Check what drives your Fargate resource charges
On this page
Fargate charges follow requested resources and billable runtime, so record both before estimating a sizing change. If you skip this, observed CPU or memory use can understate the comparison. Do not treat a supported smaller task size as proof that the workload is safe.
What you need first
- Tool
- Use a worksheet, not a command-line tool. Start with task configuration and run records from an authorized colleague, then compare the inputs with the AWS Fargate Pricing page.
- Access
- Ask an authorized AWS colleague for read-only task configuration and run records covering the selected ECS Fargate tasks, Region and review period.
- If you do not use that tool
- Send the ECS workload owner the task set, Region and review period. Request vCPU, memory, operating system, architecture, ephemeral storage, image-download start, termination time, and Spot or Savings Plan context.
Why this is worth a look
Skipping image-download time, extra ephemeral storage or Windows charges can distort the estimate. AWS examples separate CPU in vCPU-seconds, memory and additional ephemeral storage in GB-seconds, with a Windows operating-system charge per vCPU-second. Region, operating system, CPU architecture, Spot pricing and Savings Plan context also affect a like-for-like comparison.
Run this check
CHECKLISTComplete this worksheet from approved task records and the AWS Fargate Pricing page. It makes no commands or resource changes.
Scope: ECS (Amazon Elastic Container Service) Fargate tasks in one AWS Region.
Time window: record the task runs or review period, including start and end dates.
Access: ask an authorized AWS colleague for read-only configuration and run records for that scope.
Units: vCPU; GB for memory and storage (1 GB = 1024^3 bytes); seconds for duration.
[ ] Record the task set, Region and review period.
[ ] For each configuration, record requested vCPU, memory, operating system, CPU architecture and configured ephemeral storage.
[ ] Compare each current and proposed vCPU/memory pair with Supported Configurations on the AWS Fargate Pricing page.
[ ] Record additional chargeable ephemeral storage as configured storage above the included 20 GB. Use zero when there is no additional storage.
[ ] For each run, record image-download start and task-termination times. Round up to the nearest second, then apply a 60-second minimum, or a 300-second minimum for Windows containers.
[ ] Match pricing to the Region, operating system and architecture. Record whether Fargate Spot or a Savings Plan applies.
[ ] Keep CPU (vCPU-seconds), memory (GB-seconds) and additional ephemeral storage (GB-seconds) separate. Include Windows operating-system charges per vCPU-second where applicable.
[ ] List CloudWatch, public IPv4, data transfer and other AWS service charges separately from task resource charges.
[ ] Mark missing inputs before comparing current and proposed configurations. Do not infer safe sizing from requested resources alone.How to confirm it
- 01
Choose tasks and get records
Choose one ECS Fargate task or task set, one Region and a review period. Ask the workload owner for the records listed in the worksheet. Keep missing inputs visible instead of treating them as zero.
- 02
Check the requested size
Record the requested CPU, memory, operating system, architecture and storage. Compare current and proposed CPU/memory pairs with AWS Supported Configurations before estimating the change.
- 03
Check billable runtime
Record image-download start and termination for each run. Round each duration up to the nearest second and apply the documented minimum to each task run, not to the whole review period.
- 04
Compare like-for-like costs
Match Region, operating system, architecture and pricing context before comparing configurations. Keep CloudWatch, public IPv4, data transfer and other service charges separate, then ask the workload owner to validate capacity before changing size.
Before making changes
This review assumes complete configuration and run records for ECS Fargate tasks in one Region over a stated period. It checks pricing inputs, not observed utilization or actual billed totals. A supported CPU/memory pair shows an allowed configuration, not that the workload can handle it.