Reliability engineering

GitHub’s August 17 outage: how retries delayed recovery for hours

One unobserved limit prevented autoscaling, while retries turned latency into fresh load. The incident shows why spare servers are not enough without retry controls and a tested way to work through a platform failure.