Skip to main content
A deploy fails and the ECS service events show:

​
Why this happens

Fargate has an account-level, per-region quota on the total vCPU running at once — one for On-Demand and a separate one for Spot. New accounts start low (often 6 vCPU) and AWS raises it gradually as usage grows. Rolling deploys briefly run old and new tasks side by side, so a service can fail to deploy even though its steady-state footprint fits. Size the quota for at least twice the sum of vcpu × max_capacity (or vcpu × desired_count for services without autoscaling) across every Fargate service in the region. That covers a full-scale rollout while scaled to the maximum.

​
Fix: request a quota increase

Open Service Quotas for Fargate in the region you deploy to and request an increase: Moderate increases are approved within minutes; large jumps may need a support case. Retry the deployment once the new quota shows as applied.
In a brand-new account, the first Fargate task in a region may fail even at tiny sizes until AWS finishes activating the service. Wait a few minutes and retry before requesting an increase.

​
Fix: reduce the footprint

If you cannot wait for the increase:
  • Lower fargate_size on services that do not need the default 2 vCPU.
  • Lower max_capacity or desired_count temporarily so the rollout fits.
  • Scale down preview or staging environments in the same region.

​
Fix: move to EC2 capacity

The Fargate quota does not apply to tasks placed on EC2 capacity. Set capacity_provider: ec2 on the service and attach it to a cluster with an EC2 capacity provider — see rvn-ecs-cluster. EC2 has its own instance quotas, but they are separate and typically larger.