Re: [patch NetworkManager 0/6] add support for network team devices

[Jiri Pirko <jiri resnulli us>]

Tue, Jul 16, 2013 at 04:32:10PM CEST, dcbw redhat com wrote:
> On Tue, 2013-07-16 at 10:53 +0200, Jiri Pirko wrote:
> > Mon, Jul 15, 2013 at 08:22:27PM CEST, danw gnome org wrote:
> > > On 07/14/2013 11:29 AM, Jiri Pirko wrote:
> > > > teamd config is non-trivial tree structure (JSON). At this point, there is
> > > > really no point to have NM to be aware of the teamd config structure.
> > > > Once NM will need that, this can be changed.
> > >
> > > NM will need to be aware of it once we want to add support for team
> > > connections to nm-connection-editor. We'll probably have to have some
> > > sort of dual-mode interface, where if we recognize all the options, you
> > > get a nice UI with translated strings and proper widgets and such, but
> > > if you need to edit something NM doesn't know about, you can still edit
> > > the JSON directly...
> >
> > Yep. That is the plan for the future.
> >
> > > > No function for that. Teamd will scream when you pass non-valid JSON to
> > > > it. Also will scream in case some configuration is not done correctly.
> > > > At this point, I would make NM non-aware at all of the config structure.
> > > > Just leave it as string and NM to blindly pass it through...
> > >
> > > Same case as above; when the user creates a new team connection, we want
> > > to be able to detect if the configuration is invalid before the user
> > > saves the connection, rather than just accepting it at that point and
> > > then telling them later that they messed up.
> >
> > There's no way to do that other than to execute teamd with that
> > config...
>
> Well, there is; we hardcode that verification logic into NetworkManager
> and then we wait for some "libteam" to expose config verification
> functions that we could use instead.

Well if you hardcore the verification and I add some option to the
config in teamd (which I often do), NM would wrongly tell the user that the
config is invalid.

Introducing such functionality in teamd (libteamd) would be non-trivial
because very often the options are runner-dependent or link-watch
dependent. In fact, I see really no point in doing that, since you get
the verification for free when you just "run" teamd with the config.


> It's just really bad UI to accept random options that we know don't
> work, and not tell the user that they won't work, but instead wait until
> they try to start the interface and then say "oh no, something is
> invalid!".  
>
> Dan
>
> > > On 07/15/2013 01:05 PM, Pavel Simerda wrote:
> > > > How do you store the port config then?
> > >
> > > There's a separate NMSettingTeamPort setting for the ports, like with
> > > bridges.
> >
> > yep
> >
> > > -- Dan
> > _______________________________________________
> > networkmanager-list mailing list
> > networkmanager-list gnome org
> > https://mail.gnome.org/mailman/listinfo/networkmanager-list