On 13-Oct-2017, at 10:52 AM, Anil Vishnoi <[email protected]> wrote:



On Thu, Oct 12, 2017 at 9:34 PM, Shuva Kar <[email protected]>
wrote:

>
> On 13-Oct-2017, at 9:20 AM, Abhijit Kumbhare <[email protected]>
> wrote:
>
> First of all - yes - CSIT needs to be added for the hitless resync.
> Secondly hitless resync is different from bundles. Hitless resync is a
> reconciliation mechanism which uses bundles in order to be - errr hitless.
> The basic idea is that the switch will continue using the original flow
> table/s, the controller replays the flows/groups from the config into a
> bundle, the switch receives the bundle in a copy of flow table/s - and once
> the bundle programming is done makes the copy active. Basically to avoid
> datapath disruption during reconciliation.
>
> The bundles feature as present is available from OVS 2.6 and above.
>
> Gobinath has added a reconciliation framework which would allow to plug in
> a different resync mechanism in the patches:
>
> https://git.opendaylight.org/gerrit/#/q/owner:gobinath%2540e
> ricsson.com+status:merged+project:openflowplugin
>
> I don't believe he has actually added the hitless resync (looking at the
> patches) - I will ask him to shed more light.
>
> About:
>
> - How do we verify this feature? is there a way to check bundle
> programming has really happened?
>
>
> As Abhijit rightfully mentioned, Hitless resync is a feature and ofp
> bundles is a way of implementing the same. To test this implementation, we
> need to
> 1. enable the hitless resent feature, its not enabled by default,
> 2. Provision some flows on a set of switches/end-points
> 3. Ensure that there is traffic flowing through them, ping could be used
> 4. Cause a tcp disconnect between the controller and the switches and then
> connect back
>  There should be no blip in traffic . The only thing to ensure is that Arp
> entries on the switches do not timeout when the control plane is dissected.
>
​Not sure, if this is really a right test to do or not. Because traffic
blip will depend on when bundle is committed on each switch. You can use
bundle to commit all the flow/groups atomically, but doing it across the
switches is controllers job, and you can still see the disruption in the
traffic. I think the only test we can do that will verify that bundle based
flow/installation is working is, to run all the test suits with
bundle-based-reconciliation-enabled=true. So ideally after enabling
bundles, we should not see any behavioral change in the existing tests.​


>
>
Absolutely correct. I was speaking from an E2E perspective, my bad! Wearing
the wrong hat yet again !!!

Correct ! Behaviourally there should not be any change before and after
enabling the hitless resync. The applications on the controller should take
care of pushing the flows/groups across the switches on sensing a switch
connect.

>From the plugin perspective, we can just check, if the bundle based
provisioning/de-provisioning is working correctly.

Hope that helps :)
>
>
> We would need to have tests verifying reconciliation with zero datapath
> disruption - so we will need to think about it a bit.
>
>
> On Thu, Oct 12, 2017 at 1:15 PM, Anil Vishnoi <[email protected]>
> wrote:
>
>>
>>
>> On Thu, Oct 12, 2017 at 10:53 AM, Luis Gomez <[email protected]> wrote:
>>
>>> I have few questions on this feature:
>>>
>>> - Is this feature enabled by default?
>>>
>> ​Yes, its a feature, but as of now no application is basically using it.
>> We do have a sample application that shows  how user can use this feature.
>> https://github.com/opendaylight/openflowplugin/tree/master/s
>> amples/sample-bundles
>> ​
>>
>>
>>> - Which OVS version supports bundles? I believe we currently use 2.5.2
>>> in latest tools VM
>>>
>> ​I believe 2.5.2 don't have support, but i am not sure. ​
>>
>>
>>> - How do we activate bundle programming? is it enough to push a bunch of
>>> flows/group to FRM using REST?
>>>
>> ​I will get back to you on that.
>>
>>> - How do we verify this feature? is there a way to check bundle
>>> programming has really happened?
>>>
>> ​Bundle failure generates the error, but i think the easiest way probably
>> is to program different set of flows with bundle and verify that all the
>> flows are present in the switch.
>>
>>
>>>
>>> BR/Luis
>>>
>>>
>>> On Oct 12, 2017, at 10:26 AM, Anil Vishnoi <[email protected]>
>>> wrote:
>>>
>>> No, there is no CSIT test for bundles.
>>>
>>> On Thu, Oct 12, 2017 at 10:15 AM, Jamo Luhrsen <[email protected]>wr
>>> ote:
>>>
>>>> it's the thing that uses openflow bundles. I didn't follow the dev cycle
>>>> with it, but I get the sense that it works now.
>>>>
>>>> https://wiki.opendaylight.org/view/OpenDaylight_OpenFlow_Plu
>>>> gin:Hitless_resync
>>>>
>>>> JamO
>>>>
>>>> On 10/12/2017 10:04 AM, Luis Gomez wrote:
>>>> > What is hitless resync in the OF plugin context?
>>>> >
>>>> >> On Oct 12, 2017, at 8:22 AM, Jamo Luhrsen <[email protected]>
>>>> wrote:
>>>> >>
>>>> >> Hi OFP,
>>>> >>
>>>> >> We talked a little about this hitless resync feature from
>>>> openflowplugin. It's of
>>>> >> interest to netvirt. I'm wondering if there are any CSIT jobs that
>>>> validate it in
>>>> >> the openflowplugin CSIT jobs?
>>>> >>
>>>> >> We'll be adding some netvirt test cases that will also test it, but
>>>> just from a
>>>> >> higher level than maybe it would be tested at your level.
>>>> >>
>>>> >> Thanks,
>>>> >> JamO
>>>> >
>>>> _______________________________________________
>>>> openflowplugin-dev mailing list
>>>> [email protected]
>>>> https://lists.opendaylight.org/mailman/listinfo/openflowplugin-dev
>>>>
>>>
>>>
>>>
>>> --
>>> Thanks
>>> Anil
>>>
>>>
>>>
>>
>>
>> --
>> Thanks
>> Anil
>>
>> _______________________________________________
>> openflowplugin-dev mailing list
>> [email protected]
>> https://lists.opendaylight.org/mailman/listinfo/openflowplugin-dev
>>
>>
> _______________________________________________
> openflowplugin-dev mailing list
> [email protected]
> https://lists.opendaylight.org/mailman/listinfo/openflowplugin-dev
>
>
>


-- 
Thanks
Anil
_______________________________________________
openflowplugin-dev mailing list
[email protected]
https://lists.opendaylight.org/mailman/listinfo/openflowplugin-dev

Reply via email to