[openstack-dev] [Netstack] [Quantum] plugin -> backend
Dan Wendlandt
dan at nicira.com
Mon Aug 6 16:25:20 UTC 2012
On Sat, Aug 4, 2012 at 3:46 PM, Bill Shetti <billshetti at gmail.com> wrote:
> Hi, catching up, and sorry for the request for history... but here it is.
You can see a short explanation of this on the FAQ here:
http://wiki.openstack.org/QuantumDevelopment . Longer explanation below.
>
> What compelled the original decision to be made on enabling only ONE
> plugin/backend (whatever you want to call it)?
>
> Ideally you will want to instantiate multiple services from different
> vendors etc.
>
This is the big misconception that I mentioned earlier in the thread.
There is nothing that limits a plugin to dealing with a single vendor
technology. A plugin defines a strategy for mapping from logical-layer
network resources (quantum networks, ports, etc.) to provider-layer network
constructs (e.g., VLANs, hypervisor NICs). For example, a simple plugin
might map each quantum network to a VLAN ID, which it assumes it trunked to
all switches. Even if these switches are from multiple vendors, a plugin
just needs a "driver" that allows it to configure VLANs on each type of
switch. Using multiple drivers at once is definitely possible in the
current model.
Stepping back, a helpful way to think about a "plugin" is really "what
chunk of code do I invoke when API CRUD operations are performed for
networks, subnets, and ports"? Is it code that allocates a VLAN/tunnel-id
in a database and tries to communicate with switches? Is it code that just
proxies this call to an OpenFlow controller, which does the heavy lifting?
>
> Ueno-san's metaplugin should basically - baseline in quantum.
>
The ability to create "metaplugins" was actually part of the original
Quantum design... its just one more "strategy" for dispatching API calls.
The tricky part is that to date I've heard multiple different strategies
for how a metaplugin might decide which sub-plugin to dispatch to (Nachi's
is one reasonable approach), so hard-coding a particular meta-plugin
strategy seems unwise and unnecessary, given that a metaplugin works
perfectly fine already in the existing architecture. If at some point in
the future a particular approach becomes standard, then perhaps baking it
into the architecture would make sense.
Dan
Bill
>
>
> On Thu, Aug 2, 2012 at 2:07 PM, Nachi Ueno <nachi at nttmcl.com> wrote:
>
>> 2012/8/2 Dan Wendlandt <dan at nicira.com>
>>
>>> Suffice it to say we're not going to make any drastic changes with the
>>> time remaining on F-3, but I think we can talk about this at the Grizzly
>>> summit.
>>>
>>>
>> I agree. This is reason why I implemented this function my plugin and
>> extension.
>>
>>
>>> We actually put a lot of thought into whether IPAM should have a
>>> separate plugin or not, and decided that the two where so tightly coupled
>>> that it didn't make sense. We will be moving toward a model where higher
>>> level services (routers, loadbalancers, firewalls, etc.) can be implemented
>>> by plugins other than the core L2 + IPAM plugin.
>>>
>>>
>> I got use point ,but Coupling L2 + IPAM only makes sense when we use one
>> L2 plugin.
>> It is normal large infrastructure uses multiple L2 technologies.
>>
>> Nachi
>>
>>
>>> Dan
>>>
>>>
>>> On Thu, Aug 2, 2012 at 1:03 PM, Nachi Ueno <nachi at nttmcl.com> wrote:
>>>
>>>> Hi Hua
>>>>
>>>> I agree with you. Current plugin architecture is kind of silo.
>>>> My concern is about IPAM and L3.
>>>>
>>>> - IPAM
>>>> IPAM plugin and L2 plugin can be different. However it is combined in
>>>> current structure.
>>>>
>>>> - L3 function
>>>> It needed to be connected each L2 plugin in L3.
>>>>
>>>> They are a reason I'm proposing Metaplugin.
>>>> https://review.openstack.org/#/c/10181/ ( I'm very welcome your
>>>> reviews! :) )
>>>>
>>>> By implementation of Metaplugin, I realized if each plugin will
>>>> inherits QuantumDBPluginV2, and
>>>> they do not use same table. We can use multiple plugin at once.
