In the case of JuiceFS the file is split into chunks and those are the objects uploaded to S3 - treating it like a high latency block device. The downside is t

In the case of JuiceFS the file is split into chunks and those are the objects uploaded to S3 - treating it like a high latency block device. The downside is that metadata here isn't just a cache, it is necessary to operate in this fashion and must be backed up.