[openstack-dev] [neutron][api][grapql] Proof of Concept

Gilles Dubreuil gdubreui at redhat.com
Thu Jun 7 00:35:13 UTC 2018


The branch is now available under feature/graphql on the neutron core 
repository [1].

Just to summarize our initial requirements:

- GraphQL endpoint to be added through a new WeBoB/WSGI stack
- Add graphene library [2]
- Unit tests and implementation for GraphQL schema for networks, subnets 
and ports Types.

I think we should support Relay by making the Schema Relay compliant and 
support Node ID, cursor connections and .
This will offer re-fetch, automated pagination and caching out of the 
box and not only will show the power of GraphQL but also because on the 
long run it would more likely what would be needed for complex API 
structures like we have across the board.

Any thoughts?

[1] https://git.openstack.org/cgit/openstack/neutron/log/?h=feature/graphql
[2] http://graphene-python.org/

On 31/05/18 17:27, Flint WALRUS wrote:
> Hi Gilles, Ed,
>
> I’m really glad and thrilled to read such good news!
>
> At this point it’s cool to see that many initiatives have the same 
> convergent needs regarding GraphQL as it will give us a good traction 
> from the beginning if our PoC manage to sufficiently convince our peers.
>
> Let me know as soon as the branch have been made, I’ll work on it.
>
> Regards,
> Fl1nt.
> Le jeu. 31 mai 2018 à 09:17, Gilles Dubreuil <gdubreui at redhat.com 
> <mailto:gdubreui at redhat.com>> a écrit :
>
>     Hi Flint,
>
>     I wish it was "my" summit ;)
>     In the latter case I'd make the sessions an hour and not 20 or 40
>     minutes, well at least for the Forum part. And I will also make
>     only one summit a year instead of two (which is also a feed back I
>     got from the Market place). I've passed that during the user
>     feedback session.
>
>     Sorry for not responding earlier, @elmiko is going to send the
>     minutes of the API SIG forum session we had.
>
>     We confirmed Neutron to be the PoC.
>     We are going to use a feature branch, waiting for Miguel Lavalle
>     to confirm the request has been acknowledge by the Infra group.
>     The PoC goal is to show GraphQL efficiency.
>     So we're going to make something straightforward, use Neutron
>     existing server by  adding the graphQL endpoint and cover few core
>     items such as network, subnets and ports (for example).
>
>     Also the idea of having a central point of access for OpenStack
>     APIs using GrahpQL stitching and delegation is exciting for
>     everyone (and I had obviously same feedback off the session) and
>     that's something that could happen once the PoC has convinced.
>
>     During the meeting, Jiri Tomasek explained how GraphQL could help
>     TripleO UI. Effectively they struggle with APIs requests and had
>     to create a middle(ware) module in JS to do API work and
>     reconstruction before the Javascript client can use it. GraphQL
>     would simplify the process and allow to get rid of the module. He
>     also explained, after the meeting, how Horizon could benefit as
>     well, allowing to use only JS and avoid Django altogether!
>
>     I've also been told that Zuul nees GraphQL.
>
>     Well basically the question is who doesn't need it?
>
>     Cheers,
>     Gilles
>
>
>
>     On 31/05/18 03:34, Flint WALRUS wrote:
>>     Hi Gilles, I hope you enjoyed your Summit!?
>>
>>     Did you had any interesting talk to report about our little
>>     initiative ?
>>     Le dim. 6 mai 2018 à 15:01, Gilles Dubreuil <gdubreui at redhat.com
>>     <mailto:gdubreui at redhat.com>> a écrit :
>>
>>
>>         Akihiro, thank you for your precious help!
>>
>>         Regarding the choice of Neutron as PoC, I'm sorry for not
>>         providing much details when I said "because of its specific
>>         data model",
>>         effectively the original mention was  "its API exposes things
>>         at an individual table level, requiring the client to join
>>         that information to get the answers they need".
>>         I realize now such description probably applies to many
>>         OpenStack APIs.
>>         So I'm not sure what was the reason for choosing Neutron.
>>         I suppose Nova is also a good candidate because API is quite
>>         complex too, in a different way, and need to expose the data
>>         API and the control API plane as we discussed.
>>
>>         After all Neutron is maybe not the best candidate but it
>>         seems good enough.
>>
>>         And as Flint say the extension mechanism shouldn't be an issue.
>>
>>         So if someone believe there is a better candidate for the
>>         PoC, please speak now.
>>
>>         Thanks,
>>         Gilles
>>
>>         PS: Flint,  Thank you for offering to be the advocate for
>>         Berlin. That's great!
>>
>>
>>         On 06/05/18 02:23, Flint WALRUS wrote:
>>>         Hi Akihiro,
>>>
>>>         Thanks a lot for this insight on how neutron behave.
>>>
>>>         We would love to get support and backing from the neutron
>>>         team in order to be able to get the best PoC possible.
>>>
>>>         Someone suggested neutron as a good choice because of it
