[ovs-dev] [PATCH] bonding: Ignore updelay if there is no active slave.

Ben Pfaff blp at nicira.com
Mon Nov 9 22:49:06 UTC 2009


Jesse Gross <jesse at nicira.com> writes:

>> Now that I understand that part, I'm surprised that this patch
>> changes bond_choose_iface().  That's a pretty low-level routine
>> in the flow-translation logic.  I expected that this commit would
>> just change, for example, bond_run(), so that if all the
>> interfaces are disable but at least one has carrier up, it brings
>> one of those that has carrier.  Is there a good reason to do it
>> elsewhere?
>
> The only place that would be affected by moving it to bond_run is the
> enable_slave unixctl, so it's not that important that it be where it
> is.  However, I think that it is simpler and more robust the way it is
> now.  If it was bond_run that was managing the process of bringing up
> a new slave, then the logic would be more spread out between bond_run
> and bond_enable_slave and things like disabling the sending of
> learning packets in certain situations would be less clear.  I looked
> at other options before choosing this one and found that this was the
> simplest because it didn't result in lots of confusing parameters
> being passed around.  The other benefit is that this is more likely to
> just work in the future.  If there are no available slaves when one is
> needed the best one is brought up immediately without any additional
> work.  As an example, the enable_slave command just works.

OK, I can see that you have thought everything out more
thoroughly than I have.  Great.  Push it when you're ready then.




More information about the dev mailing list