Open Stack

On Fri, Jun 10, 2016 at 6:37 AM, Henry Nash <henrynash9 at mac.com> wrote:

> On further reflection, it seems to me that we can never simply enable
> either of these approaches in a single release. Even a v4.0 version of the
> API doesn’t help - since presumably a sever supporting v4 would want to be
> able to support v3.x for a signification time; and, already discussed, as
> soon as you allow multiple none-names to have the same name, you can no
> longer guarantee to support the current API.
>
> Hence the only thing I think we can do (if we really do want to change the
> current functionality) is to do this over several releases with a typical
> deprecation cycle, e.g.
>
> 1) At release 3.7 we allow you to (optionally) specify path names for
> auth….but make no changes to the uniqueness constraints. We also change the
> GET /auth/projects to return a path name. However, you can still auth
> exactly the way we do today (since there will always only be a single
> project of a given node-name). If however, you do auth without a path (to a
> project that isn’t a top level project), we log a warning to say this is
> deprecated (2 cycles, 4 cycles?)
> 2) If you connect with a 3.6 client, then you get the same as today for
> GET /auth/projects and cannot use a path name to auth.
> 3) At sometime in the future, we deprecate the “auth without a path”
> capability. We can debate as to whether this has to be a major release.
>
> If we take this gradual approach, I would be pushing for the “relax
> project name constraints” approach…since I believe this leads to a cleaner
> eventual solution (and there is no particular advantage with the
> hierarchical naming approach) - and (until the end of the deprecation)
> there is no break to the existing API.
>
> Henry
>
>
How do you handle relaxed project name constraints without completely
breaking 3.6 auth - regardless of future microversion that requires the
full path. This is a major api change and will result in very complex
matrix of project name mapping (old projects that can be accessed without
the full path, new that must always have the path)?

Simply put, I do not see relaxing project name constraints as viable
without a major API change and projects that simply are unavailable for
scoping a token to under the base api version (pre-microversions) of 3.6

I am certain that if all projects post API version 3.6 are created with the
full-path name only and that is how they are represented to the user for
auth, we get both things for free. Old projects pre-full-path would need
optional compatibility for deconstructing a full-path for them.  Basically
you end up with "path" and "name", in old projects these differ, in new
projects they are the same.  No conflicts, not breaking "currently working
auth", no "major API version" needed.