>>>         simple database model. As GraphQL can manage your behavior
>>>         of an extension declaring its own schemes I don’t think it
>>>         would take that much time to implement it.
>>>
>>>         @Gilles, if I goes to the berlin summitt I could definitely
>>>         do the networking and relationship work needed to get
>>>         support on our PoC from different teams members. This would
>>>         help to spread the world multiple time and don’t have a long
>>>         time before someone come to talk about this subject as what
>>>         happens with the 2015 talk of the Facebook speaker.
>>>
>>>         Le sam. 5 mai 2018 à 18:05, Akihiro Motoki
>>>         <amotoki at gmail.com <mailto:amotoki at gmail.com>> a écrit :
>>>
>>>             Hi,
>>>
>>>             I am happy to see the effort to explore a new API mechanism.
>>>             I would like to see good progress and help effort as API
>>>             liaison from the neutron team.
>>>
>>>             > Neutron has been selected for the PoC because of its
>>>             specific data model
>>>
>>>             On the other hand, I am not sure this is the right
>>>             reason to choose 'neutron' only from this reason. I
>>>             would like to note "its specific data model" is not the
>>>             reason that makes the progress of API versioning slowest
>>>             in the OpenStack community. I believe it is worth
>>>             recognized as I would like not to block the effort due
>>>             to the neutron-specific reasons.
>>>             The most complicated point in the neutron API is that
>>>             the neutron API layer allows neutron plugins to declare
>>>             which features are supported. The neutron API is a
>>>             collection of API extensions defined in the neutron-lib
>>>             repo and each neutron plugin can declare which subset(s)
>>>             of the neutron APIs are supported. (For more detail, you
>>>             can check how the neutron API extension mechanism is
>>>             implemented). It is not defined only by the neutron API
>>>             layer. We need to communicate which API features are
>>>             supported by communicating enabled service plugins.
>>>
>>>             I am afraid that most efforts to explore a new mechanism
>>>             in neutron will be spent to address the above points
>>>             which is not directly related to GraphQL itself.
>>>             Of course, it would be great if you overcome
>>>             long-standing complicated topics as part of GraphQL
>>>             effort :)
>>>
>>>             I am happy to help the effort and understand how the
>>>             neutron API is defined.
>>>
>>>             Thanks,
>>>             Akihiro
>>>
>>>
>>>             2018年5月5日(土) 18:16 Gilles Dubreuil <gdubreui at redhat.com
>>>             <mailto:gdubreui at redhat.com>>:
>>>
>>>                 Hello,
>>>
>>>                 Few of us recently discussed [1] how GraphQL [2],
>>>                 the next evolution
>>>                 from REST, could transform OpenStack APIs for the
>>>                 better.
>>>                 Effectively we believe OpenStack APIs provide
>>>                 perfect use cases for
>>>                 GraphQL DSL approach, to bring among other
>>>                 advantages, better
>>>                 performance and stability, easier developments and
>>>                 consumption, and with
>>>                 GraphQL Schema provide automation capabilities never
>>>                 achieved before.
>>>
>>>                 The API SIG suggested to start an API GraphQL Proof
>>>                 of Concept (PoC) to
>>>                 demonstrate the capabilities before eventually
>>>                 extend GraphQL to other
>>>                 projects.
>>>                 Neutron has been selected for the PoC because of its
>>>                 specific data model.
>>>
>>>                 So if you are interested, please join us.
>>>                 For those who can make it, we'll also discuss this
>>>                 during the SIG API
>>>                 BoF at OpenStack Summit at Vancouver [3]
>>>
>>>                 To learn more about GraphQL, check-out
>>>                 howtographql.com <http://howtographql.com> [4].
>>>
>>>                 So let's get started...
>>>
>>>
>>>                 [1]
>>>                 http://lists.openstack.org/pipermail/openstack-dev/2018-May/130054.html
>>>                 [2] http://graphql.org/
>>>                 [3]
>>>                 https://www.openstack.org/summit/vancouver-2018/summit-schedule/events/21798/api-special-interest-group-session
>>>                 [4] https://www.howtographql.com/
>>>
>>>                 Regards,
>>>                 Gilles
>>>
>>>
>>>
>>>                 __________________________________________________________________________
>>>                 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://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://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/20180607/4d9cf527/attachment.html>


More information about the OpenStack-dev mailing list