Re: [RFC][PATCH 2/4] sunrpc: Use utsnamespaces

["J. Bruce Fields" <bfields@fieldses.org>]

On Tue, Jan 06, 2009 at 07:38:26PM -0500, Trond Myklebust wrote:
> On Tue, 2009-01-06 at 19:26 -0500, J. Bruce Fields wrote:
> > On Tue, Jan 06, 2009 at 07:20:07PM -0500, Trond Myklebust wrote:
> > > As for the NFSv4 clientid, I can't see how you would ever want to use
> > > anything other than the init->utsname(), since the requirement is only
> > > that the clientid string be unique and preserved across reboots. The
> > > server isn't allowed to interpret the contents of the clientid string.
> > > Ditto for the RPCSEC_GSS machine creds that are used to establish the
> > > clientid.
> > 
> > If people eventually expect to be able to, say, migrate a container to
> > another host while using an nfs mount as their storage, then they'd need
> > the name to be that of the container, not of the host.
> 
> Why?
>
> > Obviously we'd also need to ensure the container got its own nfsv4
> > client state, etc., etc., and it sounds like we're far from that.
> 
> Again, why? Are you seriously proposing a plan to transport all NFS and
> locking state directly from one kernel to another?

If people seem to think they can do live process and container
migration:

	http://ols.fedoraproject.org/OLS/Reprints-2008/mirkin-reprint.pdf

then moving the NFS state strikes me as not the greatest of their
troubles.

I'm not volunteering!

--b.