Re: [fuse-devel] NFS and FUSE

[David S. <ds...@ja...>]

On Thu, Feb 17, 2005 at 10:57:51AM +0100, Miklos Szeredi wrote:
> > Recently we have been experimenting with exporting the FUSE filesystem
> > with NFS, this also appears to work fine but as we want to export it to
> > hundreds of clients we are speculating as to what would happen if we use
> > multiple servers, ie several machines all running the FUSE overlay
> > filesystem and all exporting over NFS, should this work? I can't think
> > of a good reason why it wouldn't.
> 
> Don't get your hopes too high on NFS exporting with current FUSE.  It
> works, but it's far from perfect.  For example it can happen that a
> rarely accessed CWD becomes stale (when the inode is purged from the
> cache on the server), which is only a nuisance in some situations but
> unacceptable in others.
> 
> It is possible to fix this, but requires extending the kernel
> interface, and as yet nobody complained.

What does this entail?  We have been migrating to 2.6 partially to get
NFS ability with FUSE.

What would be the user-visible situation if a CWD went stale?  A
"stale NFS file handle" for the user in that directory?

David