13:00:39 <lpetrut> #startmeeting hyper-v
13:00:40 <openstack> Meeting started Wed Feb  3 13:00:39 2016 UTC and is due to finish in 60 minutes.  The chair is lpetrut. Information about MeetBot at http://wiki.debian.org/MeetBot.
13:00:42 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
13:00:45 <openstack> The meeting name has been set to 'hyper_v'
13:01:10 <lpetrut> Hi guys. Alessandro will join us in a few minutes, but we can start the meeting now
13:01:28 <itoader> Hi
13:01:29 <abalutoiu> Hi
13:02:51 <sagar_nikam> Hi All
13:02:51 <lpetrut> sagar_nikam: anyone else from HP to join the meeting?
13:03:07 <sagar_nikam> lpetrut: me and vinod
13:03:11 <sagar_nikam> we can start
13:03:16 <kvinod> Hi sagar_nikam
13:03:39 <primeministerp> hi guys
13:04:21 <lpetrut> great, sure. so, as I previously said, Alessandro will join us in a few minutes. Would you like to propose a topic to begin with?
13:04:41 <sagar_nikam> yes, cluster driver
13:04:54 <lpetrut> #topic Hyper-v clustering
13:04:55 <kvinod> I have following topic from neutron front
13:05:08 <kvinod> Native thread Fixes in liberty branch
13:05:08 <sagar_nikam> we started that discussion last week towards the end of the meeting
13:05:20 <kvinod> Enhanced RPC merge?
13:05:33 <kvinod> OVS kernel panic
13:05:50 <lpetrut> yep, so, on the clustering topic: Adelina was working on this. She said that she got in touch with you
13:06:10 <sagar_nikam> yes, i did ping Adelina yesterday
13:06:22 <sagar_nikam> we are setting up the hyperv cluster
13:06:29 <sagar_nikam> should be ready in couple of days
13:06:47 <sagar_nikam> once that is ready, we can start testing the cluster driver
13:07:17 <sagar_nikam> only issue, our setup is liberty, so it would be good if we have the cluster driver which is liberty based
13:07:23 <lpetrut> that's great. unfortunately, she's not in the meeting right now, otherwise it would've been useful if she could've provide some updates on the status
13:07:33 <sagar_nikam> sure
13:07:58 <sagar_nikam> she mentioned to me that liberty based code can be provided
13:08:03 <sagar_nikam> so waiting for that
13:08:03 <alexpilotti> hey guys
13:08:14 <sagar_nikam> hi alexpilotti:
13:08:25 <atuvenie> Hi all, sorry, my internet is acting up
13:09:10 <sagar_nikam> alexpilotti: we are discussing the cluster driver
13:09:25 <alexpilotti> cool, was just back reading
13:09:33 <sagar_nikam> as mentioned few mins back, we can start testing it
13:09:49 <sagar_nikam> currently setting up the hyperv cluster on 2 machines
13:10:01 <alexpilotti> sagar_nikam: great!
13:10:02 <sagar_nikam> i did chat with adelina yesterday
13:10:09 <sagar_nikam> on this topic
13:10:19 <alexpilotti> atuvenie told me that you guys spoke
13:10:33 <sagar_nikam> our setup is liberty based, so it would be good if that patch is liberty based
13:10:40 <sagar_nikam> spoke about it to adelina
13:10:56 <sagar_nikam> once she provides it, we will start
13:11:00 <atuvenie> I am currently working on master, a liberty patch may be a little bit tricky given no os-win there
13:11:08 <alexpilotti> sagar_nikam they introduced changes (versioned objects) afterwards that are not in Liberty
13:11:25 <alexpilotti> so adapting the patch to Liberty requires extra work
13:11:41 <sagar_nikam> ok
13:11:52 <sagar_nikam> all our setups are liberty based
13:12:05 <sagar_nikam> difficult to find a mitaka setup
13:12:21 <sagar_nikam> will it be very difficult to rebase ?
13:12:36 <sagar_nikam> or, we had a liberty patch sometime back
13:12:40 <sagar_nikam> does it work ?
