-
Notifications
You must be signed in to change notification settings - Fork 80
Open
Labels
Description
Right now locations store either the entire dataset or just certain directories (and the latter use case is somewhat ad-hoc). It would be nice to have caching-only locations, which allow you to mount the filesystem locally in a data center with a small amount of storage, and which would fetch data only on-demand (which is already implemented to make sure that we can always read files from all locations), and in a caching fashion.
One possible difficulty here is what to do with the metadata. We currently store all of the metadata in every location, and it would be somewhat tricky to not do so. Now, the metadata is ~3000x smaller than the file contents, so it's not so bad, but it is a factor to consider.