Open Stack

On 8/3/2016 1:44 AM, Alex Xu wrote:
>
>
> 2016-08-03 14:40 GMT+08:00 Alex Xu <soulxu at gmail.com
> <mailto:soulxu at gmail.com>>:
>
>
>
>     2016-08-02 22:09 GMT+08:00 Matt Riedemann
>     <mriedem at linux.vnet.ibm.com <mailto:mriedem at linux.vnet.ibm.com>>:
>
>         On 8/2/2016 2:41 AM, Alex Xu wrote:
>
>             A little strange we have two API endpoints, one is
>             '/servers/{uuid}/os-interfaces', another one is
>             '/servers/{uuid}/os-virtual-interfaces'.
>
>             I prefer to keep os-attach-interface. Due to I think we
>             should deprecate
>             the nova-network also. Actually we deprecate all the
>             nova-network
>             related API in the 2.36 also. And os-attach-interface didn't
>             support
>             nova-network, then it is the right choice.
>
>             So we can deprecate the os-virtual-interface in newton. And
>             in Ocata, we
>             correct the implementation to get the vif info and tag.
>             os-attach-interface actually accept the server_id, and there
>             is check
>             ensure the port belong to the server. So it shouldn't very
>             hard to get
>             the vif info and tag.
>
>             And sorry for I missed that when coding patches also...let
>             me if you
>             need any help at here.
>
>
>
>                 --
>
>                 Thanks,
>
>                 Matt Riedemann
>
>
>
>
>             __________________________________________________________________________
>                 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://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>         Alex,
>
>         os-interface will be deprecated, that's the APIs to show/list
>         ports for a given server.
>
>         os-virtual-interfaces is not the same, and was never a proxy for
>         neutron since before 2.32 we never stored anything in the
>         virtual_interfaces table in the nova database for neutron, but
>         now we do because that's where we store the VIF tags.
>
>         We have to keep os-attach-interface (attach/detach interface
>         actions on a server).
>
>         Are you suggesting we drop os-virtual-interfaces and change the
>         behavior of os-interfaces to use the nova virtual_interfaces
>         table rather than proxying to neutron?
>
>
>     Yes, but I missed the point your point out as below. The reason is
>     that if we only deprecate the GET of os-interface, then when user
>     want to add interface, user needs to send request 'POST
>     /servers/{uuid}/os-interface'. When user want to query the interface
>     which attach to the server, user needs send request to 'GET
>     /servers/{uuid}/os-virtual-interfaces'. That means user access one
>     resource but they are under different API endpoint.
>
>     Initially I think we can use virtual_interface table to reimplement
>     the GET of os-interface'. But as you pointed out, neutron ports
>     won't show if it is created before Newton. That means we change the
>     os-interface behaviour in old Microversion. Emm... I'm a little
>     hesitate.
>
>
>
>         Note that with os-virtual-interfaces even if we start showing
>         VIFs for neutron ports, any ports created before Newton won't be
>         in there, which might be a bit confusing.
>
>
>     If we keep the os-virtual-interfaces, but we can't ensure this API
>     works for all the instances forever. Due to we can't ensure there
>     won't any old instance which created before Newton in user's cloud.
>
>     So.. one idea, we keep the GET of os-interface, deprecate the
>     os-virtual-interfaces in Newton. In Ocata, we use virtual_interface
>     table/object instead of neutron API proxy, but we fallback to proxy
>     neutron API when the instance haven't virtual_interface table( this
>     is detected by compare the network_info_cache, instance have vif in
>     network_info_cace, but no entry in virtual_interface table.).
>
>
> And remove the code about fallback in the future, when we believe there
> isn't any old instance.
>
>
>
>
>
>
>         --
>
>         Thanks,
>
>         Matt Riedemann
>
>
>         __________________________________________________________________________
>         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://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>

John suggested we deprecate os-virtual-interfaces and keep the GET 
methods to list/show interfaces in os-interface, but use the 
instance.info_cache and provide a minimal set of data, like the port 
uuid. The details on the port can then be looked up in Neutron (although 
we can possibly maintain the limited response we have for the show 
method today with what we have in instance.info_cache). That makes 
os-interface behave more like os-volume-attachments, which I like.

Regardless, this has been deferred to Ocata.

-- 

Thanks,

Matt Riedemann