--Morgan


> On 7 Jun 2016, at 09:47, Henry Nash <henrynash9 at mac.com> wrote:
>
> OK, so thanks for the feedback - understand the message.
>
> However, in terms of compatibility, the one thing that concerns me about
> the hierarchical naming approach is that even with microversioing, we might
> still surprise a client. An unmodified client (i.e. doesn’t understand 3.7)
> would still see a change in the data being returned (the project names have
> suddenly become full path names). We have to return this even if they don’t
> ask for 3.7, since otherwise there is no difference between this approach
> and relaxing the project naming in terms of trying to prevent auth
> breakages.
>
> In more detail:
>
> 1) Both approaches were planned to return the path name (instead of the
> node name) in GET /auth/projects - i.e. the API you are meant to use to
> find out what you can scope to
> 2) Both approaches were planned to accept the path name in the auth
> request block
> 3) The difference in hierarchical naming is the if I do a regular GET
> /project(s) I also see the full path name as the “project name”
>
> if we don’t do 3), then code that somehow authenticates, and then uses the
> regular GET /project(s) calls to find a project name and then re-scopes (or
> re-auths) to that name, will fail if the project they want is not a top
> level project. However, the flip side is that if there is code that uses
> these same calls to, say, display projects to the user (e.g. a home grown
> UI) - then it might get confused until it supports 3.7 (i.e. asking for the
> old microversion won’t help it) since all the names include the
> hierarchical path.
>
> Just want to make sure we understand the implications….
>
> Henry
>
> On 4 Jun 2016, at 08:34, Monty Taylor <mordred at inaugust.com> wrote:
>
> On 06/04/2016 01:53 AM, Morgan Fainberg wrote:
>
>
> On Jun 3, 2016 12:42, "Lance Bragstad" <lbragstad at gmail.com
> <mailto:lbragstad at gmail.com <lbragstad at gmail.com>>> wrote:
>
>
>
>
> On Fri, Jun 3, 2016 at 11:20 AM, Henry Nash <henrynash9 at mac.com
>
> <mailto:henrynash9 at mac.com <henrynash9 at mac.com>>> wrote:
>
>
>
> On 3 Jun 2016, at 16:38, Lance Bragstad <lbragstad at gmail.com
>
> <mailto:lbragstad at gmail.com <lbragstad at gmail.com>>> wrote:
>
>
>
>
> On Fri, Jun 3, 2016 at 3:20 AM, Henry Nash <henrynash9 at mac.com
>
> <mailto:henrynash9 at mac.com <henrynash9 at mac.com>>> wrote:
>
>
>
> On 3 Jun 2016, at 01:22, Adam Young <ayoung at redhat.com
>
> <mailto:ayoung at redhat.com <ayoung at redhat.com>>> wrote:
>
>
> On 06/02/2016 07:22 PM, Henry Nash wrote:
>
>
> Hi
>
> As you know, I have been working on specs that change the way we
>
> handle the uniqueness of project names in Newton. The goal of this is to
> better support project hierarchies, which as they stand today are
> restrictive in that all project names within a domain must be unique,
> irrespective of where in the hierarchy that projects sits (unlike, say,
> the unix directory structure where a node name only has to be unique
> within its parent). Such a restriction is particularly problematic when
> enterprise start modelling things like test, QA and production as
> branches of a project hierarchy, e.g.:
>
>
> /mydivsion/projectA/dev
> /mydivsion/projectA/QA
> /mydivsion/projectA/prod
> /mydivsion/projectB/dev
> /mydivsion/projectB/QA
> /mydivsion/projectB/prod
>
> Obviously the idea of a project name (née tenant) being unique
>
> has been around since near the beginning of (OpenStack) time, so we must
> be cautions. There are two alternative specs proposed:
>
>
> 1) Relax project name
>
> constraints: https://review.openstack.org/#/c/310048/
>
> 2) Hierarchical project
>
> naming: https://review.openstack.org/#/c/318605/
>
>
> First, here’s what they have in common:
>
> a) They both solve the above problem
> b) They both allow an authorization scope to use a path rather
>
> than just a simple name, hence allowing you to address a project
> anywhere in the hierarchy
>
> c) Neither have any impact if you are NOT using a hierarchy -
>
> i.e. if you just have a flat layer of projects in a domain, then they
> have no API or semantic impact (since both ensure that a project’s name
> must still be unique within a parent)
>
>
> Here’s how the differ:
>
> - Relax project name constraints (1), keeps the meaning of the
>
> ‘name’ attribute of a project to be its node-name in the hierarchy, but
> formally relaxes the uniqueness constraint to say that it only has to be
> unique within its parent. In other words, let’s really model this a bit
> like a unix directory tree.
>
>
> I think I lean towards relaxing the project name constraint. The
>
> reason is because we already expose `domain_id`, `parent_id`, and `name`
> of a project. By relaxing the constraint we can give the user everything
> the need to know about a project without really changing any of these.
> When using 3.7, you know what domain your project is in, you know the
> identifier of the parent project, and you know that your project name is
> unique within the parent.
>
>
> - Hierarchical project naming (2), formally changes the meaning
>
> of the ‘name’ attribute to include the path to the node as well as the
> node name, and hence ensures that the (new) value of the name attribute
> remains unique.
>
>
> Do we intend to *store* the full path as the name, or just build it
>
> out on demand? If we do store the full path, we will have to think about
> our current data model since the depth of the organization or domain
> would be limited by the max possible name length. Will performance be
> something to think about building the full path on every request?
>
>
> I now mention this issue in the spec for hierarchical project naming
>
> (the relax naming approach does not suffer this issue). As you say,
> we’ll have to change the DB (today it is only 64 chars) if we do store
> the full path . This is slightly problematic since the maximum depth of
> hierarchy is controlled by a config option, and hence could be changed.
> We will absolutely have be able to build the path on-the-fly in order to
> support legacy drivers (who won’t be able to store more than 64 chars).
> We may need to do some performance tests to ascertain if we can get away
> with building the path on-the-fly in all cases and avoid changing the
> table.  One additional point is that, of course, the API will return
> this path whenever it returns a project - so clients will need to be
> aware of this increase in size.
>
>
>
> These are all good points that continue to push me towards relaxing
>
> the project naming constraint :)
>
>
>
