AI companies commonly maintain two storage systems: training data lives in S3, an object-storage protocol that is inexpensive and well suited to massive numbers of small files, while office files use WebDAV, a file-sharing protocol that provides access like a local folder. Copying and synchronizing data between the two systems has long been the industry default.

What This Is

JuiceFS is a China-developed, open-source distributed file system. The core of this update is that its S3 Gateway, which translates S3 requests into file-system operations, and WebDAV now share the same client initialization path while reading the same metadata, cache, and file namespace. Technically, S3 uploads use file-system Rename operations to ensure atomicity, while multipart uploads are merged through CopyFileRange. WebDAV properties are written to file extended attributes, but file-lock status currently exists only in the service process's memory.

Industry View

Supporters see this as a pragmatic choice: moving data between two systems creates latency and consistency risks. There are two main objections. First, in multi-instance deployments, WebDAV file-lock status cannot be synchronized across instances, potentially causing conflicting overwrites in collaborative documents. Second, after an S3 Gateway IAM configuration change, there is a refresh window before the new permissions take effect, requiring additional operational safeguards. Some observers also note that this solves integration on the read side. In high-concurrency AI write scenarios, the bottleneck is not at the protocol layer, so teams should assess write traffic before upgrading their architecture.

Impact on Regular People

  • For enterprise IT: Maintenance should decrease, but file-lock synchronization and permission refresh windows need separate validation. This cannot directly replace an existing dual-system architecture.
  • For individual employees: Colleagues working in AI training or data-related roles may find that training data and shared-drive files now come from the same underlying storage.
  • For the consumer market: End users will barely notice any change. This is a backend infrastructure optimization.