[openstack-dev] [Neutron][LBaaS] Should TLS settings for listener be set through separate API/model?
Evgeny Fedoruk
EvgenyF at Radware.com
Thu Jun 26 10:44:10 UTC 2014
Hi guys,
Stephen, I understand your concerns regarding misleading names.
Here are my thoughts:
default_tls_container_id
This name is the same for API and database model and I think this name explains its meaning well.
sni_container_ids(for API) and listenersniassociations (for database table)
These two comes to specify the same thing - TLS container ids list for listener's SNI function,
Still there is a difference: in API it's just a list of IDs contained in listener's API call,
while in database it becomes to specify association between listener ID and TLS container ID in separate database table.
As Brandon posted, database table names in Neutron are derived from data model class names defining them.
Listenersniassociations table name is actually comes from ListenerSNIAssociation class that defines the table.
I understand there is no table for SNI object in neutron schema but I did not think of a better name for this association table name.
It could be named ListenerContainerAssociation but this name does not clarify that this is for SNI and there is no Containers table in Neutron's schema neither)
Calling it ListenerSNIContainerAssociation may be too long..
These are my thoughts but I may miss something, please propose alternative names you think of
Thanks,
Evg
-----Original Message-----
From: Brandon Logan [mailto:brandon.logan at RACKSPACE.COM]
Sent: Wednesday, June 25, 2014 11:00 PM
To: openstack-dev at lists.openstack.org
Subject: Re: [openstack-dev] [Neutron][LBaaS] Should TLS settings for listener be set through separate API/model?
Hi Stephen,
The <entity><entity>associations table name is consistent with the rest of neutron's table names, as is not breaking the table name words up by an underscore. I think this stems from the sqlalchemy models getting the table name for free because of inheriting from a base model that derives the table name based on the model's class name.
However, with markmcclain's blessing the new loadbalancing tables will be prefixed with lbaas_, but the model names will be LoadBalancer, Listener, etc.
I would agree though that since sni will not be a separate table then that will be a bit odd to have an association table's name implying a join of a table that doesn't exist.
Thanks,
Brandon
On Wed, 2014-06-25 at 09:55 -0700, Stephen Balukoff wrote:
> What's the point of putting off a potential name change to the actual
> code (where you're going to see more friction because names in the
> code do not match names in the spec, and this becomes a point where
> confusion can happen). I understand the idea that code may not exactly
> match the spec, but when it's obvious that it should, why use the
> wrong name in the spec?
>
>
> Isn't it more confusing when the API does not match database object
> names when it's clear the API is specifically meant to manipulate
> those database objects?
>
>
> Is that naming convention actually documented anywhere? And why are
> you calling it a 'listenersniassociations'? There is no "SNI" object
> in the database. (IMO, this is a terrible name that needs to be
> re-read three times just to pick out where the word breaks should be!
> As written it looks like "Listeners NI Associations" what the heck is
> an 'NI'?)
>
>
> They say that there are two hard problems in Computer Science:
> * Cache invalidation
> * Naming things
> * Off-by-one errors
>
>
> And far be it from me to pick nits about a name (OK, I guess it's
> isn't that far fetched for me to pick nits. :P ), but it's hard for me
> to imagine a worse name than 'listenersniassocaitions' being
> considered. :P
>
>
> Stephen
>
>
>
>
> On Wed, Jun 25, 2014 at 2:05 AM, Evgeny Fedoruk <EvgenyF at radware.com>
> wrote:
> Hi folks
>
>
>
> Regarding names, there are two types of them: new API
> attributes for REST call, and new column name and table name
> for the database.
>
> When creating listener, 2 new attributes will be added to the
> REST call API:
>
> 1. default_tls_container_id - Barbican TLS container uuid
>
> 2. sni_container_ids (I removed the “_list” part to make
> it shorter) – ordered list of Barbican TLS container uuids
>
> For the database, these will be translated to:
>
> 1. default_tls_container_id- new column for listeners
> table
>
> 2. listenersniassociations (changed it from
> vipsniassociations which is a mistake) – new associations
> table, holding: id(generated), listener_id, TLS_container_id,
> and position(for ordering)
>
> This kind of a name comes to comply current neutron’s table
> name convention, like pollmonitorassociation or
> providerresourceassociation
>
>
>
> I think names may always be an issue for the actual code
> review, the document is just a functional specification
>
> Since new objects model code is not landed yet, naming
> conventions may be changed while implementing this spec.