> While whichever approach we chose would only be included in a new
>
> microversion (3.7) of the Identity API, although some relevant APIs can
> remain unaffected for a client talking 3.6 to a Newton server, not all
> can be. As pointed out be jamielennox, this is a data modelling problem
> - if a Newton server has created multiple projects called “dev” in the
> hierarchy, a 3.6 client trying to scope a token simply to “dev” cannot
> be answered correctly (and it is proposed we would have to return an
> HTTP 409 Conflict error if multiple nodes with the same name were
> detected). This is true for both approaches.
>
>
> Other comments on the approaches:
>
> - Having a full path as the name seems duplicative with the
>
> current project entity - since we already return the parent_id (hence
> parent_id + name is, today, sufficient to place a project in the
> hierarchy).
>
>
>
> The one thing I like is the ability to specify just the full path
>
> for the OS_PROJECT_NAME env var, but we could make that a separate
> variable.  Just as DOMAIN_ID + PROJECT_NAME is unique today,
> OS_PROJECT_PATH should be able to fully specify a project
> unambiguously.  I'm not sure which would have a larger impact on users.
>
>
> Agreed - and this could be done for both approaches (since this is
>
> all part of the “auth data flow").
>
>
>
> - In the past, we have been concerned about the issue of what we
>
> do if there is a project further up the tree that we do not have any
> roles on. In such cases, APIs like list project parents will not display
> anything other than the project ID for such projects. In the case of
> making the name the full path, we would be effectively exposing the name
> of all projects above us, irrespective of whether we had roles on them.
> Maybe this is OK, maybe it isn’t.
>
>
>
> I think it is OK.  If this info needs to be hidden from a user,
>
> the project should probably be in a different domain.
>
>
> - While making the name the path keeps it unique, this is fine if
>
> clients blindly use this attribute to plug back into another API to
> call. However if, for example, you are Horizon and are displaying them
> in a UI then you need to start breaking down the path into its
> components, where you don’t today.
>
> - One area where names as the hierarchical path DOES look right
>
> is calling the /auth/projects API - where what the caller wants is a
> list of projects they can scope to - so you WANT this to be the path you
> can put in an auth request.
>
>
> Given that neither can fully protect a 3.6 client, my personal
>
> preference is to go with the cleaner logical approach which I believe is
> the Relax project name constraints (1), with the addition of changing
> GET /auth/projects to return the path (since this is a specialised API
> that happens before authentication) - but I am open to persuasion (as
> the song goes).
>
>
> There are those that might say that perhaps we just can’t change
>
> this. I would argue that since this ONLY affects people who actually
> create hierarchies and that today such hierarchical use is in its
> infancy, then now IS the time to change this. If we leave it too long,
> then it will become really hard to change what will by then have become
> a tough restriction.
>
>
> Henry
>
>
>
>
>
> __________________________________________________________________________
>
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe:
>
> OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org/?subject:unsubscribe>>
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
>
> __________________________________________________________________________
>
> OpenStack Development Mailing List (not for usage questions)
>
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org/?subject:unsubscribe>>
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
>
>
> __________________________________________________________________________
>
> OpenStack Development Mailing List (not for usage questions)
>
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org/?subject:unsubscribe>>
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
> __________________________________________________________________________
>
> OpenStack Development Mailing List (not for usage questions)
>
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org/?subject:unsubscribe>>
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
>
>
> __________________________________________________________________________
>
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe:
>
> OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org/?subject:unsubscribe>>
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe:
>
> OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org?subject:unsubscribe
> <http://OpenStack-dev-request@lists.openstack.org/?subject:unsubscribe>>
>
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
> My main concern with relaxing the uniqueness is you now have a
> significant issue with auth changing. Fundamentally you cannot scope to
> a project under a domain by name and domain name alone. This will break
> current auth paths and auth mechanisms.
>
> I personally think we are a bit wedged even with microversions without
> providing the full path as a name for projects created as such (and
> current projects retaining the uniqueness or if created via the old api
> maintaining the uniqueness constraint and new api projects not being
> represented at all).
>
> So in short, I am strongly against relaxing uniqueness.
>
>
> I agree.
>
> This would break code that exists currently. A microversion is fine for
> creating a new api call. A microversion is not ok for fundamentally
> breaking the semantics of an existing API. This would be a major API
> version bump. It would break existing users with existing code, and
> since there _is_ another option on the table that does not break
> existing users with existing code, I believe that it is imperative to
> select the non-breaking option.
>
> New (microversion) would make projects that are only represented by full
> path. Old projects could be represented either way. (For auth/crud that
> uses project names).
>
> Old microversion api (today) would require uniqueness constraint and not
> show/allow full pathname projects.
>
> This way we do not break auth that works today.
>
> Most everyone consumes and stores by id (nova, cinder, etc) so we likely
> won't break much with extended project names.
>
>
> End users work by name - but yes, I agree the impact cross-openstack is
> unlikely.
>
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> 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/20160610/0df3395f/attachment.html>