Request to Enable Use of Fargate Spot for Fleet ECS Services

[vireshsolanki]

Hi Team,

We would like to request approval to transition our ECS service deployment for the Fleet application to use AWS Fargate Spot in order to optimize operational costs.

Currently, the Fleet ECS tasks are running on standard Fargate capacity. Moving to Fargate Spot would help significantly reduce compute costs, especially for workloads that are fault-tolerant and can gracefully handle interruptions.

Key Notes:

No application data is stored locally on the containers (stateless).

We've ensured that tasks can safely restart without data loss or impact to user experience.

We will configure the service with a fallback to FARGATE in case Spot capacity is unavailable, ensuring high availability.
Requested Change:
Update ECS capacity provider strategy to:
fargate_capacity_providers = {
FARGATE_SPOT = {
default_capacity_provider_strategy = {
weight = 100
}
}
FARGATE = {
default_capacity_provider_strategy = {
weight = 0
}
}
}
Please let us know if you need further validation, logs, or a testing window before applying the change.

Best regards,
Viresh

[BCTBB]

@vireshsolanki - without having tested the suggestion that you provided, it looks like it should work. Are you looking for Fleet to add support for this or are you looking to make the change on your own and run Fleet in ECS with this configuration?

[vireshsolanki]

@BCTBB I did try the suggestion, but unfortunately setting the Fargate Spot to 100 weight didn’t work for me. Ideally, I’m looking for Fleet to add native support for ECS Fargate Spot so we can optimize costs. Since our containers are stateless, using Spot instances would be ideal. However, to ensure availability, we’d need to deploy across all subnets so that if one Spot capacity goes down, another one in a different AZ or subnet can take over.

[BCTBB]

@vireshsolanki Thanks trying it out and providing feedback.

Inspecting the Terraform docs and our (Fleet) Terraform code a little more deeply, it looks like we do allow the default capacity providers for the cluster to be set via a variable. However, even if you set the capacity providers, it looks like our aws_ecs_service block has the launch_type hardcoded to FARGATE, which would currently prevent us from setting FARGATE_SPOT as a capacity provider.

Ideally, I can see the default staying how we have it configured, while providing administrators, like yourself, the ability to overwrite the default configuration to allow for FARGATE, FARGATE_SPOT, or both to be configured.

I'll bring it up internally with the team to see when we can look at adding support for the capacity providers.

[vireshsolanki]

Thanks for the update @BCTBB — I had noticed earlier that the launch_type was hardcoded to FARGATE, which is likely why capacity providers like FARGATE_SPOT aren’t being picked up even if configured.

If you're open to it, I’d be happy to help make the necessary changes to support both FARGATE and FARGATE_SPOT through capacity providers. I can update the Terraform code to keep the current default behavior while allowing flexibility for admins to override it when needed.

Just let me know if you'd like me to take that on!

[BCTBB]

@vireshsolanki Absolutely, go for it.

Make the necessary changes to support both FARGATE and FARGATE_SPOT through capacity providers.

Like you noted, we definitely want to maintain FARGATE as the default and add support for FARGATE_SPOT.

I'm thinking

1. if launch_type = FARGATE, then logic is added to avoid capacity providers in aws_ecs_service.
2. if capacity_providers object is populated, then logic is added to avoid setting launch_type in aws_ecs_service.
3. If neither launch_type or capacity_providers are defined/populated, then we default to launch_type = FARGATE in aws_ecs_service.

[vireshsolanki]

@BCTBB Got it. I will start working on it