Open Stack

Hi Brandon!

Please see inline..




On Wed, May 28, 2014 at 12:01 PM, Brandon Logan <brandon.logan at rackspace.com
> wrote:

> Hi Vijay,
>
> On Tue, 2014-05-27 at 16:27 -0700, Vijay B wrote:
> > Hi Brandon,
> >
> >
> > The current reviews of the schema itself are absolutely valid and
> > necessary, and must go on. However, the place of implementation of
> > this schema needs to be clarified. Rather than make any changes
> > whatsoever to the existing neutron db schema for LBaaS, this new db
> > schema outlined needs to be implemented for a separate LBaaS core
> > service.
> >
> Are you suggesting a separate lbaas database from the neutron database?
> If not, then I could use some clarification. If so, I'd advocate against
> that right now because there's just too many things that would need to
> be changed.  Later, when LBaaS becomes its own service then yeah that
> will need to happen.
>

v> Ok, so as I understand it, in this scheme, there is no new schema or db,
there will be a new set of tables resident in neutron_db schema itself,
alongside legacy lbaas tables. Let's consider a rough view of the
implementation.

Layer 1 - We'll have a new lbaas v3.0 api in neutron, with the current
lbaas service plugin having to support it in addition to the legacy lbaas
extensions that it already supports. We'll need to put in new code anyway
that will process the v3.0 lbaas api no matter what our approach is.
Layer 2 - Management code that will take care of updating the db with
entities in pending_create, then invoking the right provider driver,
choosing/scheduling the plugin drivers or the agent drivers, invoking them,
getting the results, and updating the db.
Layer 3 - The drivers themselves (either plugin drivers (like the HAProxy
namespace driver/Netscaler) or plugin drivers + agent drivers).

While having the new tables sit alongside the legacy tables is one way to
implement the changes, I don't see how this approach leads to a lesser
amount of changes overall. Layer 2 above will be the major place where
changes can be complicated. Also, it will be confusing to have two sets of
lbaas tables in the same schema.

I don't want a separate lbaas database under neutron, and neither do I want
it within neutron. I'm not suggesting that we create a db schema alone, I'm
saying we must build it with the new LBaaS service (just like neutron
itself when it got created). If we don't do this now, we'll end up
reimplementing the logic implemented in neutron for the new lbaas v3.0 API
all over again for the new core LBaaS service. We'd rather do it in the new
one in one effort.

I could be missing some constraints that drive taking the former approach -
please help me understand those - I don't want to be discounting any one
approach without thorough consideration. Right now, it looks to me like
this approach is being taken only to accommodate the HAProxy namespace
driver. Really that is the only driver which seems to be very intertwined
with neutron in the way it uses namespaces.


>
> > What we should be providing in neutron is a switch (a global conf)
> > that can be set to instruct neutron to do one of two things:
> >
> >
> > 1. Use the existing neutron LBaaS API, with the backend being the
> > existing neutron LBaaS db schema. This is the status quo.
> > 2. Use the existing neutron LBaaS API, with the backend being the new
> > LBaaS service. This will invoke calls not to neutron's current LBaaS
> > code at all, rather, it will call into a new set of proxy "backend"
> > code in neutron that will translate the older LBaaS API calls into the
> > newer REST calls serviced by the new LBaaS service, which will write
> > down these details accordingly in its new db schema. As long as the
> > request and response objects to legacy neutron LBaaS calls are
> > preserved as is, there should be no issues. Writing unit tests should
> > also be comparatively more straightforward, and old functional tests
> > can be retained, and newer ones will not clash with legacy code.
> > Legacy code itself will work, having not been touched at all. The
> > blueprint for the db schema that you have referenced
> > (
> https://review.openstack.org/#/c/89903/5/specs/juno/lbaas-api-and-objmodel-improvement.rst)
> should be implemented for this new LBaaS service, post reviews.
> >
> I think the point of this blueprint is to get the API and object model
> less confusing for the Neutron LBaaS service plugin.  I think it's too
> early to create an LBaaS service because we have not yet cleaned up the
> tight integration points between Neutron LBaaS and LBaaS.  Creating a
> new service would require only API interactions between Neutron and this
> LBaaS service, which currently is not possible due to these tight
> integration points.
>

v> The tight integration points between LBaaS and neutron that I see are:

1. The usage of namespaces.
2.  L2 and L3 plumbing within the namespaces and tracking them in the
neutron and lbaas tables,
3. Plugin driver and agent driver scheduling framework/mechanism for LB
drivers.
4. The way drivers directly update the neutron db, which I think makes for
a lack of clear functional demarcation.

Regardless of how we use the new API and db model, will namespaces be used?
If they still need to be supported, the tight integration isn't going to go
anywhere.

This is why I think it will be best to keep the legacy drivers within
neutron, and not give an option to newer deployments to use that
concurrently with the new lbaas core service. The changes will be lesser
this way because we won't touch legacy code.

