first changes.
Under Service Auto Scaling, select the option Configure Service Auto Scaling to adjust your service’s desired count. Now, use the following values on the fields:
Minimum number of tasks: 1
Desired number of tasks: 1
Maximum number of tasks: 5
After adding this information, let's add a new scaling policy, which will execute the scaling actions in your cluster based in some specific metrics. Click in the button Add scaling policy.
In this screen, give your policy the name containers-workshop-ecs-scaling-policy. In this workshop, we are going to scale the tasks every time the metric ECSServiceAverageCPUUtilization reaches 10%. Add the value 10 in Target value, change the Scale-out cooldown period and the Scale-in cooldown period to 30.
NOTE: We are using lower values just to see the scaling in action very quickly, and also, we will be generating load using just one EC2 instance. In a real world scenario, you might want to use higher values to scale your application.
Leave the other parameters with the default values and then click in Save:
After saving the policy, click in Next step, review your configurations and then click in Update Service.
You can check the Auto Scaling configurations of your service by clicking in the service containers-workshop-ecs-service under your cluster, and then selecting the tab Auto Scaling:
Now that we have our Service Auto Scaling configured, we need to generate load in order to scale the number of running tasks in our cluster. To do so, we will be using an open source tool called Locust. You can find more information about Locust in their documentation.
Since the focus of this workshop is understanding how to run containers on AWS, we will provide you a CloudFormation template that will create an EC2 instance, install and configure Locust in it.
You can use one of the following templates to provision the load testing stack in your account.
NOTE: You must deploy this CloudFormation template in the same region where you created your VPC otherwise it will not work.
| Deploy | Region |
|---|---|
 |
US East (N. Virginia) |
|
US East (Ohio) |
|
US West (Oregon) |
|
EU (Ireland) |
|
Asia Pacific (Singapore) |
When provisioning the template, you will be asked to input some information. For the HTTP Location parameter, you can add you public IP Address or just leave it as the default.
Since we will be generating a large ammount of load, it's recommended that you use a large instance type. We are setting the default as m5.xlarge.
In order to provision the EC2 instance, you will need to have a keypar created under your EC2 console. If you are using a new account and don't have a keypar created, you can follow this link to learn how to create it.
Under Load Balancer URL, you should add the hostname of your Load Balancer. This is the same name that you used to access your application in the step 04-Deploy ECS Cluster.
Then, click in Next and in the Options screen, click in Next again. Under the Review screen, validate your configurations and click in Create.
After creating your new Stack, you should see a URL that point to your EC2 instance with the Locust installed:
Click in this URL and you will be redirected to the Locust page in your instance. In this screen, you can see that the HOST is pointing to your Load Balancer hostname. Now, under Number of users to simulate you can add 20000 and under Hatch rate add 500:
After configuring Locust, you can click in Start swarming.
This will start to simulate 20000 users accessing your application. You can follow the results in the following screen:
After generating load against your application, you should be able to see it scaling after a few seconds. Go to you containers-workshop-ecs-cluster and click in the service containers-workshop-ecs-service. Go to the Events tab and you will find the message saying that the desired count of tasks have changed:
