On Dec 20, Han Zhou wrote:
> On Mon, Dec 20, 2021 at 10:32 AM Lorenzo Bianconi <
> [email protected]> wrote:
> >
> > > On Mon, Dec 20, 2021 at 5:53 AM Lorenzo Bianconi <
> > > [email protected]> wrote:
> > > >
> > > > Remove global state variable and move move inc-proc code in an
> isolated
> > > > strucuture. This is a preliminary patch to add the capability to run
> > > > multiple inc-proc engines.
> > > >
> > >
> > > Thanks Lorenzo! Could you tell more about the use case when multiple
> > > inc-proc engines are required?
> >
> > Hi Han,
> >
> > I will rely on this patch to add an incremental processing engine for ovn
> > meters since in the current codebase ovs meters are updated when an ovn
> meter
> > is added or deleted but when it is updated.
> 
> Ok, thanks for the information. For meters, there are different types, ones
> directly specified in lflows, and ones referencing SB meter table records.
> Are you talking about the SB meter records handling? For either case, it
> seems that we need to handle lflow together. Are we sure it is better to
> have a separate engine than handling in the existing engine? What would be
> the engine nodes and dependencies for the new engine?

I guess this patch is orthogonal to the ovn-meter case, it would be useful
even for northd incremental processing started by Mark Gray. What do you think?

Regarding the ovn-meter use case I am referencing to the SB meter table but we
want to just update the meter bands w/o any change to the flow using the meter
(the lflow just refers to the meter UUID, not to the bands).
If we use a single engine we would be stopped by a force recompute before
processing the SB meter table, losing in this case the new band info.
Am I missing something?

Regards,
Lorenzo

> 
> Thanks,
> Han
> 
> >
> > Regards,
> > Lorenzo
> >
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to