ECS task placement strategy to deploy across group of ec2 instances

Question

Scenario - 2 t3.medium machines registered to the ecs cluster. I want to run 9 java microservices containers
Should I create a single task definition and have all the containers in it and have a service with placement strategy set to spread across instanceids? - IS ecs smart enough to decide the container placement across instances based on the cpu and memory available in them?
OR
Should I create a task definition for each microservice?
Please advise on what is the better approach and best practices.

Answer

Hi,

Is it a real requirement for you to run with ECS on your own EC2 instances?
If not, I would suggest to switch to Fargate, which is the serverless version of ECS. It will handle scaling, placement, HA, etc. for you.

So, less "heavy lifting" on your side.

See https://aws.amazon.com/fargate/

Additionally, for some use cases, Fargate may prove cheaper then EC2: it depends on the utilization of your instances. See this blog post for some calculations to see which one is better: https://aws.amazon.com/blogs/containers/theoretical-cost-optimization-by-amazon-ecs-launch-type-fargate-vs-ec2/

Best,

Didier

Answer

All containers in a task are run on the same host. If you want to spread those 9 containers over both hosts, you should put them into minimum of 2 tasks, or have task per container if it doesn't make difference where individual container is run (same or different host).

> The task definition is a document that describes what container images to run together, and what settings to use when running the container images.

See https://docs.aws.amazon.com/AmazonECS/latest/bestpracticesguide/application.html

ECS task placement strategy to deploy across group of ec2 instances

Relevant content