- SVFS supports authentication versions 1/2/3, is stable and fully connected.
- SVFS won't play with unsafe pointers all the time, avoiding segfaults.
- SVFS will make read/write operations atomic : it won't return to the caller until data has been read/written from/to the network.
- SVFS doesn't use temporary files. An interesting consequence is that you always get the actual operation progress with your favorite tools and you don't need any extra local space.
- An SVFS file is always a stream on the network. So you can play media content directly as a stream, seek to a specific part in the file and so on. These features also work whith encrypted content.
This is not the role of a network filesystem to chose how your data should be accessed : it should be consistent across operations. If you are looking for local speed rates, then this means you are looking for a local filesystem and svfs is no more than an easier way to achieve synchronization between both. In this case you should rely on an appropriate, journalized, battle-hardened local filesystem. This is also where you should manage ownership, permissions and other ACL/extended attributes information, relatively to your local users and groups.
By default, rsync
works with blocks. SVFS abstracts object storage.
You need to tell rsync
to work with entire files :
- mount your svfs device with
attr
option rsync -rtW --inplace --progress <source> <destination>
- profit
Openstack Swift does not handle file ownership or permissions in a way which is familiar to filesystems. It is actually relying on some form of ACLs which can't be a good match for filesystem permissions. Also, it has little sense to set file ownership or permissions over object storage : there's no such thing as uid/gid when you store an object : it's your object. These informations come from your local filesystem while you are storing data on a remote location. Given that, svfs doesn't support setting this information per file but provides per mountpoint options.
Openstack Swift generates and stores modification time so that users can't change
it. In svfs we use metadata to store this information if you supply a specific
mount option (attr
). This has a performance impact since fetching
metadata is only possible by requesting extra details on each node.
So if you want the best performance, you shouldn't use it. Note that mtime
can't be set on a directory/container/mountpoint because every change occuring
within one of this node would trigger too many requests. Usually that's not an
issue for backup tools as they don't rely on directory metas.
Openstack Swift can support directories as standard objects when they are uploaded without content. However, most of the time swift clients will not proceed this way. In this case, deleting an object will mean deleting all empty intermediate directories within the object path as well.
Yes, pick the latest pkg, install it with ruby and osxfuse and there you go !
You just have to follow unit tests guidelines.