GitHub, autoscaling, and the component substitution fallacy
Summary
The post analyzes a GitHub outage with a focus on autoscaling and Istio sidecar behavior, arguing that autoscaling policies must consider cross-service interactions rather than CPU alone. It introduces the 'component substitution fallacy' and emphasizes examining how traffic, autoscaling, sidecars, retry logic, and load distribution interact to impact reliability. It also recommends load testing and deeper internal incident reviews to prevent similar failures.