>
> I will commit the document with all comments addressed and
> mentioned above names.
>
> Please review it and give your feedback, I think we are close
> to complete this one )
>
>
>
> Thanks,
>
> Evg
>
>
>
>
>
>
>
> From: Vijay Venkatachalam
> [mailto:Vijay.Venkatachalam at citrix.com]
> Sent: Wednesday, June 25, 2014 8:34 AM
>
>
> To: OpenStack Development Mailing List (not for usage
> questions)
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should TLS
> settings for listener be set through separate API/model?
>
>
> Thanks for the details Evg!
>
>
>
> I understand there was no TLS settings API originally planned.
>
>
>
> From: Stephen Balukoff [mailto:sbalukoff at bluebox.net]
> Sent: Wednesday, June 25, 2014 5:46 AM
> To: OpenStack Development Mailing List (not for usage
> questions)
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should TLS
> settings for listener be set through separate API/model?
>
>
>
>
> Evgeny--
>
>
>
>
> Two minor nits:
>
>
>
>
>
> * Your spec lists the new SNI related settings 'sni_list' (and
> it contains more than just IDs, so calling it
> 'sni_container_ids_list' is misleading). Please be precise in
> the terms you use, and don't switch them mid discussion. :)
>
>
> * Also, I personally really hate long table names when they're
> unnecessary. "vipsniassociations" isn't mentioned in your spec
> anywhere, and frankly, is a lot worse than "sni_list." I
> personally prefer "SNIPolicies", but I'm also OK with a short
> name like "sni_list".
>
>
>
>
>
> Otherwise I agree with you on all points.
>
>
>
>
>
> Stephen
>
>
>
>
>
>
>
>
>
>
>
>
>
> On Tue, Jun 24, 2014 at 3:26 AM, Evgeny Fedoruk
> <EvgenyF at radware.com> wrote:
>
> Vijay, there is no intension for a new TLS settings
> API.
>
> Creation of a listener with TLS offloading will be
> one-step.
>
>
>
> When tenant creates listener with TERMINATED-HTTPS
> protocol he must supply default_tls_container_id for
> offloading.
>
> Not supplying default TLS container id for offloading
> for TERMINATED-HTTPS listener will raise an error.
>
> SNI list may or may not be supplied by the tenant.
> Default value for SNI certificates list is an empty
> list.
>
>
>
> So listener resource will have another two attributes:
> default_tls_container_id and sni_container_ids_list.
> These are relevant for TERMINATED-HTTPS protocol
> listeners only. In other cases its default value are
> ‘None’ and empty list.
>
> In schema, Default_tls_container_id will be added to
> listener object as another column.
>
> Sni_container_ids_list wil be managed by new table
> “vipsniassociations” which has listener_id,
> container_id, and position (for ordering) columns
>
>
>
> Does it make sense?
>
>
>
> Thanks,
>
> Evg
>
>
>
>
>
>
>
>
>
> From: Vijay Venkatachalam
> [mailto:Vijay.Venkatachalam at citrix.com]
> Sent: Tuesday, June 24, 2014 12:31 PM
>
>
> To: OpenStack Development Mailing List (not for usage
> questions)
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should
> TLS settings for listener be set through separate
> API/model?
>
>
>
>
>
>
> To clarify, the request is for a new TLS settings API
> with “default_tls_container_id” & “sni_list”.
>
>
>
> If there is a new API, then we would have an object
> model reflecting this as a separate entity.
>
>
>
> The tenant would do the following
>
>
>
> 1. Create a listener with TERMINATED_HTTPS
>
> 2. Set the TLS settings for the listener
> using /v2.0/listener/<listenerid>/tlssettings (if at
> all we are having some default values this can be
> reflected here)
>
>
>
> The only good thing is the separation of the TLS
> settings out of the listener API.
>
>
>
> But, I can see 2 downsides
>
> 1. The loadbalancer creation is a 2 step
> procedure
>
> 2. We cannot enforce certificate attachment as
> part of the create of listener.
