# ECS Fargate Spot Opportunity for Dev/Test

> Fargate Spot cuts up to 70% off on-demand Fargate rates, but a reclaim arrives with about 2 minutes of notice. ZopNight recommends the switch only for dev/test-named services running at least 2 tasks behind a load balancer, and prices the saving from live tier rates, never a flat fraction.

Source: https://zop.dev/integrations/aws/recommendations/ecs-fargate-spot-opportunity-for-dev-test
Updated: 2026-08-19

---

## Interruption tolerance is the axis, not utilization

AWS reclaims Fargate Spot capacity on roughly 2 minutes' notice. Whether a workload
survives that has nothing to do with how busy its CPU is. Utilization questions belong to
the rightsizing rules. What this rule asks is narrower: is the service discretionary
(a dev, test, qa, staging, or sandbox name, and nothing production-flavoured about it), is
it actually launched on Fargate per the discovered launch type, and can it prove it
tolerates losing a task mid-flight?

## Two tasks and a load balancer as the safety bar

Tolerance is required as evidence, not assumed from the environment name. The service must
run a desired count of at least 2, so a reclaim leaves at least 1 task serving, and it must
sit behind a load balancer with registered targets, so traffic reroutes around the
reclaimed task while its replacement starts. A single-task QA service on Spot turns every
reclaim into an outage for a team that is actively using it. Those services abstain here, and
their honest adoption path is a mixed capacity-provider strategy with an on-demand base,
which this rule does not model and will not pretend to.

## Live Spot rates or no recommendation

The dollar figure is derived from live on-demand versus Spot tier rates for the service's
actual SKU, supplied by the billing pipeline. When live rates are absent the rule abstains
instead of falling back to a hardcoded fraction. The ~70% you see in AWS marketing copy
is [the ceiling AWS quotes for Fargate Spot](https://aws.amazon.com/ecs/pricing/), not a measured saving. A missing or
zero-priced monthly cost also abstains.

## Confirming the launch type and task count

```bash
aws ecs describe-services --cluster dev-cluster --services my-service \
  --query 'services[0].{launchType:launchType,desired:desiredCount,loadBalancers:loadBalancers}'
```

The rule's firing shape is `launchType` FARGATE, `desired` of 2 or more, and a non-empty
`loadBalancers` list, on a service whose name reads dev/test.

## Adopting FARGATE_SPOT with a fallback

Switch the service's capacity provider strategy to FARGATE_SPOT as primary and keep plain
FARGATE as a weighted fallback, so steady-state capacity survives a Spot drought. Then
watch task stop reasons for a week: occasional reclaims are the deal you signed up for,
but frequent ones on a latency-sensitive test suite mean the base weight needs raising.