>>>> So at least IPAM will be solved.
>>>>
>>>> Best
>>>> Nachi Ueno
>>>>
>>>> 2012/8/1 Hua ZZ Zhang <zhuadl at cn.ibm.com>
>>>>
>>>>> just add my cents here.
>>>>>
>>>>> "Driver" concept make sense to my understaning. The current quantum
>>>>> underline plugins works and behaves more like network connectivity
>>>>> provider on top of specific type of device, from hardware and
>>>>> software, from vendors to open source. You can only enable ONE of it to
>>>>> provide virtual network service, but can't deploy without it.Just like
>>>>> database driver, it provide access of data backend and can't be absent.
>>>>> However plugin is not a essential part. Multiple plugins can be enabled at
>>>>> the same time in many software cases. They can work together with host to
>>>>> provide more functionalities.
>>>>>
>>>>> *Best Regards, *
>>>>>
>>>>> ------------------------------
>>>>>
>>>>> *Edward Zhang(张华)*
>>>>> Staff Software Engineer
>>>>> Travel&Transportation Standards
>>>>> Emerging Technology Institute(ETI)
>>>>> IBM China Software Development Lab
>>>>> e-mail: zhuadl at cn.ibm.com
>>>>> Notes ID: Hua ZZ Zhang/China/IBM
>>>>> Tel: 86-10-82450483
>>>>>
>>>>>
>>>>> 地址:北京市海淀区东北旺西路8号 中关村软件园28号楼 环宇大厦3层 邮编：100193
>>>>> Address: 3F Ring, Building 28 Zhongguancun Software Park, 8
>>>>> Dongbeiwang West Road, Haidian District, Beijing, P.R.C.100193
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> [image: Inactive hide details for Dan Wendlandt ---2012-07-31
>>>>> 14:50:45---Yes, we've had this discussion many times :) I agree that peop]Dan
>>>>> Wendlandt ---2012-07-31 14:50:45---Yes, we've had this discussion many
>>>>> times :) I agree that people find the term "plugin" confusing, but each
>>>>> time we've talked
>>>>>
>>>>>
>>>>> *Dan Wendlandt <dan at nicira.com>*
>>>>>
>>>>> 2012-07-31 14:45
>>>>> Please respond to
>>>>> OpenStack Development Mailing List <
>>>>> openstack-dev at lists.openstack.org>
>>>>>
>>>>>
>>>>>
>>>>> To
>>>>>
>>>>>
>>>>> "Sumit Naiksatam (snaiksat)" <snaiksat at cisco.com>
>>>>>
>>>>>
>>>>> cc
>>>>>
>>>>>
>>>>> OpenStack Development Mailing List <
>>>>> openstack-dev at lists.openstack.org>, "netstack at lists.launchpad.net"
>>>>> <netstack at lists.launchpad.net>, Willian Molinari <
>>>>> willian.molinari at locaweb.com.br>
>>>>>
>>>>>
>>>>> Subject
>>>>>
>>>>>
>>>>> Re: [openstack-dev] [Netstack] [Quantum] plugin -> backend
>>>>>
>>>>>
>>>>> Yes, we've had this discussion many times :) I agree that people find
>>>>> the term "plugin" confusing, but each time we've talked about it, we've
>>>>> failed to find a single term that is substantially better to warrant the
>>>>> confusion likely to be caused by renaming.
>>>>>
>>>>> In some cases I've started using the term "engine" when describing the
>>>>> plugin concept to people, since its really about a "pluggable backend" that
>>>>> powers the generic quantum API layer. The name "driver" was very
>>>>> intentionally not chosen, as driver implies that it is specific to a
>>>>> particular type of back-end device, whereas a Quantum plugin is really more
>>>>> about an overall strategy of creating logical networks, etc. For example,
>>>>> you could have a generic VLAN plugin that has drivers to talk to many
>>>>> different types of switches.
