Open Stack

Ishii-san,

Re-reading the proposal you sent on Feb. 2nd, I realized that you didn't
separate between generic operations and plugin-specific operations. I
thought you did, sorry. I propose to rewrite your spec into one API and two
(optional) extensions:

- network service API (plugin-agnostic):
    - list_vnics(): [vnic_id]
        Return the list of vnic_id created by the tenant (project), where
vnic_id is the ID of a VNIC.
    - destroy_vnic(vnic_id)
        Remove a VNIC from its VM, given its ID, and destroy it.
    - plug(vnic_id, port_id)
        Plug the VNIC with ID vnic_id into the port with ID port_id, both
managed by this network service.
    - unplug(vnic_id)
        Unplug the VNIC from its port, previously plugged by calling plug().
    - list_ports(): [port_id]
        Return the list of IDs of ports created by the tenant (project).
    - destroy_port(port_id)
        Destroy port with ID port_id.

- Ethernet VNIC API extension:
    - create_vnic([mac_address]): vnid_id
        Create a VNIC and return the ID of the created VNIC. Associate the
given MAC address with the VNIC, or associate a random unique MAC with it if
not given.
        This cannot be part of the API because it violates criterion 4): we
plan to implement non-Ethernet virtual devices to connect VMs, so this
operation cannot be implemented in that specific plugin.

- NATed IPv4 network API extension:
    - create_network(address_prefix): network_id
        Create a new logical network with floating addresses in the given
address range, and return the network ID.
        This cannot be part of the API because it violates criterion 4): a
network service plugin may not support NATted IPv4 networks, for instance a
plugin that supports only pure L2 private Ethernet networks to interconnect
VMs. Moreover, the notion of "logical network" doesn't make sense in all
cases: one can imagine a network plugin where every port is implemented by a
separate Layer-2 OpenVPN connection to a tenant's private physical Ethernet
network, in which case there is no notion of "logical network" (or it would
require users to create a separate logical network for every port / VNIC,
which would be very inconvenient).
    - list_networks(): [network_id]
        Return the list of IDs of logical networks created by the tenant
(project).
        This cannot be part of the API because it violates criterion 4):
idem, the notion of "logical network" is not plugin-agnostic.
    - destroy_network(network_id)
        Destroy the logical network with ID network_id.
        This cannot be part of the API because it violates criterion 4):
idem, the notion of "logical network" is not plugin-agnostic.
    - create_port(network_id): port_id
        Create a port in the logical network with ID network_id, associate a
floating address to it, and return the port's ID.
        This cannot be part of the API because it violates criterion 4):
idem, the notion of "logical network" is not plugin-agnostic.

What do you think of that new version of the API spec?
Do you agree with the split into API+extensions?

