GitHub, autoscaling, and the component substitution fallacy

GitHub experienced an outage caused by an Istio sidecar pod that reached its concurrency limits and failed to autoscale correctly due to a misconfigured policy that monitored the host service's load but not the sidecar's limits. The incident highlights a common problem where autoscaling policies based on standard metrics like CPU utilization may fail to detect saturation in other components, such as sidecars or thread pools. Since each service behaves differently under load, teams must create custom autoscaling policies and validate them through load testing, adding operational complexity to service ownership.