GitHub Went Dark for 8 Hours, and the Cause Wasn't Even a Bug

Keypoints:

• • GitHub suffered a nearly eight-hour outage on August 17, lasting 7 hours and 47 minutes
  • Web and API error rates hit roughly 20%, while file downloads failed at closer to 50%
  • The root cause was a capacity shortage, not a code or configuration change
  • A separate Visual Studio Code bug triggered a wave of retries that worsened the outage
  • GitHub's monthly commit volume doubled from 1.4 billion to 2.9 billion between April and August 2026

On August 17, GitHub suffered a nearly eight-hour outage that disrupted GitHub.com, Issues, Pull Requests, Actions, Copilot, and authentication services worldwide. At its worst point, web and API error rates hit roughly 20%, while archive and raw-content downloads failed at closer to 50%, leaving millions of developers unable to push code, open pull requests, or run CI pipelines for most of a workday.

GitHub's own postmortem found the root cause wasn't a bad code deployment or a configuration change, which is usually the first thing engineers suspect when a major platform goes down. It was a pure capacity shortage: record-high traffic overwhelmed load balancers in the company's Central US data center, and a piece of internal networking infrastructure called an Istio sidecar hit its processing limit without triggering the auto-scaling system that was supposed to catch it. That gap between actual load and the system's ability to detect it is what turned a traffic spike into a multi-hour outage.

A separate bug in Visual Studio Code then made recovery slower than it needed to be. When Copilot authentication started failing, the editor kept retrying automatically at a massive scale, pushing traffic to the token service from its usual 7,000 to 9,000 requests per second up to somewhere between 70,000 and 100,000 requests per second. That retry storm piled even more load onto systems that were already struggling to recover, which is part of why Actions and Copilot stayed degraded well after most other services came back.

The scale behind the failure is what makes this worth understanding. GitHub's monthly commit volume doubled from 1.4 billion in April to 2.9 billion by August, driven heavily by AI coding tools generating far more commits than human developers typically would on their own.

 GitHub's CTO, Vlad Fedorov, acknowledged as much directly, saying rapid usage growth explains the load on the system but doesn't excuse the outage, a distinction that's becoming increasingly relevant as AI tools push traffic patterns past what infrastructure built just a year or two ago was ever designed to handle.