Why You Can't Just Mount an S3 Bucket and Start Editing

Short summary

Object storage like S3 is not a filesystem — it has no directory tree, stat(), or POSIX partial reads — so mounting it for media editing fails under random-access workloads like timeline scrubbing. The solution involves stub files that report correct metadata without local data, plus read interception layers that fetch only the byte ranges an application requests. FUSE and Linux's fanotify API offer different tradeoffs for implementing this partial-hydration layer.

• •S3 and other object stores lack POSIX semantics — naive FUSE mounts turn each seek into a separate HTTP request
• •Stub files report correct metadata without local data, enabling partial hydration of specific byte ranges
• •FUSE intermediates all syscalls with overhead; fanotify offers a lighter alternative by subscribing to specific filesystem events

Generated with AI, which can make mistakes.