Use this checklist to reduce your monthly bill without sacrificing reliability.
Compute
Use Fargate Spot for non-critical workloads. Spot provides 70%
savings. Use it for development, staging, and worker tasks that tolerate
interruption.
Right-size Fargate tasks. Monitor CPU and memory utilization via
CloudWatch. If utilization stays below 30%, step down to a smaller
configuration.
Consider Fargate Savings Plans. A 1-year Compute Savings Plan
reduces on-demand costs by up to 50% for predictable baseline workloads.
Enable auto-scaling. Scale down during off-peak hours rather than
running peak capacity 24/7.
Networking
Replace NAT Gateway with VPC endpoints. If you only need access to
SSM, SQS, ECR, and CloudWatch Logs, two or three VPC endpoints cost less
than a NAT Gateway.
Use a single AZ for development. VPC endpoint costs double with two
AZs. Use one AZ for non-production environments.
Deploy in a public subnet for dev/test. Eliminates NAT and VPC
endpoint costs entirely.
Observability
Set log retention to the minimum needed. 7 days for dev, 30 days
for staging, 90 days for production.
Use metric filters instead of custom metrics. CloudWatch metric
filters extract metrics from logs at no additional cost beyond ingestion.
Sample X-Ray traces. Use 1–10% sampling in production. Full
sampling is rarely necessary and costs $5 per million traces.
Export old logs to S3. For retention beyond 90 days, export to S3
at $0.023/GB-month instead of keeping them in CloudWatch at $0.03/GB-month.
API Layer
Use HTTP API instead of REST API. HTTP API costs $1/M requests
versus $3.50/M for REST API. Switch to REST only when you need usage plans
or request validation.
Skip ALB for low-traffic deployments. At fewer than 500K requests
per day, HTTP API Gateway is significantly cheaper than the ALB fixed cost.
Storage
Review EFS throughput mode. Elastic throughput is the best default.
Never provision throughput for a config-only mount.