On 4/27/23 19:54, Han Zhou wrote:
>> (Random thought) NB could have an API (ChassisDeprovisionRequest?)
>> that would be used by CMS to request cleanup for a chassis by name.
>> Northd could then update the object with the status of the request, or
>> delete it once it's processed.
>>
> This may work, but I'd avoid any imperative approach if possible. NB (and
> also SB) should just maintain the desired states and let the
> controllers/daemons to converge.
> A "request" may be failed/timeout/invalid/out-of-date, which needs to be
> transactional and will be relatively complex to handle.
> 
> If the direct connection to SB is really the only concern, I think the most
> viable approach is still to add a chassis table in NB just for CMS to
> specify the valid chassis that should exist, and ovn-northd (probably with
> a new thread) will take care of the cleanup. This is simple and scalable.

Do we really need a new thread?  We don't really expect chassis related
changes to happen often.

> The only question is whether it's worth making this change.
> 

The fact that it's a relatively simple patch is an argument for making
the change IMO.

Regards,
Dumitru

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to