[openstack-dev] [Neutron] Improving dhcp agent scheduling interface

Armando M. armamig at gmail.com
Wed Nov 5 15:24:28 UTC 2014

Hi Eugene thanks for bringing this up for discussion. My comments inline.

On 5 November 2014 12:07, Eugene Nikanorov <enikanorov at mirantis.com> wrote:

> Hi folks,
> I'd like to raise a discussion kept in irc and in gerrit recently:
> https://review.openstack.org/#/c/131944/
> The intention of the patch is to clean up particular scheduling
> method/interface:
> schedule_network.
> Let me clarify why I think it needs to be done (beside code api
> consistency reasons):
> Scheduling process is ultimately just a two steps:
> 1) choosing appropriate agent for the network
> 2) adding binding between the agent and the network
> To perform those two steps one doesn't need network object, network_id is
> satisfactory for this need.

I would argue that it isn't, actually.

You may need to know the state of the network to make that placement
decision. Just passing the id may cause the scheduling logic to issue an
extra DB query that can be easily avoided if the right interface between
the caller of a scheduler and the scheduler itself was in place. For
instance we cannot fix [1] (as you pointed out) today because the method
only accepts a dict that holds just a partial representation of the
network. If we had the entire DB object we would avoid that and just
passing the id is going in the opposite direction IMO

> However, there is a concern, that having full dict (or full network
> object) could allow us to do more flexible things in step 1 like deciding,
> whether network should be scheduled at all.

That's the whole point of scheduling, is it not? If you are arguing that we
should split the schedule method into two separate steps
(get_me_available_agent and bind_network_to_agent), and make the caller of
the schedule method carry out the two step process by itself, I think it
could be worth exploring that, but at this point I don't believe this is
the right refactoring.

> See the TODO for the reference:


> https://github.com/openstack/neutron/blob/master/neutron/scheduler/dhcp_agent_scheduler.py#L64
> However, this just puts an unnecessary (and actually, incorrect)
> requirement on the caller, to provide the network dict, mainly because
> caller doesn't know what content of the dict the callee (scheduler driver)
> expects.

Why is it incorrect? We should move away from dictionaries and passing
objects so that they can be reused where it makes sense without incurring
in the overhead of re-fetching the object associated to the uuid when
needed. We can even hide the complexity of refreshing the copy of the
object every time it is accessed if needed. With information hiding and
encapsulation we can wrap this logic in one place without scattering it
around everywhere in the code base, like it's done today.

> Currently scheduler is only interested in ID, if there is another
> scheduling driver,

No, the scheduler needs to know about the state of the network to do proper
placement. It's a side-effect of the default scheduling (i.e. random). If
we want to do more intelligent placement we need the state of the network.

> it may now require additional parameters (like list of full subnet dicts)
> in the dict which may or may not be provided by the calling code.
> Instead of making assumptions about what is in the dict, it's better to go
> with simpler and clearer interface that will allow scheduling driver to do
> whatever makes sense to it. In other words: caller provides id, driver
> fetches everything it
> needs using the id. For existing scheduling drivers it's a no-op.

Again, the problem lies with the fact that we're passing dictionaries

> I think l3 scheduling is an example of interface done in the more right
> way; to me it looks clearer and more consistent.

I may argue that the l3 scheduling api is the bad example for the above
mentioned reasons.

> Thanks,
> Eugene.

At this point I am still not convinced by the arguments provided that the
patch 131944 <https://review.openstack.org/#/c/131944/> should go forward
as it is.

> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20141105/0f82343d/attachment.html>

More information about the OpenStack-dev mailing list