Enhance distributed cache layer for write requests

[zhitaoli-6]

What would you like to be added:

I understand the distributed cache feature as follows. It is just a cache for read instead of write. File write requests still firstly upload a new object into the object store and then the write is thought as COMMITTED.

Could we enhance the distributed cache layer so that after write is done in this layer, the write can be seen as COMMITTED. Of course, this layer should use replicas/EC for data reliability and availability. The data can be written to object store if they becomes cold.

Why is this needed:
The enhancement will greatly improve file write performance. Actually the new design makes JuiceFS a real tiering file system product.

[jiefenghuang]

Thank you for your feedback. However, this document is for the cloud service, and this requirement can be raised through the official contact channels on the website.

There are currently no plans to support distributed cache in the community edition. Distributed caching can be implemented via multiple mount points through a cloud disk, but at the moment there is no coordination or handling capability across multiple clients for cache updates and awareness. This may be considered in the future.