AWS Fargate serverless container compute for ECS. Covers Fargate vs EC2 decision guide, CPU/memory sizing, platform versions, Fargate Spot cost optimization, Graviton/ARM architecture, networking, and EFS integration. Use when deploying serverless containers, optimizing Fargate costs, sizing Fargate tasks, or choosing between Fargate and EC2 launch types.
64
76%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Passed
No findings from the security scan
Fix and improve this skill with Tessl
tessl review fix ./.claude/skills/ecs-fargate/SKILL.mdThe canonical home for this skill is ecs-fargate in fernandezbaptiste/Skrillz
Complete guide to AWS Fargate serverless container compute, including sizing, optimization, and best practices.
| Aspect | Fargate | EC2 |
|---|---|---|
| Management | Serverless | Manual |
| Scaling | Automatic | ASG-based |
| Pricing | Per vCPU-second | Per instance-hour |
| Overhead | None | High |
| Control | Limited | Full |
| Best For | Variable load | Steady state |
Use Fargate when:
Use EC2 when:
| CPU (vCPU) | Memory Options |
|---|---|
| 0.25 | 512 MB, 1 GB, 2 GB |
| 0.5 | 1 GB - 4 GB (1 GB increments) |
| 1 | 2 GB - 8 GB (1 GB increments) |
| 2 | 4 GB - 16 GB (1 GB increments) |
| 4 | 8 GB - 30 GB (1 GB increments) |
| 8 | 16 GB - 60 GB (4 GB increments) |
| 16 | 32 GB - 120 GB (8 GB increments) |
# Sizing decision tree
def recommend_fargate_size(app_type: str, expected_memory_mb: int):
"""Recommend Fargate CPU/Memory based on workload"""
if app_type == "web_api":
# Typically CPU-bound
if expected_memory_mb <= 512:
return {"cpu": "256", "memory": "512"}
elif expected_memory_mb <= 1024:
return {"cpu": "512", "memory": "1024"}
else:
return {"cpu": "1024", "memory": str(min(expected_memory_mb, 2048))}
elif app_type == "worker":
# Typically memory-bound
return {"cpu": "256", "memory": str(max(512, expected_memory_mb))}
elif app_type == "batch":
# Need burst capacity
return {"cpu": "1024", "memory": str(max(2048, expected_memory_mb))}
# Default conservative
return {"cpu": "512", "memory": "1024"}# Development: Minimum viable
cpu = "256"
memory = "512"
# Small API: Balanced
cpu = "512"
memory = "1024"
# Production API: Standard
cpu = "1024"
memory = "2048"
# Memory-intensive: Data processing
cpu = "1024"
memory = "4096"
# Compute-intensive: ML inference
cpu = "4096"
memory = "8192"| Version | Status | Features |
|---|---|---|
| 1.4.0 | Recommended | containerd runtime, jumbo frames, UDP support |
| 1.3.0 | Deprecated (Mar 2026) | Docker runtime |
| LATEST | Auto-updates | Always latest revision |
# Always pin explicitly
resource "aws_ecs_service" "app" {
platform_version = "1.4.0" # NOT "LATEST"
# ...
}# Boto3
ecs.create_service(
platformVersion='1.4.0',
# ...
)# Mixed strategy: reliable baseline + cost-effective scaling
resource "aws_ecs_service" "app" {
capacity_provider_strategy {
capacity_provider = "FARGATE"
weight = 1
base = 2 # Always keep 2 on-demand
}
capacity_provider_strategy {
capacity_provider = "FARGATE_SPOT"
weight = 4 # 4x weight = ~80% on Spot
}
}# Boto3
ecs.create_service(
capacityProviderStrategy=[
{'capacityProvider': 'FARGATE', 'weight': 1, 'base': 2},
{'capacityProvider': 'FARGATE_SPOT', 'weight': 4}
],
# ...
)# Application-level graceful shutdown
import signal
import sys
def shutdown_handler(signum, frame):
print("Received termination signal, shutting down gracefully...")
