GitHub Availability Report: September 2026 Details Azure Migration Progress and Incident Post-Mortems

GitHub’s September 2026 availability report details progress in reducing shared-database pressure and expanding Azure migration, while acknowledging significant incidents and a missed migration target. The company reduced load on its oldest shared database by moving 24 authentication-core tables, removing approximately one million queries per second from its replicas. Additionally, cleanup of 980 million unnecessary edit-history rows freed an estimated 1.8 terabytes of storage, and weekly deadlocks fell approximately 77% from their June peak.

Azure Migration and Capacity

By October 7, 120 of 160 production services were deployed in Central US, and 13 of 51 primary MySQL databases had moved. Azure handled 65.4% of read traffic for GitHub’s core application, 48.3% of Git read traffic, and 80.2% of stream-processing traffic. However, GitHub paused Central US deployments on September 11 due to uneven capacity across availability zones. Although permanent ramps resumed on September 18, the company missed its aggressive September 30 goal to deploy every production service in the region.

Resilience and Performance Improvements

GitHub implemented overload controls and failure containment measures. Half of sampled pull request traffic can now continue with limited functionality when a shared dependency fails. Database traffic routing automatically added capacity during spikes, with one group expanding from 14 to 19 instances during a September 23 increase in activity. For GitHub Actions, production changes reduced measured Redis operations by approximately 28% and relative CPU by approximately 30%. A separate infrastructure cache reduced Redis bandwidth by 84% and median response time for the primary inventory request by 96%.

Incident Post-Mortems

The report includes post-mortems for two incidents:

• September 13 Incident: A background data-cleanup job increased load on a shared database, causing the primary server to run out of available connections. This impacted ~96% of customer requests creating issues through the web interface and 90% of new signups. The safeguard monitored replica lag but failed to account for incoming load on the primary server.
• September 28 Incident: A change to the service running Copilot code review caused full review copies to be sent only for some customers, leaving others in a reviewing state without results. This impacted Copilot code review across github.com and GHEC data-residency regions for approximately 2 hours and 45 minutes. The issue was resolved by reverting the change.