[Gluster-devel] RFC: FUSE kernel features to be adopted by GlusterFS
chenk at redhat.com
Thu Nov 9 16:25:19 UTC 2017
On Thu, Nov 9, 2017 at 1:33 PM, Niels de Vos <ndevos at redhat.com> wrote:
> On Thu, Nov 09, 2017 at 01:04:48AM +0100, Csaba Henk wrote:
> > Hello list,
> > here is an overview of recent additions to the
> > FUSE protocol that are not yet supported by
> > the GlusterFS FUSE implementation and
> > what sh/could be done about them.
> > https://docs.google.com/document/d/1VGxMC7Db7Rdd19VXlBe0tI-frdjICJdzxx4EsntuPDU/edit
> > Comments on the doc or in this thread are most
> > warmly welcome.
> I would really love to see us use libfuse (not a forked or bundled
> version) with libgfapi. That should make it easier for us to follow
> upstream enhancements. A proof of concept was once written by Olegsandr
> Natalenko and is available at https://github.com/gluster/xglfs
> Maintaining three access xlators/protocols (FUSE, gNFS, gfapi) adds a
> lot of overhead. Having a single one (gfapi) with additional projects
> for the other 'bindings' to protocols would be much better to maintain.
> There was a bug or issue for this somewhere... I'm not able to find it
I harbored an unresolved unease about depending on libfuse, kind of a
memory of a thought that it is does not do good to us to have it, but
forgot about the reason. Recently I managed to pin it down - it was from
the pre-acquisition days, when on customers' setups libfuse was either
unavailable or only a fairly out of date version, which was the cause
of many headaches.
So - nothing inherent to libfuse and nothing that would be relevant as
But then, let me put it like this: what reason could we have to *not* go
with xglfs in 4.0? It's true that the deliverables are present in libfuse and
libgfapi and it's just a thin glue. As such it seems to be almost devoid
of design concerns, it just bridges the two interfaces in a
straightforward manner. Superficially it seems to be a superior approach
- what snag holds us back to embrace it wholeheartedly?
More information about the Gluster-devel