Sharing local files with Kubernetes clusters

Sammanfattning: Data science development largely revolves around working in notebooks, and these usually run in a cluster environment. These cloud notebooks do however come with limitations. Support for tools like version control, testing, refactoring and linting is limited, which negatively impacts developer experience and software quality. We propose a networked file system for sharing local files and directories with a remote cloud server. This allows the developer to work in a cloud notebook without having to upload files to an external storage provider, and enables the use of traditional development tools since all files are stored on the local machine. The networked file system is implemented as a FUSE file system and is made available to Kubernetes as a custom volume driver. We show that it is possible to build a widely compatible file sharing solution for any Kubernetes workload. It is easy to use and has reasonable performance.