On Thu, Feb 02, 2012 at 11:42:50AM -0800, Justin Pettit wrote:
>
> On Feb 1, 2012, at 5:19 PM, Ben Pfaff wrote:
>
> > On Wed, Feb 01, 2012 at 04:57:35PM -0800, Justin Pettit wrote:
> >> On Feb 1, 2012, at 2:58 PM, Ben Pfaff wrote:
> >>
> >>> + xvu = row.external_ids.get("xs-vif-uuid")
> >>> + if xvu:
> >>> + iface_id = (new_iface_ids.get(xvu)
> >>> + or iface_ids.get(xvu)
> >>> + or get_iface_id(row.name, xvu))
> >>
> >> Doesn't this prefer an old iface_id (in "iface_ids") over a new one
> >> (through get_iface_id())? Is that intended behavior? Is your
> >> concern about making a bunch of XAPI checks to get
> >> "nicira-iface-id"?
> >
> > The intent is that the new caching behavior should be the same as the
> > old caching behavior. In either case, after we translate a
> > xs-vif-uuid to an iface-id once, we cache the translation until the
> > xs-vif-uuid is no longer in use. That's the intention, anyhow. Do
> > you see a change?
>
> It was a general comment, but it seems like the current behavior has
> been okay, so no need to change it--especially since those XAPI
> calls aren't necessarily cheap.
>
> This code does change things slightly because if "xs-vif-uuid" goes
> away, the "iface-id" will, too. However, that seems reasonable to
> me.
Thanks.
I did some basic testing on a XenServer and pushed this.
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
