maas-devel team mailing list archive

[Blake Rouse <blake.rouse@xxxxxxxxxxxxx>]

On Aug 24, 2015 9:45 PM, "Christian Robottom Reis" <kiko@xxxxxxxxxxxxx>
wrote:
>
> On Mon, Aug 24, 2015 at 03:55:33PM -0400, Blake Rouse wrote:
> > There is never a case where an Interface will have multiple MAC
addresses.
> > One interface has one MAC address. But the MAC address field cannot be
> > unique, this is because a bond interface can use the MAC address of one
of
> > its parents.
>
> Great! That's what I was hoping -- IMV in fact interface is the proper
> abstraction, with MAC being one of its properties. However:
>
> > 3. A "vlan" or "alias" interface should never have a MAC address as this
> > comes from its parent.
>
> In practice will it have a NULL MAC or the same MAC as the parent
> interface?

I am working on this now, it would be best to have the MAC the same as the
parent that way the lease parsing can easily identify the interface to
update its assigned ip address.

>
> > Also making this change will require the Interface to link to a Node.
Which
> > makes more since then the current implementation. A constraint will also
> > need to exist for this change where only a "physical" or "unknown"
> > interface should have a Node. All other interface types "bond", "vlan",
> > "alias" should derive their node from there parents.
>
> Really? I realize it's a bit of a normalization violation, but I think
> it will make the code quite odd to have some Interfaces have Nodes and
> others not!

Yeah after working on this I agree. The node will be set for all
interfaces. If its a child interface the validation will make sure its the
same node as its parents.

> --
> Christian Robottom Reis   | [+1] 612 888 4935    |
http://launchpad.net/~kiko
> Canonical VP Hyperscale   | [+55 16] 9 9112 6430