>>>>>
>>>>> Dan
>>>>>
>>>>> On Mon, Jul 30, 2012 at 7:55 PM, Sumit Naiksatam (snaiksat) <*
>>>>> snaiksat at cisco.com* <snaiksat at cisco.com>> wrote:
>>>>>
>>>>> Hi,
>>>>>
>>>>>
>>>>>
>>>>> I believe there are two topics of discussion here, one of which is
>>>>> the terminology. The way things are implemented today, I agree that the
>>>>> “plugin” terminology seems a bit confusing. However, probably the bigger
>>>>> topic of discussion is what kind of a design is preferable, “backend”
>>>>> versus “plugin”? As Yong points out, today’s Quantum service completely
>>>>> relies on the plugin for providing all functionality, including
>>>>> functionality that is probably common across plugins (like state management
>>>>> of logical resources, IPAM, etc.). Going forward, would it make sense to
>>>>> push some of the common functionality into the Quantum service, and have
>>>>> plugins which actually behave like the name suggests?
>>>>>
>>>>>
>>>>>
>>>>> Thanks,
>>>>>
>>>>> ~Sumit.
>>>>>
>>>>>
>>>>>
>>>>> *From:* netstack-bounces+snaiksat=*cisco.com at lists.launchpad.net*<cisco.com at lists.launchpad.net>
>>>>> [mailto:*netstack-bounces+snaiksat* <netstack-bounces%2Bsnaiksat>=
>>>>> *cisco.com at lists.launchpad.net* <cisco.com at lists.launchpad.net>] *On
>>>>> Behalf Of *Yong Sheng Gong*
>>>>> Sent:* Monday, July 30, 2012 7:05 PM*
>>>>> To:* Willian Molinari*
>>>>> Cc:* OpenStack Development Mailing List; *
>>>>> netstack at lists.launchpad.net* <netstack at lists.launchpad.net>*
>>>>> Subject:* Re: [Netstack] [Quantum] plugin -> backend
>>>>>
>>>>>
>>>>>
>>>>> Hi,
>>>>> Add it into openstack-dev and [quantum] into the subject.
>>>>>
>>>>> Yes, 'backend' seems better than 'plugin' for our case here.
>>>>>
>>>>> Our plugin is a must for quantum server to work, while 'plugin'
>>>>> tends to make us think it will provide more functionalities if we plug it
>>>>> in.
>>>>> And I don't think our plugin is 'pluggable backend'. I prefer to
>>>>> call it 'replaceable or configurable' 'backend' or 'dirver'.
>>>>>
>>>>> Thanks
>>>>> Yong Sheng Gong
>>>>>
>>>>>
>>>>> *
>>>>> **-----netstack-bounces+gongysh=cn.ibm.com at lists.launchpad.net*<-----netstack-bounces+gongysh=cn.ibm.com at lists.launchpad.net> wrote:
>>>>> -----
>>>>>
>>>>> To: *"netstack at lists.launchpad.net"* <netstack at lists.launchpad.net>
>>>>> *<netstack at lists.launchpad.net>* <netstack at lists.launchpad.net>
>>>>> From: Willian Molinari
>>>>> Sent by: *netstack-bounces+gongysh=cn.ibm.com at lists.launchpad.net*<netstack-bounces+gongysh=cn.ibm.com at lists.launchpad.net>
>>>>> Date: 07/31/2012 07:26AM
>>>>> Subject: [Netstack] plugin -> backend
>>>>>
>>>>> Æ!!
>>>>>
>>>>> Hi folks!
>>>>>
>>>>> I was concerned to bring the "plugins" discussion because it looks
>>>>> like a bikeshedding
>>>>> and it probably was discussed before, but I think it will be
>>>>> beneficial at all.
>>>>>
>>>>> What motivated me to bring the discussion was the Metaplugin
>>>>> implementation
>>>>> (*https://review.openstack.org/#/c/10181/*<https://review.openstack.org/#/c/10181/>)
>>>>> that looks like a quantum backend implementing
>>>>> support for plugins.
