not receiving LACP PDUs should suspend the link, this behavior should
be specified as per LACP standard. LACP PDUs are used for 2 purposes,
forming and ensuring that the bond membership is correct, as well as
implicit keepalive so as to ensure the bond member is still active and
therefore part of the bond.
thx
On Sat, Aug 6, 2011 at 4:25 PM, Ethan Jackson <et...@nicira.com> wrote:
> Ah, that's a very good point. I suppose the correct thing to do would
> be fall back to active-backup as you suggest. If we are going to do
> that for LACP we may as well do it for the standard SLB case as well.
> I think for now we should just warn as implemented in the current
> version of the patch. I wan't to look up what linux bonding and
> hardware implementations do in the case of failed LACP negotiations.
> I'd prefer to mimic them.
>
> Sanjay, do you know what hardware switches usually do when they fail
> to receive LACP PDUs on any of the slaves in a bond? Do they just
> disable the link altogether, or just assume the bond is working
> correctly, or fall back to an active-backup configuration?
>
> Ethan
>
>
> On Sat, Aug 6, 2011 at 08:55, Ben Pfaff <b...@nicira.com> wrote:
>> How about the case where LACP fails to negotiate? Should we fall back
>> to active-backup if flood_vlans are configured?
>>
>> On Fri, Aug 05, 2011 at 10:55:29PM -0700, Ethan Jackson wrote:
>>> Personally I think it's better to just fail. This will force someone
>>> configuring a bond to notice the problem and deal with it. I'm
>>> worried about people deploying active-backup bond's in production when
>>> they really intend slb bonds.
>>>
>>> That said, it's mostly an aesthetic issue and I don't feel
>>> particularly strongly about it.
>>>
>>> Ethan
>>>
>>> On Fri, Aug 5, 2011 at 21:17, Ben Pfaff <b...@nicira.com> wrote:
>>> > On Fri, Aug 05, 2011 at 06:22:07PM -0700, Ethan Jackson wrote:
>>> >> This seems fine, I would go slightly further though.
>>> >>
>>> >>
>>> >> > + ? ?if (s->balance == BM_SLB && port->bridge->cfg->n_flood_vlans) {
>>> >> > + ? ? ? ?VLOG_WARN("port %s: SLB bonds are incompatible with
>>> >> > flood_vlans, "
>>> >> > + ? ? ? ? ? ? ? ? ?"please use another bond type or disable
>>> >> > flood_vlans",
>>> >> > + ? ? ? ? ? ? ? ? ?port->name);
>>> >> > + ? ?}
>>> >>
>>> >> I would change this warning to an error, and actually fail to create
>>> >> the port in this case. ?I'm afraid people are going to ignore the
>>> >> warning in the log. ?If you disagree, go ahead and merge.
>>> >
>>> > What do you think of forcing the bond to active-backup mode? ?Then it
>>> > will still work, at least.
>>> >
>>
>
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
