GitHub said an outage on August 17 lasted seven hours and 47 minutes, disrupting github.com, authentication, GitHub Actions, its APIs, pull requests, issues and Copilot worldwide, according to a post-incident writeup published on the GitHub Blog.
The company said a critical infrastructure component in its Central U.S. data center failed to scale as traffic reached unprecedented levels, and that the resulting capacity shortage cascaded into authentication failures across services. During recovery, Copilot clients entered retry loops that added more traffic and delayed full restoration, GitHub said.
GitHub said it was the platform's second significant incident in August, and noted that monthly commits on the platform nearly doubled since April, from 1.4 billion to 2.9 billion, adding pressure to systems built for a lower load. "That growth explains the pressure on our systems. It does not excuse the outage," the company said.
GitHub said it is adding more than 3 million CPU cores and 120 petabytes of storage, expanding its Azure footprint to handle 58% of platform load, scaling its monorepo architecture for more linear read capacity, and adding consistent retry limits and stronger alerting across services.
For teams that build CI/CD around GitHub, the retry-loop detail is the practical takeaway: a client that hammers a struggling service on failure can turn a recoverable capacity problem into a multi-hour outage, which is worth checking in your own integrations regardless of what GitHub fixes on its end.