Re: virtiofsd: Where should it live?

[Dr. David Alan Gilbert]

* Daniel P. Berrangé (address@hidden) wrote:
> On Mon, Nov 25, 2019 at 06:50:21PM +0000, Dr. David Alan Gilbert wrote:
> > Hi,
> >   There's been quite a bit of discussion about where virtiofsd, our
> > implemenation of a virtiofs daemon, should live.  I'd like to get
> > this settled now, because I'd like to tidy it up for the next
> > qemu cycle.
> > 
> > For reference it's based on qemu's livhost-user+chunks of libfuse.
> > It can't live in libfuse because we change enough of the library
> > to break their ABI.  It's C, and we've got ~100 patches - which
> > we can split into about 3 chunks.
> > 
> > Some suggestions so far:
> >   a) In contrib
> >      This is my current working assumption; the main objection is it's
> >      a bit big and pulls in a chunk of libfuse.
> 
> My main objection to 'contrib/' is actually the perceived notions
> about what the contrib directory is for. When I see 'contrib/'
> code in either QEMU, or other open source projects, my general
> impression is that this is largely unsupported code which is just
> there as it might be interesting to someone, and doesn't typically
> get much ongoing dev attention.
> 
> Parts that are fully supported & actively developed by projects
> usually live elsewhere like a src/ or lib/ or tools/ directory.
> 
> This has kind of been the case with QEMU historically, with
> the vhost-user-blk, vhost-user-scsi not being real production
> quality implementations. Rather they are just technology demos
> to show what you might do.   vhost-user-gpu/input blurred this
> boundary a bit as they're more supported tools, and so I'd
> argue contrib/ probably wasn't the right place for them either
> in hindsight.
> 
> virtiofsd is definitely different as it is intended to be a
> fully production quality supported tool with active dev into
> the future IIUC.
> 
> IOW, if we did decide we want it in QEMU, then instead of
> '$GIT/contrib/virtiofsd', I'd prefer to see '$GIT/virtiofsd'.

I'm not sure it deserves a new top level for such a specific tool.

> >   b) In a submodule
> > 
> >   c) Just separate
> > 
> > Your suggestions/ideas please.  My preference is (a).
> 
> What I'm wondering is just how much sharing / overlap of code and concepts
> and community operation there is going otbetween QEMU and virtiofsd. From
> the tech POV, IIUC, the main blocker it would need to be in QEMU is because
> it links to libvhost-user and we've not declared that to be a stable API
> for 3rd party linking.
> 
> Personally I'm always biased towards self-contained apps being in their
> own repositories, rather than bundling too much stuff into one repo. You
> can see that in the way we've created independant git repos for any libvirt
> module that didn't need to be part of the main libvirt.git.
> 
> To me the key benefit this gives is flexibility in approach. ie the app
> doesn't need to blindly follow every precedent that QEMU has set. It
> can instead take the most appropriate path for its needs. For example...
> 
> It could use meson for its build system already. This would be good as
> builds will be done in a matter of seconds. For contributors it would
> be a much less daunting project to join as it wouldn't be lost in the
> firehose of other non-virtiofsd contributions on qemu-devel.
> 
> It doesn't have to follow QEMU's 3-times a year release model, with 6
> week long freeze periods. It can be more agile releasing 6 times a year
> with 1 week freezes if desired, I personally think tihs would be quite
> desirable for a young project like virtiofsd which is evolving rapidly
> as it would get new work available to users much more rapidly.

Form virtiofsd's point of view I'm not that worried about the release
cycle;  Given that features have to go through virtio standardisation,
the release ycle is unlikely to be a bottleneck.

> It doesn't have to follow QEMU's API stability & deprecation policies.
> It could be more flexible in taking non-compatible changes, which again
> may be valuable for a young rapidly evolving app.
> 
> 
> 
> Anyway to be clear, I'm not a contributor to virtiofsd, nor likely to
> be one in the future, so just consider this a personal POV. From QEMU's
> POV I don't think it'll matter whether virtiofsd in or out of QEMU git.
> It is more about the impact & burden QEMU's dev process & standards might
> impose on virtiofsd itself.

As a qemu contributor, your opinion is welcome!  No need to sit on the
fence.

> I'm fine with whatever option above is chosen, with the only caveat
> being that if its in qemu.git, I don't think it belongs under contrib/
> it should be a top level dir of its own.
> 

Dave

> Regards,
> Daniel
> -- 
> |: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
> |: https://libvirt.org         -o-            https://fstop138.berrange.com :|
> |: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK