Enhancing auto scaling resilience by tracking worker utilization metrics

This article explains how to improve auto scaling resilience by tracking worker utilization instead of system resources like CPU.

• Traditional CPU-based scaling fails when resource consumption doesn't correlate with application capacity
• Worker utilization measures ratio of active work to available processing capacity
• Formula: total work (backlog + in-flight) divided by total workers
• Works across mixed instance types, variable latencies, and evolving applications
• Implement using CloudWatch metric math in target tracking scaling policies
• Recommended starting target utilization value is 0.7
• Approach works for SQS-based, ECS, and synchronous API workloads
• Enables independent optimization of availability and cost without policy changes

Worker utilization-based scaling provides more resilient auto scaling that automatically adapts to capacity constraints regardless of instance type or application changes.