While I fully understand that we're trying to change the way we look at the
lbaas deployments, and the db object model is an effort towards that, we
need to ensure that the execution is kept elegant as well. For drivers for
lb solutions like f5 or Netscaler, these pain points can be done away with
because they do their own network provisioning and we keep track of them
only to clean up (especially for virtual appliance solutions).

It will however mean that we'll have the additional task of implementing
the new core service before we can use the new db object model. I say we
should just go for that effort and make it happen.



> >
> > The third option would be to turn off neutron LBaaS API, and use the
> > new LBaaS core service directly, but for this we can simply disable
> > neutron lbaas, and don't need a config parameter in neutron.
> >
> >
> > Implementing this db schema within neutron instead will be not just
> > complicated, but a huge effort that will go waste in future once the
> > new LBaaS service is implemented. Also, migration will unnecessarily
> > retain the same steps needed to go from legacy neutron LBaaS to the
> > new core LBaaS service in this approach (twice, in succession) in case
> > for any reason the version goes from legacy neutron LBaaS -> new
> > neutron LBaaS -> new LBaaS core service.
> I totally agree that this is technical debt, but I believe it is the
> best option we have right now since LBaaS needs to live in the Neutron
> code and process because of the tight integration points.  Since this
> object model refactor has been slated for Juno, and these tight
> integration points may or may not be cleaned up by Juno, staying within
> Neutron seems to be the best option right now.
>

v> As I described above, I think the tight integration points are best kept
in legacy code and not carried over to the new implementation. The cleanest
way to do it would be to clearly demarcate neutron related operations
(L2/L3) from LBaaS. But I am keen to get your views on what the difficult
integration points are so that I get a better understanding of the
motivations behind keeping the new tables in neutron.


Regards,
Vijay


>
> >
> > Going forward, the legacy neutron LBaaS API can be deprecated, and the
> > new API that directly contacts the new LBaaS core service can be used.
> >
> >
> > We have discussed the above architecture previously, but outside of
> > the ML, and a draft of the blueprint for this new LBaaS core service
> > is underway, and is a collation of all the discussions among a large
> > number of LBaaS engineers including yourself during the summit - I
> > will be posting it for review within a couple of days, as planned.
> >
> >
> >
> >
> > Regards,
> > Vijay
> >
> >
> > On Tue, May 27, 2014 at 12:32 PM, Brandon Logan
> > <brandon.logan at rackspace.com> wrote:
> >         Referencing this blueprint:
> >
> https://review.openstack.org/#/c/89903/5/specs/juno/lbaas-api-and-objmodel-improvement.rst
> >
> >         Anyone who has suggestions to possible issues or can answer
> >         some of
> >         these questions please respond.
> >
> >
> >         1. LoadBalancer to Listener relationship M:N vs 1:N
> >         The main reason we went with the M:N was so IPv6 could use the
> >         same
> >         listener as IPv4.  However this can be accomplished by the
> >         user just
> >         creating a second listener and pool with the same
> >         configuration.  This
> >         will end up being a bad user experience when the listener and
> >         pool
> >         configuration starts getting complex (adding in TLS, health
> >         monitors,
> >         SNI, etc). A good reason to not do the M:N is because the
> >         logic on might
> >         get complex when dealing with status.  I'd like to get
> >         people's opinions
> >         on this on whether we should do M:N or just 1:N.  Another
> >         option, is to
> >         just implement 1:N right now and later implement the M:N in
> >         another
> >         blueprint if it is decided that the user experience suffers
> >         greatly.
> >
> >         My opinion: I like the idea of leaving it to another blueprint
> >         to
> >         implement.  However, we would need to watch out for any major
> >         architecture changes in the time itis not implemented that
> >         could make
> >         this more difficult than what it needs to be.
> >
> >         2. Pool to Health Monitor relationship 1:N vs 1:1
> >         Currently, I believe this is 1:N however it was suggested to
> >         deprecate
> >         this in favor of 1:1 by Susanne and Kyle agreed.  Are there
> >         any
> >         objections to channging to 1:1?
> >
> >         My opinion: I'm for 1:1 as long as there aren't any major
> >         reasons why
> >         there needs to be 1:N.
> >
> >         3. Does the Pool object need a status field now that it is a
> >         pure
> >         logical object?
> >
> >         My opinion: I don't think it needs the status field.  I think
> >         the
> >         LoadBalancer object may be the only thing that needs a status,
> >         other
> >         than the pool members for health monitoring.  I might be
> >         corrected on
> >         this though.
> >
> >         _______________________________________________
> >         OpenStack-dev mailing list
> >         OpenStack-dev at lists.openstack.org
> >
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >
> >
> > _______________________________________________
> > OpenStack-dev mailing list
> > OpenStack-dev at lists.openstack.org
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
> _______________________________________________
> 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/20140528/d3128640/attachment.html>