Regards,
--
Romain Lenglet

2011/2/16 Romain Lenglet <romain at midokura.jp>

> Hi Erik,
>
> Thanks for your comments.
>
> There doesn't seem to be a consensus to use "core API + extensions" vs.
> multiple APIs?
> Anyway, I don't see any issues with specifying a "core API" for network
> services, and a "core API" for network agents, corresponding exactly to
> NTT's Ishii-san's "generic APIs", and specifying all the non-generic,
> plugin-specific operations in extensions.
> If the norm becomes to have a core API + extensions, then the network
> service spec will be modified to follow that norm. No problem.
>
> The important point we need to agree on is what goes into the API, and what
> goes into extensions.
>
> Let me rephrase the criteria that I proposed, using the "API" and
> "extensions" terms:
> 1) any operation called by the compute service (Nova) directly MUST be
> specified in the API;
> 2) any operation called by users / admin tools MAY be specified in the API,
> but not necessarily;
> 3) any operation specified in the API MUST be independent from details of
> specific network service plugins (e.g. specific network models, specific
> supported protocols, etc.), i.e. that operation can be supported by every
> network service plugin imaginable, which means that:
> 4) any operation that cannot be implemented by all plugins MUST be
> specified in an extension, i.e. if one comes up with a counter-example
> plugin that cannot implement that operation, then the operation cannot be
> specified in the API and MUST be specified in an extension.
>
> Do we agree on those criteria?
>
> I think Ishii-san's proposal meets those criteria.
> Do you see any issues with Ishii-san's proposal regarding the split between
> core operations and extension operations?
> If you think that some operations that are currently defined as extensions
> in Ishii-san's proposal should be in the API, I'll be happy to try to give
> counter-examples of network service plugins that can't implement them. :)
>
> Regards,
> --
> Romain Lenglet
>
>
> 2011/2/16 Erik Carlin <erik.carlin at rackspace.com>
>
>  My understanding is that we want a single, canonical OS network service
>> API.  That API can then be implemented by different "service engines" on
>> that back end via a plug-in/driver model.  The way additional features are
>> added to the canonical API that may not be core or for widespread adoption
>> (e.g. something vendor specific) is via extensions.  You can take a look at
>> the proposed OS compute API spec<http://wiki.openstack.org/OpenStackAPI_1-1>to see how extensions are implemented there.  Also, Jorge Williams has done
>> a good write up of the concept here<http://wiki.openstack.org/JorgeWilliams?action=AttachFile&do=view&target=Extensions.pdf>
>> .
>>
>>  Erik
>>
>>   From: Romain Lenglet <romain at midokura.jp>
>> Date: Tue, 15 Feb 2011 17:03:57 +0900
>> To: 石井 久治 <ishii.hisaharu at lab.ntt.co.jp>
>> Cc: <openstack at lists.launchpad.net>
>>
>> Subject: Re: [Openstack] Network Service for L2/L3 Network Infrastructure
>> blueprint
>>
>>   Hi Ishii-san,
>>
>> On Tuesday, February 15, 2011 at 16:28, 石井 久治 wrote:
>>
>>  Hello Hiroshi-san
>>
>> >> Do you mean that the former API is an interface that is
>> >> defined in OpenStack project, and the latter API is
>> >> a vendor specific API?
>> > My understanding is that yes, that's what he means.
>>
>> I also think so.
>>
>> In addition, I feel it is issue that what network functions should be
>> defined as generic API, and what network functions should be defined as
>> plugin specific API.
>> How do you think ?
>>
>> I propose to apply the following criteria to determine which operations
>> belong to the generic API:
>> - any operation called by the compute service (Nova) directly MUST belong
>> to the generic API;
>> - any operation called by users (REST API, etc.) MAY belong to the generic
>> API;
>> - any operation belonging to the generic API MUST be independent from
>> details of specific network service plugins (e.g. specific network models,
>> specific supported protocols, etc.), i.e. the operation can be supported by
>> every network service plugin imaginable, which means that if one can come up
>> with a counter-example plugin that cannot implement that operation, then the
>> operation cannot belong to the generic API.
>>
>>  How about that?
>>
>>  Regards,
>> --
>> Romain Lenglet
>>
>>
>>
>> Thanks
>> Hisaharu Ishii
>>
>>
>> (2011/02/15 16:18), Romain Lenglet wrote:
>>
>> Hi Hiroshi,
>> On Tuesday, February 15, 2011 at 15:47, Hiroshi DEMPO wrote:
>> Hello Hisaharu san
>>
>>
>> I am not sure about the differences between generic network API and
>> plugin X specific network service API.
>>
>> Do you mean that the former API is an interface that is
>> defined in OpenStack project, and the latter API is
>> a vendor specific API?
>>
>>
>> My understanding is that yes, that's what he means.
