Re: [PATCH Version 3 0/2] Add multihostname support for NFSv4.1,2

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

On Fri, Apr 29, 2016 at 03:39:35PM -0400, Anna Schumaker wrote:
> On 04/29/2016 02:05 PM, J. Bruce Fields wrote:
> > On Fri, Apr 29, 2016 at 11:24:30AM -0400, Chuck Lever wrote:
> >>
> >>> On Apr 29, 2016, at 10:46 AM, Steve Dickson <SteveD@redhat.com> wrote:
> >>>
> >>>
> >>>
> >>> On 04/29/2016 10:27 AM, Adamson, Andy wrote:
> >>>> Hi Steve
> >>>>
> >>>> Yes, if we decide to keep the multiple hostname option, then a man page update is required. I don't think we have a consensus on using the multiple hostname mount option as a CLI to express session trunking addresses. Chuck Lever made some good points around not using multiple hostnames:
> >>>>
> >>>> ---- From Chuck: ----
> >>>> - client admins can specify arbitrary hostnames on the command line; hostnames
> >>>> for instance that correspond to some other server.
> >>>>
> >>>> - network conditions can change at anytime, making
> >>>> the original set of trunks lop-sided, or some trunks
> >>>> may become unreachable. What if the server reboots
> >>>> with new i/f's or with one or more removed? The
> >>>> client would likely have to remount in these cases
> >>>> to adapt to network configuration changes.
> >>>>
> >>>> - multiple hostnames could be nailed into
> >>>> /etc/fstab on potentially hundreds of clients. When
> >>>> server or network configuration changes, there would
> >>>> have to be a manual change on all these clients.
> >>>> ----------
> >>>>
> >>>> What do you think? Should we keep the multiple hostname CLI as one method of expressing session trunking addresses?
> >>> I would think so...
> >>
> >> I don't believe a mount CLI is an obvious good choice.
> >>
> >> The client and server should provide some indication
> >> to each other that session trunking is supported. The
> >> server should provide the proper configuration
> >> parameters, which can change even while a client has
> >> mounted the server.
> >>
> >> That's why I favor having the client perform a
> >> GETATTR(fs_locations) on the server's pseudofs, via
> >> which the server provides the correct addresses to
> >> use. The client can poll for changes in the address
> >> list on a regular basis.
> >>
> >> Please, let's automate this instead of having to
> >> nail one more wonky feature into the mount CLI?
> > 
> > Yeah, I guess that makes sense.
> > 
> > My worries from the previous thread were that the fs_locations and
> > fs_locations_info don't *really* give enough information to guarantee
> > that trunking will be an improvement.
> > 
> > But I'd guess those cases aren't common (maybe fs_locations use isn't
> > even that common).  Still, might want a way to opt out.
> > 
> > Maybe it would be worth documenting what the automatic probing does so
> > that servers know how to influence it if desired.
> 
> Would it make sense to add configuration options to /etc/exports to set up how the server handles this?

We have the "replicas=" option that'll allow you to set the fs_locations
list.

If we wanted a "turn off session trunking at the server" switch, that'd
need code.  Maybe it would advertise a different minor identifier per
interface in that case?  Or you could allow finer-grained configuration
of minor numbers from userspace if we wanted to.  We could also
implement fs_locations_info, though it doesn't really seemed designed to
communicate this kind of information.

Not really knowing what people might want to do this, I'm inclined to
leave things as is and wait to see what's needed.  "replicas=" is
probably good enough for now on the server side, and maybe some simple
switch on the client side would be useful.

--b.