>
>
>
> If the new API itself has “-1”s then I am perfectly OK
> with the current object model with
> default_tls_container_id in listener table.
>
>
>
> Thanks,
>
> Vijay V.
>
>
>
>
>
> From: Evgeny Fedoruk [mailto:EvgenyF at Radware.com]
> Sent: Tuesday, June 24, 2014 2:19 PM
> To: OpenStack Development Mailing List (not for usage
> questions)
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should
> TLS settings for listener be set through separate
> API/model?
>
>
>
>
> Vipsniassociations table:Line 147 in last patch of the
> document
>
>
>
> From: Vijay Venkatachalam
> [mailto:Vijay.Venkatachalam at citrix.com]
> Sent: Tuesday, June 24, 2014 10:17 AM
> To: OpenStack Development Mailing List (not for usage
> questions)
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should
> TLS settings for listener be set through separate
> API/model?
>
>
>
>
>
>
> >>SNI list is managed by separate entity
>
> What is this entity?
>
>
>
> From: Evgeny Fedoruk [mailto:EvgenyF at Radware.com]
> Sent: Tuesday, June 24, 2014 12:25 PM
> To: OpenStack Development Mailing List (not for usage
> questions)
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should
> TLS settings for listener be set through separate
> API/model?
>
>
>
>
> +1 for option 1. SNI list is managed by separate
> entity, default TLS container is part of a listener
> object. It will have None value when listener does not
> offloads TLS.
>
> Managing another entity for 1:0-1 relationship just
> for future use seems not right to me. Breaking TLS
> settings apart from listener can be done when needed,
> if needed.
>
>
>
> Thanks,
>
> Evg
>
>
>
>
>
> From: Stephen Balukoff [mailto:sbalukoff at bluebox.net]
> Sent: Tuesday, June 24, 2014 4:26 AM
> To: OpenStack Development Mailing List (not for usage
> questions)
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should
> TLS settings for listener be set through separate
> API/model?
>
>
>
> Ok, so we've got opinions on both sides of the
> argument here. I'm actually pretty ambivalent about
> it. Do others have strong opinions on this?
>
>
>
>
> On Mon, Jun 23, 2014 at 6:03 PM, Doug Wiegley
> <dougw at a10networks.com> wrote:
>
> Put me down for being in favor of option 1.
>
>
>
>
>
> A single attribute in a 1:1 relationship? Putting
> that in a new table sounds like premature optimization
> to me; design the database change for the future
> feature when you can see the spec for it.
>
>
>
>
>
> Thanks,
>
>
> Doug
>
>
>
>
>
>
>
>
> From:Stephen Balukoff <sbalukoff at bluebox.net>
> Reply-To: "OpenStack Development Mailing List (not for
> usage questions)" <openstack-dev at lists.openstack.org>
> Date: Monday, June 23, 2014 at 5:25 PM
> To: "OpenStack Development Mailing List (not for usage
> questions)" <openstack-dev at lists.openstack.org>
> Subject: Re: [openstack-dev] [Neutron][LBaaS] Should
> TLS settings for listener be set through separate
> API/model?
>
>
>
>
>
> Also to add to pros for 2:
>
>
>
>
> * Keeping the TLS stuff contained to its own objects
> means we can have separate development resources on
> each and not worry as much about overlapping domains.
> (TLS-related knowledge and knowledge of dealing with
> TCP / UDP listeners are separate knowledge domains. Or
> at least, the former is a more specialized subset of
> the latter.)
>
>
>
>
>
> Note that what we're proposing means there's
> essentially a 1:0-1 relationship between Listener and
> this new yet-to-be-named object. (0 in case the
> Listener is not terminating TLS.)
>
>
>
>
>
> Stephen
>
>
>
>
> On Mon, Jun 23, 2014 at 3:38 PM, Brandon Logan
> <brandon.logan at rackspace.com> wrote:
>
> Whoops, [Neutron][LBaaS] got taken out of the subject
> line here.
> Putting it back in.
>
>
> On Mon, 2014-06-23 at 21:10 +0000, Brandon Logan
> wrote:
> > Okay so we've talked a bit about this in IRC and now
> I'm sending this
> > out as an update. Here are the options with pros
> and cons that have
> > come from that discussion.
