Nydus is a container image filesystem framework that implements on-demand loading through a content-addressable file system format to accelerate container startup and improve storage efficiency.
The tool addresses the performance and cost limitations of standard OCI container images by implementing chunk-based on-demand loading, where only the data needed for immediate execution is downloaded rather than the entire image upfront. It uses the RAFS format to enhance the OCI image specification, enabling second-level container startup times and millisecond-level function package loading. The system includes chunk-level deduplication across layers and images to reduce storage, transport, and memory consumption, along with end-to-end data integrity verification to detect supply chain attacks at runtime.
Nydus suits organizations running containerized workloads where startup latency and bandwidth efficiency matter significantly, particularly in Kubernetes environments or serverless computing scenarios. The tool integrates with containerd as a snapshotter and supports multiple container runtimes including runC and Kata, making it compatible with existing container infrastructure. It works with various storage backends including registries, object storage, NAS, and P2P services, and maintains compatibility with OCI images while providing native support for eStargz images. Teams should evaluate it when image size and cold startup performance are bottlenecks, though adoption requires integration with containerd-based systems.
The project maintains active development with regular updates to core functionality and ecosystem integration. The codebase is written in Rust, emphasizing memory safety and performance optimization across CPU, memory, and network dimensions. Development includes ongoing work on data analyzability features such as access recording and IO amplification detection, alongside continuous refinement of POSIX compatibility through kernel EROFS and FUSE filesystem implementations.
