What the GitHub Outage Taught Us About Authentication Retries
On August 17, 2026, GitHub went through a rough day. For nearly eight hours, large parts of the platform struggled. API requests, Actions, pull requests, Issues, webhooks, and Copilot all felt the impact. Error rates hit around 20% on web and API traffic. Archive and raw content downloads failed about half the time. The official status page eventually said engineers had identified a problematic…
On August 17, 2026, GitHub experienced a significant disruption, with the platform struggling for nearly eight hours. During this time, API requests, Actions, pull requests, Issues, webhooks, and Copilot all faced issues. Error rates reached around 20% for web and API traffic, and archive and raw content downloads failed in about half of the cases.
The official status page later revealed that engineers had pinpointed a problematic component and implemented fixes, but they also decided to partially disable authentication-token retries, as those retries were exacerbating the issue.
The article highlights the importance of understanding how retries can sometimes create more problems than they solve. Most systems use retries for failed requests as a default measure. However, under stress, retries can lead to a feedback loop, where the extra traffic increases the load on the already struggling service, causing more failures and triggering additional retries.
This is known as a retry storm or metastable failure. In such situations, removing the initial trigger may not be enough to resolve the system's problems, as amplified traffic keeps the service saturated.
Microsoft and Google have developed authentication systems that aim to avoid this retry storm. Microsoft Entra ID uses a separate infrastructure for authentication, with a backup service running if the primary one is degraded. Their client libraries, MSAL, incorporate rate limits and prevent aggressive retry loops. Continuous Access Evaluation allows for longer-lived tokens while supporting near real-time revocation.
Google, on the other hand, favors short-lived access tokens and disciplined client behavior, using exponential backoff with jitter and limiting retries to specific transient errors.
GitHub's architecture, while historically tightly coupled, has been relatively independent since its acquisition by Microsoft. However, moving to Microsoft's identity stack, like Entra ID and MSAL, would have altered the product in ways beyond technical convenience. GitHub supports various user models, including personal accounts, open-source projects, and enterprises, which differ from Entra ID's focus on organizational identity.
MSAL is a client library that assists applications in acquiring tokens from Entra ID but doesn't replace the need for GitHub to manage its own tokens for Git operations, GitHub Apps, Actions, and the API.
The article concludes that while the outage was disruptive, it also shed light on a common systems problem: the retry cascade. Good authentication systems must plan for this by limiting amplification, providing fallback paths, and treating token services as critical infrastructure. GitHub will likely continue enhancing its authentication stack. Meanwhile, the broader community can learn from this experience: when designing retry logic, it's crucial to consider what happens when the dependency is already under pressure.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.