13:12:47 <alexpilotti> sagar_nikam: non dramatically difficult,  but it requires work
13:12:52 <sagar_nikam> we can pick that
13:13:12 <alexpilotti> you can take the mitaka patches and backport them sure
13:13:37 <alexpilotti> on our side we're evaluating if we are going to put them in compute-hyperv liberty
13:13:46 <sagar_nikam> ok
13:14:11 <sagar_nikam> i can pick up from compute-hyperv liberty whenever it is ready (if you decide that)
13:14:15 <alexpilotti> but we surely won't backport on Nova (as part of a community effort, at least)
13:14:51 <sagar_nikam> not a issue, if not backported to nova
13:15:05 <sagar_nikam> we just need a patch which works in liberty
13:15:07 <alexpilotti> anyway first we need it to get it top notch on Mitaka :-)
13:15:16 <sagar_nikam> right
13:15:38 <sagar_nikam> i remember that there was a patch in Liberty , does it not work ?
13:15:51 <alexpilotti> even kilo
13:16:04 <alexpilotti> we did it initially in kilo
13:16:07 <sagar_nikam> if that works, we can start with it
13:16:44 <alexpilotti> we can definitely share those patches as well
13:16:57 <sagar_nikam> sure
13:17:06 <alexpilotti> although back then it was a separate effort, as it was not merged in compute-hyperv
13:17:17 <sagar_nikam> let us know which will be better to test, liberty-nova or compute-hyperv-liberty
13:17:24 <sagar_nikam> we can wait
13:17:54 <alexpilotti> We'll send you an update as soon as we are done w the Mitaka one, if that works for you
13:18:09 <sagar_nikam> sure
13:18:23 <alexpilotti> can we move to the next topic?
13:18:29 <sagar_nikam> question on networking in cluster driver
13:18:40 <alexpilotti> sure
13:18:47 <sagar_nikam> if the instance is moved from host H1 to host H2
13:18:56 <sagar_nikam> done outside of nova
13:18:56 <sagar_nikam> by
13:19:03 <sagar_nikam> failover cluster
13:19:15 <sagar_nikam> the instance will retain the same ip
13:19:26 <sagar_nikam> we just need the switch on both hosts
13:19:33 <sagar_nikam> of same name
13:19:43 <sagar_nikam> is my understanding correct ?
13:20:15 <alexpilotti> well, the networking is fairly transparent
13:20:29 <alexpilotti> and similar to what happens in live migration
13:20:50 <sagar_nikam> ok
13:20:59 <alexpilotti> the networking-hyperv agent (or the OVS one) pick up the new port(s)
13:21:29 <sagar_nikam> and IP and security rules will work as earlier ?
13:21:29 <alexpilotti> the cluster is used for the VMs, not for IPs or other resources
13:21:53 <alexpilotti> sure, IP and sec groups are still managed by Neutron
13:22:01 <sagar_nikam> ok
13:22:23 <sagar_nikam> kvinod: do you have any questions on networking in cluster driver ?
13:22:25 <alexpilotti> our goal is to make cluster support almost "invisible"
13:22:43 <sagar_nikam> alexpilotti: thats good
13:23:04 <alexpilotti> cool
13:23:06 <alexpilotti> next
13:23:07 <sagar_nikam> sonu: kvinod: any questions from your end
13:23:19 <alexpilotti> #topic RPC networking-hyperv patch
13:23:23 <Sonu> The case of OVS will be little different
13:23:37 <kvinod> yes on neutron front the neutron agent will have to start with binding as if if a port was created?
13:23:54 <Sonu> Sorry I am bit slow on network.
13:24:28 <kvinod> i mean after migration to H2 neutron agent will get a new port to prcess for which it will have to do the binding?
13:24:29 <alexpilotti> Sonu, kvinod: correct the port itself is created differently between OVS ad networking-hyperv
13:24:41 <alexpilotti> but even it that case it does not affect the cluster itself
13:24:53 <alexpilotti> it's very similar to the live migration use case
13:25:20 <kvinod> ok , so that menas neworking will not have any impact?
13:25:35 <kvinod> and not work required for neutron
13:25:38 <kvinod> ?