# Drain connections
# Complete in-flight requests
# Save state if needed
sys.exit(0)
signal.signal(signal.SIGTERM, shutdown_handler)resource "aws_ecs_task_definition" "app" {
family = "my-app"
requires_compatibilities = ["FARGATE"]
# ARM architecture
runtime_platform {
operating_system_family = "LINUX"
cpu_architecture = "ARM64" # or "X86_64"
}
# ...
}# Dockerfile supporting both architectures
FROM --platform=$TARGETPLATFORM python:3.11-slim
# Build for multiple platforms
# docker buildx build --platform linux/amd64,linux/arm64 -t myapp:latest .# Build and push multi-arch image
docker buildx create --use
docker buildx build \
--platform linux/amd64,linux/arm64 \
--push \
-t 123456789.dkr.ecr.us-east-1.amazonaws.com/myapp:latest .Each Fargate task gets its own ENI with:
resource "aws_ecs_service" "app" {
network_configuration {
subnets = var.private_subnets
security_groups = [aws_security_group.ecs_tasks.id]
assign_public_ip = false # Use NAT for outbound
}
}# Avoid NAT costs for AWS services
resource "aws_vpc_endpoint" "ecr_api" {
vpc_id = module.vpc.vpc_id
service_name = "com.amazonaws.${var.region}.ecr.api"
vpc_endpoint_type = "Interface"
subnet_ids = module.vpc.private_subnets
}
resource "aws_vpc_endpoint" "ecr_dkr" {
vpc_id = module.vpc.vpc_id
service_name = "com.amazonaws.${var.region}.ecr.dkr"
vpc_endpoint_type = "Interface"
subnet_ids = module.vpc.private_subnets
}
resource "aws_vpc_endpoint" "s3" {
vpc_id = module.vpc.vpc_id
service_name = "com.amazonaws.${var.region}.s3"
}
resource "aws_vpc_endpoint" "logs" {
vpc_id = module.vpc.vpc_id
service_name = "com.amazonaws.${var.region}.logs"
vpc_endpoint_type = "Interface"
subnet_ids = module.vpc.private_subnets
}# EFS File System
resource "aws_efs_file_system" "app" {
creation_token = "app-storage"
encrypted = true
lifecycle_policy {
transition_to_ia = "AFTER_30_DAYS"
}
}
resource "aws_efs_mount_target" "app" {
count = length(var.private_subnets)
file_system_id = aws_efs_file_system.app.id
subnet_id = var.private_subnets[count.index]
security_groups = [aws_security_group.efs.id]
}
# Task Definition with EFS
resource "aws_ecs_task_definition" "app" {
family = "my-app"
volume {
name = "app-storage"
efs_volume_configuration {
file_system_id = aws_efs_file_system.app.id
root_directory = "/app-data"
transit_encryption = "ENABLED"
authorization_config {
access_point_id = aws_efs_access_point.app.id
iam = "ENABLED"
}
}
}
container_definitions = jsonencode([
{
name = "my-app"
mountPoints = [
{
sourceVolume = "app-storage"
containerPath = "/data"
readOnly = false
}
]
# ...
}
])
}# Default: 20 GB, configurable up to 200 GB
resource "aws_ecs_task_definition" "app" {
ephemeral_storage {
size_in_gib = 100 # For large temp files
}
}| Strategy | Savings | Effort |
|---|---|---|
| Fargate Spot | 70% | Low |
| ARM/Graviton | 20% | Medium |
| Right-sizing | 10-50% | Medium |
| Compute Savings Plans | 50% | Low |
| Schedule-based scaling | 30-60% | Medium |
def estimate_monthly_cost(cpu_vcpu: float, memory_gb: float,
hours_per_month: int = 730,
spot_percentage: float = 0.0):
"""Estimate monthly Fargate cost (us-east-1 pricing)"""
# On-demand pricing (approximate)
cpu_rate = 0.04048 # per vCPU-hour
memory_rate = 0.004445 # per GB-hour
on_demand_hours = hours_per_month * (1 - spot_percentage)
spot_hours = hours_per_month * spot_percentage
# Spot is ~70% cheaper
cpu_cost = (cpu_vcpu * cpu_rate * on_demand_hours) + \
(cpu_vcpu * cpu_rate * 0.3 * spot_hours)
memory_cost = (memory_gb * memory_rate * on_demand_hours) + \
(memory_gb * memory_rate * 0.3 * spot_hours)
return cpu_cost + memory_cost
# Example: 1 vCPU, 2 GB, 50% Spot
cost = estimate_monthly_cost(1, 2, spot_percentage=0.5)
print(f"Estimated monthly cost: ${cost:.2f}") # ~$2593ed392
Canonical home
since Sep 12, 2026
If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.