Introduction
GitHub recently faced two service interruptions that fell short of our standards, and we sincerely apologize for the disruption they caused. In response, we’ve been executing a sweeping reliability plan — from scaling infrastructure to rethinking core system design. This listicle breaks down the key facts behind GitHub’s ongoing availability improvements, the forces driving unprecedented growth, and the concrete steps we’re taking to ensure a more resilient platform. Dive into each point to understand the challenges and the solutions shaping GitHub’s future.
1. Two Incidents That Forced a Hard Look
The two recent availability incidents were unacceptable. They affected users’ ability to push code, run Actions, and access repositories. For each event, we identified root causes — ranging from database bottlenecks to cascading failures in cache layers. We’ve since implemented targeted fixes and are auditing all dependencies to prevent recurrence. Short-term remedies included session cache redesign and webhook backend migration, while long-term strategies focus on isolating services to contain blast radius. Our commitment: transparency on what went wrong and relentless improvement.
2. A 10X Capacity Plan That Quickly Became 30X
In October 2025, we launched a plan to increase GitHub’s capacity tenfold, aiming for substantial failover and reliability gains. But by February 2026, it was clear the target needed to triple: we now must design for 30 times today’s scale. This revision wasn’t arbitrary — it came directly from observing real traffic patterns and customer usage surges. The takeaway? Our infrastructure must be built to absorb massive, sudden load spikes without degradation.
3. The Real Driver: Agentic Development Workflows
Since the second half of December 2025, the way software is built has accelerated dramatically. Agentic development — where AI agents and automated tools create code, open pull requests, and trigger pipelines — is exploding. Every metric confirms this: repository creation, pull request activity, API calls, automation runs, and large-repository workloads are all climbing rapidly. This isn’t a temporary spike; it’s a fundamental shift in development velocity, and it places new demands on every part of our stack.
4. Why a Single Pull Request Stresses the Entire System
At high scale, a simple pull request doesn’t just touch Git storage. It triggers mergeability checks, branch protection rules, GitHub Actions workflows, search indexing, notifications, permission lookups, webhooks, API calls, background jobs, caches, and databases. Small inefficiencies multiply: queues deepen, cache misses turn into database load, indexes lag, retries amplify traffic, and one slow dependency can ripple across multiple product experiences. This interconnectedness is why we’re attacking coupling at every layer.
5. New Hierarchy of Priorities: Availability First
Our engineering efforts now follow a strict order: availability, then capacity, then new features. This means we’re cutting unnecessary work, improving caching strategies, isolating critical services, and removing single points of failure. We’re also moving performance-sensitive code paths into systems built for high throughput, such as transitioning from Ruby monolith to Go for scalability. The goal: graceful degradation where one subsystem under pressure doesn’t take down the rest.
6. Immediate Actions: Fixing the Biggest Bottlenecks
In the short term, we tackled bottlenecks that appeared faster than expected. We moved webhooks off MySQL to a dedicated backend, redesigned the user session cache, and reworked authentication and authorization flows to slash database load. We also leveraged our Azure migration to spin up significantly more compute resources. These changes bought us breathing room while we worked on deeper structural improvements.
7. Isolating Critical Services to Shrink Blast Radius
Next, we focused on isolating services like Git and GitHub Actions from other workloads. This involved careful dependency analysis and traffic tiering to decide what to separate and how to minimize impact from attacks or surges. Each risk was addressed in order of severity. By reducing shared infrastructure, we limit the blast radius when problems occur — ensuring that an issue in one service doesn’t cascade into a platform-wide outage.
8. Cloud Migration: From Custom Data Centers to Azure and Beyond
We were already migrating out of smaller custom data centers into public cloud when this growth hit. We accelerated the transition and began designing a multi-cloud path. This move gives us elastic capacity, better geographic distribution, and the ability to fail over across providers. Multi-cloud also reduces vendor lock-in and opens up new options for performance and resilience.
9. Distributed Systems Work: Reducing Hidden Coupling
Beyond specific migrations, this is classic distributed systems engineering. We’re reducing hidden coupling between services, limiting the ways failures can propagate, and ensuring GitHub degrades gracefully. Progress is steady, but these two incidents are reminders that there’s still work to do. Every component — from background job queues to API gateways — is being audited for weak points.
10. What’s Next: Continuous Improvement and Transparency
We’re not done. The 30X scale target means we’ll keep investing in infrastructure, tooling, and processes. Expect more frequent communications on reliability, including post‑mortems and performance dashboards. Our engineering teams are dedicated to making GitHub the most dependable platform for collaborative development. We thank you for your patience and trust as we build for a future of agentic, high‑velocity software creation.
Conclusion
GitHub’s availability incidents were a catalyst for deep, systemic changes. From reevaluating capacity targets to isolating critical services and embracing multi‑cloud, every decision is driven by the goal of keeping your work flowing. We’ll continue to share updates as we progress. Your feedback and reports remain invaluable — together we’ll make GitHub more resilient than ever.