Assess how efficiently one AWS account spends on Amazon ECS and AWS Fargate. This plan inventories the account's clusters, services, and task definitions through read-only ECS APIs, collects CloudWatch and Container Insights utilization at the finest granularity available over an observation window of at least 14 days with an explicit fallback when Container Insights is not enabled, then evaluates five opportunity areas: task sizing against observed CPU and memory utilization, idle or near-idle services, autoscaling configuration and coverage, sustained versus intermittent runtime that points to scheduling opportunities, and Fargate versus EC2-backed launch-type economics including Fargate Spot. It also captures the read-only launch sources for standalone tasks, such as scheduled tasks and EventBridge rules, so intermittent workloads are not misread as idle services. This plan is analysis-only: it reads configuration, metrics, and metadata and never creates, updates, resizes, scales, stops, or deletes any cluster, service, task, or task definition.
Inventory ECS clusters, services, and task definitions
Enumerates the account's ECS clusters, services, running and stopped tasks, and the task definitions behind them, capturing configured CPU and memory, launch type, capacity provider, desired count, and identity tags.
Collect container utilization evidence
Gathers CloudWatch and Container Insights utilization at the finest granularity available over at least the configured observation window, with an explicit fallback to task counts, service events, and billing-derived runtime patterns when Container Insights is not enabled.
Assess task sizing against observed utilization
Compares each service's configured task CPU and memory against observed utilization peaks and sustained levels, accounting for Fargate's discrete CPU and memory combinations before any resizing conclusion.
Identify idle and low-value services
Identifies services that show little or no meaningful activity across the observation window, separating genuinely idle services from low-traffic services that still serve a purpose.
Assess autoscaling configuration and coverage
Reviews Application Auto Scaling configuration per service, judging which services run a fixed desired count where demand clearly varies and which already scale appropriately.
Assess runtime patterns and scheduling opportunities
Distinguishes sustained round-the-clock workloads from intermittent or off-hours-idle workloads, and captures the read-only launch sources for standalone tasks so scheduled work is correctly attributed.
Assess launch-type and capacity-provider economics
Evaluates whether the current Fargate, Fargate Spot, or EC2-backed choice is the cheaper fit given the observed runtime, utilization, and interruption tolerance signals.