That error means the task was started on Fargate with an invalid CPU/memory combo. Fargate does not support 3 vCPU (3072). Valid Fargate CPU values are 256, 512, 1024, 2048, 4096, 8192 with specific memory ranges. See AWS’s matrix: Valid Fargate CPU/memory combinations. Since you intend to use ECS on EC2 (not Fargate), your ECS work pool’s job template is almost certainly defaulting to Fargate. Change it to EC2. What to change in your Prefect ECS work pool (Prefect 3.x): - launch_type: Set to

EC2

. Do not use

FARGATE

. - Or capacity_provider_strategy: If you use capacity providers on your cluster, set a strategy referencing your EC2 provider and omit

launch_type

entirely. Don’t set both. - cpu/memory: Pick any values that make sense for your EC2 instances (they are not restricted to the Fargate matrix). For your POC, small defaults like

cpu: 256

,

memory: 512

are fine. - subnets/security groups: Set them under the run-task network configuration (awsvpc). Example below. - task_role_arn/execution_role_arn: Provide your task role (app permissions) and the ecsTaskExecutionRole (pulling images, logs, etc.). Example base job template (edit this in Prefect UI → Work Pools → your ECS pool → Edit base job template). Use either launch_type or capacity_provider_strategy, not both:

{
  "region": "ap-southeast-1",
  "cluster": "your-ecs-cluster",
  "image": "public.ecr.aws/your-image:tag",
  "cpu": 256,
  "memory": 512,

  "launch_type": "EC2",
  // If you prefer capacity providers instead of launch_type, remove launch_type and use:
  // "capacity_provider_strategy": [
  //   {"capacityProvider": "EC2", "weight": 1}
  // ],

  "task_role_arn": "arn:aws:iam::123456789012:role/YourTaskRole",
  "execution_role_arn": "arn:aws:iam::123456789012:role/ecsTaskExecutionRole",

  "network_configuration": {
    "awsvpcConfiguration": {
      "subnets": ["subnet-abc123", "subnet-def456"],
      "securityGroups": ["sg-0123456789abcdef0"],
      "assignPublicIp": "DISABLED"
    }
  }

  // Note: Prefect’s ECS worker will construct the task definition.
  // For EC2 it will use awsvpc networking by default if you provide awsvpcConfiguration.
}

A few tips for your setup: - That Fargate error is a strong signal your pool is currently using Fargate. After switching to

EC2

(or to an EC2 capacity provider), the error should disappear. - Ensure your ECS cluster actually has EC2 capacity (container instances) registered and in ACTIVE state. If you want everything on a single host, that host must be an ECS container instance (ECS agent installed) in the cluster. - If you previously added Fargate-only fields (e.g.,

platformVersion

,

requiresCompatibilities: ["FARGATE"]

,

ephemeralStorage

), remove them for EC2. - You can override cpu/memory per deployment using deployment

job_variables

if you want different sizes by environment. Relevant docs: - Prefect ECS guide (Prefect 3): Prefect ECS guide - Job variables override (per-deployment): Customize job variables - AWS: Valid Fargate CPU/memory combos: Fargate CPU/memory If you share your current work pool base job template (redact ARNs/IDs), I can suggest the exact edits to switch it from Fargate to EC2 for your cluster in ap-southeast-1.