13:25:39 <Sonu> Yes. Nova vif must create an OVS port. Neutron will bind it in normal way
13:26:00 <alexpilotti> fully transparent
13:26:10 <kvinod> k
13:26:32 <alexpilotti> getting back to the new topic
13:26:36 <alexpilotti> RPC patch
13:26:57 <alexpilotti> Claudiu was last week at the midcycle meetup and is getting back to the office on Monday
13:27:08 <alexpilotti> but he told me that unit tests are missing on the patch
13:27:40 <alexpilotti> while abalutoiu tested the patch and results are all good
13:27:55 <Sonu> We covered some tests post claudius comment
13:27:55 <kvinod> last week I uploaded a patch set with some UT code
13:28:31 <Sonu> Request you to review the same and advise if more is needed.
13:29:32 <Sonu> Thanks for testing
13:29:43 * alexpilotti looks for the patch link..
13:29:44 <kvinod> alexpilotti: could you please convay message to Claudiu to have a look again and post comment about what additional test he is expecting
13:30:00 <alexpilotti> #link https://review.openstack.org/#/c/263865/
13:30:47 <alexpilotti> I see that you amended the commit msg and uploaded a BP, which is great
13:31:35 <claudiub> hello, just dropping in for a moment. the unit test coverage is not 100%,
13:31:42 <alexpilotti> approved BP and set to "needs review"
13:31:56 <claudiub> also, I've suggested to add some unit tests with some raw enhanced SG rules.
13:32:03 <kvinod> patch 4 had UT code
13:32:04 <Sonu> Thank you much
13:32:10 <alexpilotti> abalutoiu just conformed that he tested it again on top of the latest patchsets, so seems all good
13:32:26 <alexpilotti> hyper-v CI is happy as well
13:32:33 <kvinod> good
13:33:00 <primeministerp> woot
13:33:09 <claudiub> kvinod:  typically, you need to add a unit test per new method / function, covering all code branches
13:33:32 <claudiub> or more, if there are mutually exclusive code branches..
13:34:49 <Sonu> Vinod do you want to re check on this please.
13:34:53 <alexpilotti> claudiub welcome back! :)
13:35:03 <claudiub> o/
13:35:25 <kvinod> claudiub: will have a check again and do the needful
13:35:47 <alexpilotti> claudiub: to speed up the communication, could you please tell kvinod which are the methods / code paths that still need test coverage?
13:36:06 <claudiub> k, will re-review once the unit tests are submitted.
13:36:12 <kvinod> that would be great, thanks
13:36:48 <kvinod> claudiub: I hope you saw patchset 4
13:37:12 <kvinod> on top of it do let me know what additional UT is required
13:37:56 <claudiub> yeah, _generate_rules, _select_sgr_by_direction, _split_sgr_by_ethertype, update_security_group_rules are uncovered
13:38:41 <kvinod> ok, will cover those also
13:41:04 <alexpilotti> kvinod: in general we try to achieve max coverage
13:41:18 <alexpilotti> teh idea is that if there's a code path, a test needs to run it
13:41:27 <alexpilotti> next topic?
13:42:08 <Sonu> OVS core dump
13:42:38 <Sonu> That we faced on windows 2012 r2.
13:42:43 <alexpilotti> great
13:42:50 <alexpilotti> #topic OVS core dump
13:43:14 <alexpilotti> so, the dump shows that the error is in the underlying driver
13:44:30 <alexpilotti> and if it doesnt show up on our test environments (Mellanox, Chelsio, etc), we need to find a way to access a similar environment
13:44:44 <alexpilotti> or at least get a couple of Emulex cards
13:45:06 <primeministerp> alexpilotti I'm looking who to reach out for that as well
13:45:29 <Sonu> Ok
13:45:37 <alexpilotti> Sonu: any chance you could let us access it or send a few boxes to Cambridge so that primeministerp could add them to the test racks?
13:46:29 <alexpilotti> Sonu: is that going to be your reference architecture or just some "random hardware"?
13:46:36 <primeministerp> Sonu, we may only need the hba's
13:46:42 <Sonu> Random
13:47:05 <Sonu> I will find that out.
