ECS Fargate Rightsizing

Infralyst analyzes CloudWatch utilization for each ECS Fargate service and surfaces savings opportunities. When you're ready, generate ready to merge Terraform PRs from Slack or the dashboard.

  • Unlimited free recommendations
  • Read-only IAM role
  • Your team reviews and merges every change
ECS Fargate Recommendations
3 found
ResourceCPUMemoryTasksCurrentRecommendedSavings
api-serviceus-east-1 · 84 days17%avg 3.8%26%avg 14%21vCPU-2GB0.5vCPU-1GB$41/moGenerate PR
worker-processorus-east-1 · 63 days22%avg 6.2%29%avg 11%32vCPU-4GB1vCPU-2GB$58/moGenerate PR
cache-sidecarus-east-1 · 47 days18%avg 4.1%22%avg 9%10.5vCPU-2GB0.25vCPU-1GB$19/moGenerate PR
Analyzed 12 services across 2 regionsUpdated 2 min ago

You're paying for Fargate capacity
you don't use

Teams pick CPU and memory allocations when they first define a task and never revisit them. Fargate bills per vCPU-hour and GB-hour, so every unit of unused capacity is a direct cost. The problem compounds quietly across dozens of services.

AWS Cost Explorer can flag the waste, but it stops at a recommendation. Infralyst picks up where it leaves off: it analyzes real CloudWatch utilization at the service level, selects a conservative target configuration, and generates a PR your team can review and merge.

How the analysis works

A conservative, multi gate approach. If any check fails, it doesn't get recommended.

1. Collect utilization metrics

Infralyst pulls CPU and memory utilization from CloudWatch for each ECS Fargate service, using up to 365 days of historical data.

2. Run the analysis gates

Each service passes through data quality validation, utilization thresholds, and seasonality checks. A service only gets a recommendation if all gates pass.

3. Generate the recommendation

Infralyst recommends the next smaller CPU/memory configuration. When you're ready, an AI agent reads your Terraform, writes the change, and validates it. The analysis is fully deterministic.

Learn more about our methodology

Thresholds & methodology

Every gate and threshold Infralyst checks before recommending Fargate rightsizing. These are intentionally conservative. We'd rather miss a savings opportunity than recommend a change that causes problems.

Pre-flight

Shared task definitionNot shared across multiple services
Minimum sizeNot already smallest Fargate configuration

Data quality

Data window≥ 30 days of metric data
Uptime coverage≥ 95% within the data window

CPU & Memory

P99.5 · 1–60 days (5 min)< 40%
P95 · 60–365 days (hourly)< 40%

Seasonality

For services with 60–365 days of data, P95 is checked per weekly bucket. If any week exceeds the 40% threshold, the service is excluded.

Infralyst only recommends stepping down one size at a time. If there's room to go further, it happens across multiple analysis cycles.

What the PR looks like

Infralyst opens a Terraform pull request in your repo. Your team reviews and merges it like any other change.

Downsize ECS Fargate service api-service: 1vCPU-2GB → 0.5vCPU-1GB (-$41/mo)

Openinfralystbotopened this pull request

ECS Fargate service api-service in us-east-1 can be downsized from 1vCPU-2GB to 0.5vCPU-1GB, saving an estimated $41/mo.

MetricAvgP99.5
CPU3.8%17%
Memory14%26%

Lookback: 84 days

View on Infralyst

Trigger PRs directly from Slack

Get notified when new savings are found. Review the details and generate a PR without leaving Slack.

Infralyst
InfralystAPP10:03

New ECS Fargate savings recommendation · Save $41.00/mo

Resource

api-service (us-east-1)

Lookback

84d

CPU Avg

3.8%

CPU P99.5

17%

Mem Avg

14%

Mem P99.5

26%

Current

1vCPU-2GB

Recommended

0.5vCPU-1GB

View on Infralyst

Safety & guardrails

Built for production infrastructure. Here's what Infralyst won't do.

Conservative thresholds

Infralyst won't recommend rightsizing unless CPU and memory P99.5 (or P95 for longer windows) are comfortably below 40%. A single spike won't trigger a false recommendation.

Seasonality-aware

For services with 60+ days of data, P95 utilization is checked per weekly bucket over the full observation window. If any week exceeds 40%, the service is excluded. No recommendation, no risk.

Zero downtime deployment

Every PR includes a deployment configuration block. ECS rolls out the new task definition gradually, keeping your service available throughout.

PR-only changes

Infralyst never applies changes directly. Every recommendation is delivered as a notification in Slack or in our console. You trigger PR generation, review, test, and merge on your timeline.

Also available for EC2 Instances, Auto Scaling Groups, and EBS Volumes.

ECS Fargate Rightsizing FAQ

Start rightsizing your ECS Fargate services

Connect your AWS account and Infralyst will find Fargate services worth rightsizing. Generate the Terraform PR in one click when you're ready.

  • No credit card required
  • Read-only IAM role
  • Your team reviews and merges every change