>>
>> --
>> Romain Lenglet
>>
>>
>>
>> Thanks
>> Hiroshi
>>
>>  -----Original Message-----
>> From: openstack-bounces+dem=ah.jp.nec.com at lists.launchpad.net
>> [mailto:openstack-bounces <openstack-bounces>+dem=
>> ah.jp.nec.com at lists.launchpad.ne
>> t] On Behalf Of 石井 久治
>> Sent: Thursday, February 10, 2011 8:48 PM
>> To: openstack at lists.launchpad.net
>> Subject: Re: [Openstack] Network Service for L2/L3 Network
>> Infrastructure blueprint
>>
>> Hi, all
>>
>> As we have said before, we have started designing and writing
>> POC codes of network service.
>>
>>  - I know that there were several documents on the new network
>> service issue that were locally exchanged so far.
>> Why not collecting them into one place and share them
>>
>> publicly?
>>
>> Based on these documents, I created an image of
>> implementation (attached). And I propose the following set of
>> methods as the generic network service APIs.
>> - create_vnic(): vnic_id
>> Create a VNIC and return the ID of the created VNIC.
>> - list__vnics(vm_id): [vnic_id]
>> Return the list of vnic_id, where vnic_id is the ID of a VNIC.
>> - destroy_vnic(vnic_id)
>> Remove a VNIC from its VM, given its ID, and destroy it.
>> - plug(vnic_id, port_id)
>> Plug the VNIC with ID vnic_id into the port with ID
>> port_id managed by this network service.
>> - unplug(vnic_id)
>> Unplug the VNIC from its port, previously plugged by
>> calling plug().
>> - create_network(): network_id
>> Create a new logical network.
>> - list_networks(project_id): [network_id]
>> Return the list of logical networks available for
>> project with ID project_id.
>> - destroy_network(network_id)
>> Destroy the logical network with ID network_id.
>> - create_port(network_id): port_id
>> Create a port in the logical network with ID
>> network_id, and return the port's ID.
>> - list_ports(network_id): [port_id]
>> Return the list of IDs of ports in a network given its ID.
>> - destroy_port(port_id)
>> Destroy port with ID port_id.
>>
>> This design is a first draft.
>> So we would appreciate it if you would give us some comments.
>>
>> In parallel with it, we are writing POC codes and uploading
>> it to "lp:~ntt-pf-lab/nova/network-service".
>>
>> Thanks,
>> Hisaharu Ishii
>>
>>
>> (2011/02/02 19:02), Koji IIDA wrote:
>>
>> Hi, all
>>
>>
>> We, NTT PF Lab., also agree to discuss about network service at the
>> Diablo DS.
>>
>> However, we would really like to include network service in
>>
>> the Diablo
>>
>> release because our customers strongly demand this feature. And we
>> think that it is quite important to merge new network
>>
>> service to trunk
>>
>> soon after Diablo DS so that every developer can contribute their
>> effort based on the new code.
>>
>> We are planning to provide source code for network service
>>
>> in a couple
>>
>> of weeks. We would appreciate it if you would review it
>>
>> and give us
>>
>> some feedback before the next design summit.
>>
>> Ewan, thanks for your making new entry at wiki page (*1).
>>
>> We will also
>>
>> post our comments soon.
>>
>> (*1) http://wiki.openstack.org/NetworkService
>>
>>
>> Thanks,
>> Koji Iida
>>
>>
>> (2011/01/31 21:19), Ewan Mellor wrote:
>>
>> I will collect the documents together as you suggest, and
>>
>>  I agree that we need to get the requirements laid out again.
>>
>>
>> Please subscribe to the blueprint on Launchpad -- that way
>>
>>  you will be notified of updates.
>>
>>
>> https://blueprints.launchpad.net/nova/+spec/bexar-network-service
>>
>> Thanks,
>>
>> Ewan.
>>
>>  -----Original Message-----
>> From: openstack-bounces+ewan.mellor=citrix.com at lists.launchpad.net
>>
>>   [mailto:openstack-bounces <openstack-bounces>+ewan.mellor=
>> citrix.com at lists.launchpad.net
>>
>>  ]
>> On Behalf Of Masanori ITOH
>> Sent: 31 January 2011 10:31
>> To: openstack at lists.launchpad.net
>> Subject: Re: [Openstack] Network Service for L2/L3 Network
>> Infrastructure blueprint
>>
>> Hello,
>>
>> We, NTT DATA, also agree with majority of folks.
>> It's realistic shooting for the the Diablo time frame to have the
>> new network service.
>>
>> Here are my suggestions:
>>
>> - I know that there were several documents on the new network
>> service issue
>> that were locally exchanged so far.
>> Why not collecting them into one place and share them
>>
>>  publicly?
>>
>>
>> - I know that the discussion went into a bit
>>
>>  implementation details.
>>
>>  But now, what about starting the discussion from the
>>
>>  higher level
>>
>>  design things (again)? Especially, from the
>>
>>  requirements level.
>>
>>
>> Any thoughts?
>>
>> Masanori
>>
>>
>> From: John Purrier<john at openstack.org>
>> Subject: Re: [Openstack] Network Service for L2/L3 Network
>> Infrastructure blueprint
>> Date: Sat, 29 Jan 2011 06:06:26 +0900
>>
>>  You are correct, the networking service will be more
>>
>>   complex than
>>
>>   the