>>>>>
>>>>> When we first looked into quantum we thought that quantum plugin
>>>>> was following the same
>>>>> concept of all other plugins (ie we should install a lot of
>>>>> plugins to enhance the application)
>>>>> but we found that this is not the concept of quantum plugins,
>>>>> talking to Dan about this at
>>>>> the openstack summit I found the real concept of quantum plugins
>>>>> and I heard some people
>>>>> saying that plugins should be something like a "pluggable
>>>>> backend", so why not to call the
>>>>> plugin just "backend"?
>>>>>
>>>>> Looks natural to have just one backend at time and this backend
>>>>> should handle multiple
>>>>> plugins if needed (the metaplugin case).
>>>>>
>>>>> Sorry for bringing a non-technical discussion like this but every
>>>>> time someone asks me to
>>>>> explain what quantum does I need to show plugins as "backends" to
>>>>> make sense.
>>>>>
>>>>> I'm the only guy that think it's confusing? :P
>>>>>
>>>>> Just want to hear your ideas about this topic.
>>>>>
>>>>> --
>>>>> Willian Molinari
>>>>> (a.k.a PotHix)
>>>>>
>>>>> --
>>>>> Mailing list: *https://launchpad.net/~netstack*<https://launchpad.net/%7Enetstack>
>>>>> Post to : *netstack at lists.launchpad.net*<netstack at lists.launchpad.net>
>>>>> Unsubscribe : *https://launchpad.net/~netstack*<https://launchpad.net/%7Enetstack>
>>>>> More help : *https://help.launchpad.net/ListHelp*<https://help.launchpad.net/ListHelp>
>>>>>
>>>>> --
>>>>> Mailing list: *https://launchpad.net/~netstack*<https://launchpad.net/~netstack>
>>>>> Post to : *netstack at lists.launchpad.net*<netstack at lists.launchpad.net>
>>>>> Unsubscribe : *https://launchpad.net/~netstack*<https://launchpad.net/~netstack>
>>>>> More help : *https://help.launchpad.net/ListHelp*<https://help.launchpad.net/ListHelp>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>>> Dan Wendlandt
>>>>> Nicira, Inc: *www.nicira.com* <http://www.nicira.com/>
>>>>> twitter: danwendlandt
>>>>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>>> _______________________________________________
>>>>> OpenStack-dev mailing list
>>>>> OpenStack-dev at lists.openstack.org
>>>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>>>
>>>>>
>>>>> --
>>>>> Mailing list: https://launchpad.net/~netstack
>>>>> Post to : netstack at lists.launchpad.net
>>>>> Unsubscribe : https://launchpad.net/~netstack
>>>>> More help : https://help.launchpad.net/ListHelp
>>>>>
>>>>>
>>>>
>>>> _______________________________________________
>>>> OpenStack-dev mailing list
>>>> OpenStack-dev at lists.openstack.org
>>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>>
>>>>
>>>
>>>
>>> --
>>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>> Dan Wendlandt
>>> Nicira, Inc: www.nicira.com
>>> twitter: danwendlandt
>>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>
>>>
>>> _______________________________________________
>>> OpenStack-dev mailing list
>>> OpenStack-dev at lists.openstack.org
>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>>>
>>>
>>
>> --
>> Mailing list: https://launchpad.net/~netstack
>> Post to : netstack at lists.launchpad.net
>> Unsubscribe : https://launchpad.net/~netstack
>> More help : https://help.launchpad.net/ListHelp
>>
>>
>
> --
> Mailing list: https://launchpad.net/~netstack
> Post to : netstack at lists.launchpad.net
> Unsubscribe : https://launchpad.net/~netstack
> More help : https://help.launchpad.net/ListHelp
>
>
--
~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dan Wendlandt
Nicira, Inc: www.nicira.com
twitter: danwendlandt
~~~~~~~~~~~~~~~~~~~~~~~~~~~
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20120806/9bbbe4f2/attachment-0001.html>
More information about the OpenStack-dev
mailing list