GitHub Availability and Infrastructure Migration Report: August-September 2026
GitHub details infrastructure resilience efforts, regional migration progress, and recent production service incident resolutions.
1Infrastructure Migration Progress
GitHub continues migrating production workloads to Azure, specifically targeting the Central US region. By October 7, 120 of 160 services were deployed. Azure currently handles 65.4% of core application read traffic and 80.2% of stream-processing traffic. However, uneven capacity across availability zones forced a temporary deployment pause on September 11.
2Database Performance Optimizations
Engineering teams reduced shared-database pressure by moving 24 authentication-core tables, removing one million queries per second from replicas. Additionally, deleting 980 million unnecessary edit-history rows from the issues and pull request cluster reclaimed 1.8 terabytes of storage. Weekly deadlocks have fallen by approximately 77% since their June peak levels.
3Resilience and Overload Control
New overload controls now protect high-priority API and web workflows by rejecting lower-priority requests during saturation. GitHub Actions gained significant operating headroom, with production changes reducing Redis operations by 28% and relative CPU usage by 30%. An infrastructure cache also improved response times for primary inventory requests by 96%.
4September 13 Incident Analysis
A background data-cleanup job caused a major incident by exhausting primary database connections. The job lacked awareness of primary server load, leading to request failures across dependent services. The impact affected 96% of web-based issue creation. Recovery was delayed because request handlers lacked quick timeouts and retry loops saturated the database.
5Copilot Code Review Disruption
On September 28, a service change caused Copilot code reviews to fail silently for many customers. The system incorrectly reported successful completion while failing to deliver full review copies. Monitoring gaps delayed detection until reviews remained incomplete for 30 minutes. The team resolved the issue by reverting the faulty deployment.
6Failure Containment Strategies
GitHub is enhancing failure containment for pull requests, allowing half of sampled traffic to maintain limited functionality during shared dependency failures. Database traffic routing now automatically adds capacity during spikes; one group successfully scaled from 14 to 19 instances on September 23. These mechanisms aim to isolate faults within specific services.
7Operational Lessons Learned
Recent outages, including the August 6 and 17 events, highlight that infrastructure improvements must be validated under real-world conditions. The organization maintains a strict priority hierarchy: availability first, then capacity, and finally features. Future efforts focus on ensuring background processes respect database capacity limits and improving automated service recovery behaviors.
8Future Reliability Roadmap
The team aims to unlock all available capacity in the Central US region by the end of October. Ongoing work includes strengthening Copilot delivery safeguards and refining deployment processes to identify issues faster. These technical investments are critical to preventing the recurrence of connection-limit exhaustion and silent service failures.