We've compiled all ECS Fargate cost optimization tips in this confluence page, including the 2 new options: Fargate with Graviton (20% cost reduction) and Fargate Spot (up to 70% cost reduction). Our ECS best practices has also been updated to mention that Fargate with Graviton should be enabled for all ECS services which do not use Fargate Spot. Unfortunately, we can't combine them, for now.

Who is this announcement for?

Why do I need this?

What do you need to do?

All of these tasks are NOT urgent, there's no specific organization-wide deadline. Kindly decide the priority internally within your team.

Know the risks of Fargate Spot listed in the cost optimization page.

Then, decide which backend services to use Spot. The remaining services will use Graviton.
When you're not sure whether a particular backend service is safe for Spot, just start with Graviton.

Follow the steps to enable either Spot / Graviton, in staging and then production.
During the optimization, be aware that:

ECS memory parameter should be set for all (including app) containers defined in the task definition.
Unnaturally, the ecs.fargate.cpu.percent metric value could exceed far above 100%. We've added some notes in our BCP datadog dashboard and the ECS doc - current limitations.
It's better to start by upgrading the com.traveloka.common.gradle:java-base-plugin version to 7.3.1, which removes the aggressive default heap size settings from the earlier versions.

ECS Fargate Cost Optimizations

Who is this announcement for?

Why do I need this?

What do you need to do?