Cursor Reinvents Git Storage for Massive Scale

Cursor's new Git architecture uses a cloud-based log to achieve over 300 pushes per second, transforming Git's performance at scale.

The company Cursor has developed a new architecture for Git storage, named Continuity, that fundamentally changes how version control systems can scale. According to a report from InfoQ, this new system uses a write-ahead log (WAL) stored in Amazon's S3 cloud storage as the definitive source of truth for a repository. This design allows local repositories on fast NVMe drives to function as temporary, high-performance caches rather than the primary data store. In synthetic benchmarks using S3 Express One Zone, Cursor reported achieving more than 300 pushes per second and linear scaling for read operations across as many as 100 replicas, a significant performance leap over traditional Git server configurations.

The core innovation lies in applying a well-known database technique—the write-ahead log—to the world of version control. A WAL ensures data integrity by first recording every intended change to a sequential log before the change is actually applied to the main dataset. In this case, instead of writing to a local disk, every Git push is first recorded in S3. This simple but powerful shift decouples two critical operations: ensuring data consistency (writing to the S3 log) and coordinating updates across multiple copies, or replicas, of the repository. Because writing to a highly available service like S3 is fast and reliable, the system can confirm a push almost instantly, while the slower process of updating all the replicas happens in the background without blocking developers.

This approach stands in stark contrast to conventional methods for scaling Git. Many large organizations rely on high-performance network file systems (NFS) to host a central repository, but this often creates a single point of failure and a performance bottleneck as the number of developers and automated CI/CD processes grows. Other solutions involve complex replication schemes that can be difficult to manage and still struggle with write-heavy workloads. By leveraging cloud object storage, a trend already established in the database and data analytics fields, Cursor is applying a modern, cloud-native pattern to a foundational developer tool. This not only addresses performance but also enhances durability and resilience, inheriting the guarantees provided by services like S3.

For CTOs and infrastructure leaders, Continuity presents a compelling new model for managing version control at scale. The performance claims, if validated in real-world scenarios, could eliminate a major source of friction for large engineering organizations, especially those managing massive monorepos or running thousands of automated builds per day. While this architecture is currently specific to Cursor's platform, its underlying principles could influence the future design of other version control systems like GitLab and GitHub Enterprise. The key thing to watch is how this system performs under diverse, real-world workloads beyond synthetic tests and whether the broader industry begins to adopt similar S3-backed log architectures for other stateful systems.

This architecture challenges traditional Git server limitations, where performance bottlenecks are common in large monorepos or high-frequency CI/CD pipelines. By shifting the source of truth to a scalable cloud service like S3, it offers a new blueprint for building highly available and performant version control systems.

For companies with large engineering teams, slow Git operations are a direct drag on developer productivity and release velocity. This model could significantly reduce infrastructure complexity and costs for scaling Git, offering a competitive advantage by enabling faster development cycles at scale.