GitHub's 8-Hour Outage Revealed How Auth Retries Can Worsen Failures

On August 17, 2026, GitHub experienced a major outage lasting nearly eight hours, affecting APIs, Actions, pull requests, webhooks, and Copilot, with error rates reaching around 20% on web and API traffic. Engineers identified that authentication-token retries were amplifying the problem rather than helping, creating a feedback loop where repeated retry attempts further overloaded the already struggling authentication service. To allow recovery, GitHub partially disabled these retries — a scenario known as a retry storm or metastable failure. By contrast, companies like Microsoft and Google have long-designed authentication systems with isolated fallback infrastructure, disciplined retry logic using exponential backoff, and cached token validation to reduce live dependencies. GitHub's historically tighter service coupling made it more vulnerable to this cascading effect, and the outage highlighted the architectural trade-offs the platform still faces despite being owned by Microsoft.