> >
> > 1) default_certificate_id is an attribute of the
> Listener object.
> >
> > Pros:
> > -No extra entity needed
> >
> > Cons:
> > -May bloat Listener object when more attributes are
> needed for only TLS
> > termination. Sounds like TLS version and cipher
> selection will be
> > needed attributes in the future.
> >
> >
> > 2) A separate TLS Entity is created that is
> referenced by the Listener
> > object. This entity at first may only contain a
> certificate_id that
> > references barbican. Name and description can be
> allowed as well.
> >
> > Pros:
> > -TLS domain specific attributes contained in its own
> entity
> > -Future attributes would just be added to this
> entity and not bloat the
> > Listener object.
> >
> > Cons:
> > -It's another entity
> >
> > In IRC we (sbalukoff, myself) seemed to agree option
> 2 is right way to
> > go. Anyone agree or disagree?
> >
> > Thanks,
> > Brandon
> >
> > On Mon, 2014-06-23 at 12:15 -0700, Stephen Balukoff
> wrote:
> > > The separate entity makes sense for certificates
> participating in an
> > > SNI configuration, but probably not so much for
> the 'default'
> > > certificate used when TLS is being terminated.
> > >
> > >
> > > Vijay: You're also right that other TLS-related
> attributes will
> > > probably get added to the Listener object. This
> probably makes sense
> > > if they apply to the Listener object as a whole.
> (This includes things
> > > like TLS version and cipher selection.)
> > >
> > >
> > > I don't see much of a point in creating a separate
> object to contain
> > > these fields, since it would have a 1:1
> relationship with the
> > > Listener. It's true that for non-TLS-terminated
> Listeners, these
> > > fields wouldn't be used, but isn't that already
> the case in many other
> > > objects (not just in the Neutron LBaaS sub
> project)?
> > >
> > >
> > > Thanks,
> > > Stephen
> > >
> > >
> > >
> > >
> > >
> > >
> > > On Mon, Jun 23, 2014 at 9:54 AM, Brandon Logan
> > > <brandon.logan at rackspace.com> wrote:
> > > Vijay,
> > > I think the separate entity is still going
> to happen. I don't
> > > think it
> > > has remvoed. Or that is may just be my
> assumption.
> > >
> > > Thanks,
> > > Brandon
> > >
> > > On Mon, 2014-06-23 at 15:59 +0000, Vijay
> Venkatachalam wrote:
> > > > Hi:
> > > >
> > > >
> > > > In the “LBaaS TLS termination capability
> specification”
> > > proposal
> > > >
> > > > https://review.openstack.org/#/c/98640/
> > > >
> > > > TLS settings like default certificate
> container id and SNI
> > > cert list are part of the listener
> properties.
> > > >
> > > > I think it is better to have this as a
> separate entity so
> > > that the listener properties are clean and
> is not “corrupted”
> > > with TLS settings.
> > > >
> > > > I liked the original SSL proposal better
> where TLS settings
> > > was a separate entity.
> > > >
> > > > It is just 2 properties now but in
> future the TLS settings
> > > would grow and if we are going to
> introduce a TLS
> > > profile/params/settings entity later, it
> is better to do it
> > > now (albeit with min properties).
> > > >
> > > > Thanks,
> > > > Vijay V.
> > >
> > > >
> _______________________________________________
> > > > 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
> > >
> > >
> > >
> > >
> > > --
> > > Stephen Balukoff
> > > Blue Box Group, LLC
> > > (800)613-4305 x807
> > > _______________________________________________
> > > 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
>
>
>
>
>
>
>
> --
> Stephen Balukoff
> Blue Box Group, LLC
> (800)613-4305 x807
>
>
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
>
>
>
>
> --
> Stephen Balukoff
> Blue Box Group, LLC
> (800)613-4305 x807
>
>
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
>
>
>
> --
> Stephen Balukoff
> Blue Box Group, LLC
> (800)613-4305 x807
>
>
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
>
>
> --
> Stephen Balukoff
> Blue Box Group, LLC
> (800)613-4305 x807
> _______________________________________________
> 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
More information about the OpenStack-dev
mailing list