>>
>> volume
>>
>> service. The existing blueprint is pretty comprehensive,
>>
>>   not only
>>
>>   encompassing the functionality that exists in today's network
>> service
>>
>> in
>>
>> Nova, but also forward looking functionality around flexible
>> networking/openvswitch and layer 2 network bridging
>>
>>   between cloud
>>
>>   deployments.
>>
>> This will be a longer term project and will serve as the bedrock
>> for
>>
>> many
>>
>> future OpenStack capabilities.
>>
>> John
>>
>> -----Original Message-----
>> From: openstack-bounces+john=openstack.org at lists.launchpad.net
>>
>>   [mailto:openstack-bounces <openstack-bounces>+john=
>> openstack.org at lists.launchpad.net]
>>
>>   On
>>
>> Behalf
>>
>> Of Thierry Carrez
>> Sent: Friday, January 28, 2011 1:52 PM
>> To: openstack at lists.launchpad.net
>> Subject: Re: [Openstack] Network Service for L2/L3 Network
>>
>> Infrastructure
>>
>> blueprint
>>
>> John Purrier wrote:
>>
>> Here is the suggestion. It is clear from the response
>>
>>    on the list
>>
>>  that
>>
>> refactoring Nova in the Cactus timeframe will be too risky,
>>
>> particularly as
>>
>> we are focusing Cactus on Stability, Reliability, and
>>
>>   Deployability
>>
>>  (along
>>
>> with a complete OpenStack API). For Cactus we should leave the
>>
>> network and
>>
>> volume services alone in Nova to minimize destabilizing the code
>>
>> base. In
>>
>> parallel, we can initiate the Network and Volume Service
>>
>>   projects
>>
>>   in Launchpad and allow the teams that form around these
>>
>>   efforts to
>>
>>   move
>>
>> in
>>
>> parallel, perhaps seeding their projects from the
>>
>>   existing Nova code.
>>
>>
>> Once we complete Cactus we can have discussions at the Diablo DS
>>
>>  about
>>
>> progress these efforts have made and how best to move
>>
>>   forward with
>>
>>  Nova
>>
>> integration and determine release targets.
>>
>> I agree that there is value in starting the proof-of-concept work
>>
>> around
>>
>> the network services, without sacrificing too many developers to
>> it,
>>
>> so
>>
>> that a good plan can be presented and discussed at the
>>
>>   Diablo Summit.
>>
>>
>> If volume sounds relatively simple to me, network sounds
>>
>> significantly
>>
>> more complex (just looking at the code ,network manager code is
>> currently used both by nova-compute and nova-network to
>>
>>   modify the
>>
>>  local
>>
>> networking stack, so it's more than just handing out IP
>>
>>   addresses
>>
>>    through an API).
>>
>> Cheers,
>>
>> --
>> Thierry Carrez (ttx)
>> Release Manager, OpenStack
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~openstack
>> Post to : openstack at lists.launchpad.net
>> Unsubscribe : https://launchpad.net/~openstack
>> More help : https://help.launchpad.net/ListHelp
>>
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~openstack
>> Post to : openstack at lists.launchpad.net
>> Unsubscribe : https://launchpad.net/~openstack
>> More help : https://help.launchpad.net/ListHelp
>>
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~openstack
>> Post to : openstack at lists.launchpad.net
>> Unsubscribe : https://launchpad.net/~openstack
>> More help : https://help.launchpad.net/ListHelp
>>
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~openstack
>> Post to : openstack at lists.launchpad.net
>> Unsubscribe : https://launchpad.net/~openstack
>> More help : https://help.launchpad.net/ListHelp
>>
>>   _______________________________________________
>> Mailing list: https://launchpad.net/~openstack
>> Post to : openstack at lists.launchpad.net
>> Unsubscribe : https://launchpad.net/~openstack
>> More help : https://help.launchpad.net/ListHelp
>>
>> Attachments:
>> - smime.p7s
>>
>>
>> _______________________________________________
>> Mailing list: https://launchpad.net/~openstack
>> Post to : openstack at lists.launchpad.net
>> Unsubscribe : https://launchpad.net/~openstack
>> More help : https://help.launchpad.net/ListHelp
>>
>>
>>   _______________________________________________ Mailing list:
>> https://launchpad.net/~openstack Post to : openstack at lists.launchpad.netUnsubscribe :
>> https://launchpad.net/~openstack More help :
>> https://help.launchpad.net/ListHelp
>>
>> Confidentiality Notice: This e-mail message (including any attached or
>> embedded documents) is intended for the exclusive and confidential use of the
>> individual or entity to which this message is addressed, and unless otherwise
>> expressly indicated, is confidential and privileged information of Rackspace.
>> Any dissemination, distribution or copying of the enclosed material is prohibited.
>> If you receive this transmission in error, please notify us immediately by e-mail
>> at abuse at rackspace.com, and delete the original message.
>> Your cooperation is appreciated.
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20110216/1364a235/attachment.html>