AWS Fargate on a Budget

Photo by Alexander Grey on Unsplash

The Business Problem

At the day job, there's a new application that needs to be deployed in a manner that meets these requirements:

• the code is server side and it needs to run longer than 15 minutes
• this is an internal use app, it can tolerate occasional disruptions
• non-engineering stakeholders should be able to access this application
• the application will be used by fewer then 10 people and there will be periods of time where no one will be using it
• it is faster to get approval if we use Amazon Web Services, because we currently use AWS heavily

A Solution

For non-disclosure agreement and security reasons this is a variation of the solution, not the exact solution.

It includes the following:

1. One Fargate Spot instance behind an Application Load Balancer
2. Two EventBridge Scheduler schedules that turn the Fargate Spot instances on at a specified time and off at a specified time

Disclaimer: this is not the only way to accomplish this.

Architecture Diagram and Example Code

Here's the high level architecture diagram of this solution.

Example code with instructions on how to deploy this into your AWS account, can be found in this GitHub repo.

Disclaimer: in this example, we are using a demo Docker image. If you are setting up an application that contains information that shouldn't be open to the public internet, ensure that authentication is in place. Authentication has been intentionally omitted, because it is not the focus of this article.

Let's zoom in a bit closer on the two major components of this solution:

1. Fargate Spot
2. EventBridge Scheduler

Fargate Spot

With Fargate there are two capacity provider types: Fargate and Fargate Spot.

Fargate is the standard which gives you on-demand access to containerized compute.

Fargate Spot is similar to Fargate, but cheaper (advertised as up to 70% discounted) and can be interrupted by AWS. The reason for this is because AWS operates at a massive scale and lots of times there are instances available that will run and cost AWS money regardless. In order to make money off of this extra capacity, AWS offers this extra capacity as Spot instances at a discounted rate. Because AWS might need this capacity back as demand rises, they reserve the right to give you a two minute warning before shutting down your instance to put it back into the regular Fargate on-demand pool.

The launch blog post has an excellent overview for an AWS official description.

Launching a Fargate Spot instance involves 1. launching it and 2. setting up the container to handle a graceful shutdown given a two minute warning from AWS.

Launch the Instance

The example code uses the ApplicationLoadBalancedFargateService CDK construct.

You can specify one or multiple capacity providers with the capacityProviderStrategies property with weights. The higher the weight, the more often that launch type will be utilized.

The example code sets things up to prefer Spot, but to use a regular Fargate launch type in case there is no Spot availability.

capacityProviderStrategies: [
          {
            capacityProvider: 'FARGATE_SPOT',
            weight: 50,
          },
          // Backup in case FARGATE_SPOT is not available
          {
            capacityProvider: 'FARGATE',
            weight: 1,
          },
        ],

Handle Graceful Shutdowns

In the example code, this is done in two places:

1. Adds a stopTimeout to the container on the Fargate Task for 120 seconds (needs to be <= 2 minutes)
2. Sets the deregistration delay on the Application Load Balancer at 30 seconds to give the instance time to separate from the Load Balancer before terminating (also needs to be <= 2 minutes)

EventBridge Scheduler

In the example, there are two EventBridge Scheduler schedules - "up" and "down".

This is accomplished by using an AWS CLI command to set the desiredCount on the ECS Service to either 1 (on/up) or 0 (off/down). Here's the official documentation for the command.

If this command were run on the command line via the AWS CLI it would look something like:

aws ecs update-service --cluster <cluster-name> --service <service-name> --desired-count <desired-count-int>

For this app, we're using EventBridge's Universal Targets feature. Universal Targets allow you to run most (not all - see the link to view unsupported commands) AWS CLI commands directly in an EventBridge Schedule without the need to add (and pay for) compute (such as Lambda).

In our example, we have one Schedule that sets the desiredCount to 1 (on/up) at 9am PT Mon-Fri and another that sets the desiredCount to 0 (off/down) at 5pm PT Mon-Fri.

These cron expressions can be adjusted...be sure to ask your stakeholder(s) when they plan to use the app to ensure it is available.

Summary

This was a walkthrough of a solution that allows us to run a Fargate instance on a budget. The solution utilizes Fargate Spot and EventBridge Scheduler to periodically shut down the Fargate Tasks during periods of no usage.

The example code repository can be found on GitHub.

What tips and tricks do you have for saving money with Fargate?

Comments

[Vince Fulco (It / It's)]

Terrific walkthru. Look forward to studying further.