Aws Fargate

Fargate removes the undifferentiated heavy lifting of managing EC2 instances for containerized workloads — just define a task (container image, CPU, memory, environment), and Fargate provisions compute on-demand. For agents running batch processing jobs, background workers, or short-lived services: RunTask launches a container, WaitUntilTasksStopped blocks until completion, and DescribeTasks retrieves exit codes and logs. Fargate Spot is 50-70% cheaper for interruption-tolerant workloads. ECS + Fargate is the most common AWS pattern for running containers without Kubernetes. Confidence is docs-derived.

ECS API (via AWS SDK): RunTask launches a task ({cluster, taskDefinition, launchType: FARGATE, networkConfiguration: {awsvpcConfiguration: {subnets, securityGroups, assignPublicIp}}, overrides: {containerOverrides: [{name, environment, command}]}}). DescribeTasks retrieves task status and exit codes. StopTask terminates a running task. ListTasks lists tasks in a cluster. boto3 (Python): ecs.run_task(**params) → response includes task ARN; ecs.wait_until_tasks_stopped(cluster, tasks) blocks until done; ecs.describe_tasks(cluster, tasks) retrieves exit codes. Task definitions: define in ECS Console or via RegisterTaskDefinition API (container image, CPU/memory, port mappings, environment, secrets from SSM/Secrets Manager).

IAM-based auth: SigV4 request signing via AWS SDK / boto3. IAM role permissions: ecs:RunTask, ecs:DescribeTasks, ecs:StopTask, iam:PassRole (to pass task execution role). Task execution role: allows Fargate to pull ECR images and inject secrets. Task role: permissions for containers to call AWS services. VPC: Fargate tasks run in VPC subnets — ensure subnets have internet access (public subnet + auto-assign IP, or NAT gateway for private subnets). Secrets injection: reference SSM Parameter Store / Secrets Manager ARNs in task definition (no plaintext env var storage).

Async model: RunTask returns task ARN; task transitions PENDING → RUNNING → STOPPED. Wait for completion via waiter or polling DescribeTasks. Task exit codes in containers[0].exitCode — non-zero means failure. CloudWatch Logs: configure awslogs log driver in task definition for container stdout/stderr. Fargate cold start: 10-30 seconds for task to reach RUNNING state. Spot interruptions: 2-minute warning via EventBridge (handle gracefully). Fargate uptime tied to ECS/AWS infrastructure (99.99% SLA). Common failure: insufficient IP addresses in subnet (Fargate requires one ENI per task — plan subnet CIDR accordingly).

docs.aws.amazon.com/AmazonECS/latest/developerguide/AWS_Fargate.html covers task definition configuration, networking setup, Fargate Spot, and capacity providers. Getting started: create ECS cluster (Fargate type), register task definition, run first task via console or boto3 in under 15 minutes. boto3 (pip install boto3) is the standard Python interface. AWS CLI (aws ecs run-task). Task definition wizard in ECS console for first-time configuration. Community via AWS re:Post and Stack Overflow (amazon-ecs tag). Fargate is mature — extensive community examples and tutorials available.