13:47:07 <alexpilotti> Sonu: do you have some other NICs around? E.g. mellanox or Intel?
13:47:26 <primeministerp> Sonu: this was basically what I was asking in the previous email exchange
13:47:27 <alexpilotti> how are the nics configured? e.g. teaming, etc
13:47:38 <Sonu> I will find that out.
13:48:06 <alexpilotti> Sonu since the error is in the driver, we might need to speak with the Emulex driver team
13:48:09 <Sonu> No team
13:48:21 <alexpilotti> at Emulex I mean :-)
13:48:56 <sagar_nikam> sonu: did you mean no NIC teaming ?
13:49:03 <Sonu> I understand. Let me try with other nic model.
13:49:20 <alexpilotti> it's a closed source driver, so we have no idea what happens in it, so debugging options are limited :-)
13:49:58 <Sonu> Sagar we did not team the nics yet for OVS
13:50:06 <alexpilotti> It could be that: 1) we send some perfectly fine packets that the diver is not handling properly or 2) or packets that we send have issues
13:50:35 <Sonu> It is very consistent, so most likely it is 2
13:51:04 <sagar_nikam> Sonu: consistent on different blades ?
13:51:04 <alexpilotti> Sonu: consistent across different NIC models?
13:51:08 <Sonu> With Windows 2012 r2 data center edition we faced this issue.
13:51:25 <sagar_nikam> Sonu: can you try on another blade
13:51:54 <Sonu> Sure.
13:52:11 <Sonu> Tried only on emulex.
13:52:30 <Sonu> Should get some qlogic adapters ad test.
13:52:36 <alexpilotti> I already asked aserdean on my OVS team to look into this, so we'll try to replicate it ASAP, but since it seems like a fairly straightforward use case, I'm not expecting much
13:53:02 <Sonu> Thx
13:53:59 <alexpilotti> anything else on this topic?
13:54:09 <alexpilotti> we have 7' to go
13:54:24 <sagar_nikam> in last week's meeting
13:54:25 <alexpilotti> lpetrut: did you speak about FC already?
13:54:48 <sagar_nikam> we discussed about thala using the liberty based patches for his scale testing
13:55:01 <sagar_nikam> he is still waiting for those patches
13:55:26 <lpetrut> alexpilotti: well, I have no updates related to the FC topic, other than the fact that the Nova patches have been -2ed because of the feature freeze
13:55:36 <alexpilotti> sagar_nikam: yeah, claudiub is working on that, but as I was saying he was off the last two weeks
13:56:17 <sagar_nikam> alexpilotti: sure, we will wait and test it when available
13:57:05 <kvinod> also, thala was not able to proceed with testing due to difficulty in porting native thread and enhanced rpc patch on top of stable liberty
13:57:42 <alexpilotti> kvinod: couldn't Thala just run a test without those patches for now?
13:58:23 <kvinod> No, the idea was to run the test with all updates and available patches
13:58:27 <sagar_nikam> alexpilotti: i think he was waiting based on the discussion we had last week
13:58:29 <alexpilotti> kvinod: also, the enhanced RPC one if I'm not mistaken requires a Neutron patch, which merged in master
13:58:37 <kvinod> and that is what we discussed in last IRC
13:59:12 <kvinod> neutraon dependent patch was also taken
13:59:35 <alexpilotti> kvinod: did you backpot it to livberty already?
13:59:44 <kvinod> so with all patches on top of liberty is causing exception
13:59:54 <alexpilotti> ah ok
14:00:17 <kvinod> no I thought of dicussing this today about the plan to backport
14:00:36 <kvinod> backporting to liberty
14:00:54 <alexpilotti> the neutron patch can already be backported I guess
14:01:06 <kvinod> do we have plan for this about backporting the patches to liberty
14:01:08 <sagar_nikam> should this topic be discussed in hyperv channel now ?
14:01:19 <sagar_nikam> since we are almost done with time
14:01:22 <alexpilotti> to stable/liberty in upstream Neutron, I mean
14:01:32 <alexpilotti> uh oh out of time folks :)
14:01:40 <lpetrut> #endmeeting