'We let you down': GitHub pledges to scale up before developers give up

GitHub has been experiencing a series of outages in recent weeks, prompting the company's Chief Technology Officer, Vladimir Fedorov, to pledge improvements to the platform's reliability. The most recent outage occurred on August 17, lasting 7 hours and 47 minutes and affecting a wide range of services including Actions, pull requests, issues, Copilot, and APIs. In a statement, Fedorov acknowledged that the company "let you down" during this incident.

The August 17 outage followed another one on August 6, which also impacted the Actions service. Over time, GitHub has acknowledged issues with platform stability, having reported as far back as April that monthly commits had surged to 1.4 billion, and now exceeding 2.9 billion. The company has also seen a significant increase in the number of new repositories and merged pull requests processed monthly. Microsoft Azure currently handles around 58 percent of GitHub's platform load and manages half of all Git operations.

Fedorov outlined GitHub's strategy to address the scaling weaknesses that led to the recent outages. The company aims to implement an architecture that can linearly scale read capacity with the number of readers, potentially enabling unlimited reads. This new architecture will be rolled out gradually, initially targeting the largest repositories known as monorepos.

In the interim, GitHub is working to isolate critical systems to minimize the impact of future failures, tighten retry limits, and implement alerts to detect early signs of traffic spikes. The company has emphasized that previous outages were not caused by recent code or configuration changes, but rather were issues that had existed within the platform for some time.

Fedorov stressed the importance of addressing these underlying problems to restore developer trust, stating that his team is "responsible" for fixing these issues and regaining the confidence of GitHub's user community.