13:03:00 <alexpilotti> #startmeeting hyper-v
13:03:00 <openstack> Meeting started Wed Jan 27 13:03:00 2016 UTC and is due to finish in 60 minutes.  The chair is alexpilotti. Information about MeetBot at http://wiki.debian.org/MeetBot.
13:03:01 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
13:03:03 <openstack> The meeting name has been set to 'hyper_v'
13:03:05 <lpetrut> Hi
13:03:08 <itoader> o/
13:03:10 <itoader> Hi
13:03:18 <alexpilotti> morning folks
13:03:24 <primeministerp> o/
13:03:29 <atuvenie> o/
13:03:33 <claudiub> o/
13:03:45 <alexpilotti> kvinod: sonu and sagar are joining us today?
13:03:47 <kvinod> hi all
13:04:07 <kvinod> alexpilotti: Sagar will join in 10 min and Sonu will not be able to make it
13:04:12 <sagar_nikam> Hi All
13:04:22 <kvinod> oh Sagar is in
13:04:23 <sagar_nikam> just a bit late
13:04:48 <alexpilotti> hi all!
13:04:54 <alexpilotti> #topic FC
13:05:05 <alexpilotti> sagar_nikam: any updates on the reviews by any chance?
13:05:47 <sagar_nikam> alexpilotti: Kurt and Hemna are in cinder midcycle meetup
13:05:54 <sagar_nikam> hence reviews are delayed
13:06:07 <sagar_nikam> will ping them again
13:06:14 <alexpilotti> tx
13:07:04 <alexpilotti> about midcycle, claudiub is at the Nova one. Any updates already worth sharing here?
13:07:22 <claudiub> hello
13:07:33 <claudiub> a loooot of talk about the scheduler and resource pools basically
13:07:47 <alexpilotti> #topic Nova midcycle updates
13:07:56 <claudiub> apparently, in the future the scheduler might be a separate project itself
13:08:16 <alexpilotti> wow, right about time, if you ask me :)
13:08:24 <claudiub> as there is talk about networking resource pools and volume resource pools, basically cinder and neutron to use the same scheudler
13:08:47 <alexpilotti> sweet
13:09:01 <claudiub> anyways, using the jaypipes' resource pools, the storage problem on the cluster will be solved.
13:09:07 <alexpilotti> any target release? e.g. N or O?
13:09:32 <claudiub> but the thing is, it a 7 part spec, so, for M, they say the first 3 will be implemented
13:09:48 <claudiub> the rest will be for N.
13:10:03 <alexpilotti> ok, so our nephews will use it, got it :)
13:10:50 <alexpilotti> claudiub: are those first 3 realistically merging in M?
13:11:10 <sagar_nikam> any discussion on drivers ? separate project for nova drivers ? like we have in neutron
13:11:13 <claudiub> other than that, a lot of discussion for libvirt's live migration, they are going to move away from the current implementation with ssh, which is slow
13:11:33 <claudiub> alexpilotti: yeah, that is the plan
13:11:58 <alexpilotti> claudiub: hyper-v has by far the best live migration, hopefully that wont require bit changes on our side
13:12:03 <claudiub> alexpilotti: but i don't think those 3 will be enough to solve the problem, as the last part is the scheduler part, which is the most important for solving this issue
13:12:25 * alexpilotti sobs
13:12:37 <claudiub> as for splitting the drivers, no, same things, unfortunately
13:12:48 <sagar_nikam> ok
13:12:50 <claudiub> although, there's a new project on the block, called os-vif
13:12:50 <alexpilotti> claudiub: any other relevant updates from the midcycle front?
13:13:07 <claudiub> which is going to refactor the vif stuff from nova and neutron into a separate project
13:13:26 <claudiub> so, I think any ovs vif plugging we have in nove would go there.
13:13:42 <sagar_nikam> ok
13:13:53 <sagar_nikam> from which release ? N or O ?
13:14:02 <claudiub> as for other updates, specs for N are open, going to try to get live-resize approved.
13:14:51 <claudiub> sagar_nikam: you mean the os-vif? there are still a couple of issues that have to be addressed before it officially replace the vifs in nova
13:15:07 <sagar_nikam> ok
13:15:16 <claudiub> so, it's a work in progress at the moment
13:16:08 <claudiub> as for other things, for the second day, its been mostly cross-project topics with neutron and cinder mostly
13:16:53 <claudiub> i guess one thing that we'll have to take care of in the future, is the neutron hyper-v ml2 driver
13:17:13 <claudiub> as it will have to be changed, to return an os-vif versioned object
13:17:33 <claudiub> from what I understood.
13:17:59 <claudiub> questions?
13:19:33 <alexpilotti> if there are no other questions, I'd go w the next topic
13:19:34 <sagar_nikam> no from my side
13:19:37 <alexpilotti> cool
13:20:02 <alexpilotti> #topic networking-hyperv scalability
13:20:35 <alexpilotti> the pymi + threading patches are pretty good to go
13:20:55 <sagar_nikam> ok
13:21:05 <alexpilotti> we're keeping the original BP that kvinod created
13:21:22 <alexpilotti> since anyway it does not refer to implementation details
13:21:24 <sagar_nikam> tested for all scenarios ? will we now have a version of pyMI ?
13:21:33 <kvinod> cool
13:22:15 <kvinod> what I understand from claudiub is that the existing rst file will be modified with the implemented approach
13:22:15 <alexpilotti> sagar_nikam: we are testing on all the scenarios that we already discussed
13:22:34 <sagar_nikam> ok
13:22:41 <alexpilotti> kvinod: yep claudiub is rewriting the spec to match the current implementation
13:22:47 <claudiub> kvinod: yeah, I knew I was forgetting something last night. :(
13:23:25 <alexpilotti> sagar_nikam: do you think you could test the patches in the meantime?
13:23:38 <alexpilotti> we're also testing sonu's patch
13:23:48 <kvinod> I have already conveyed this in review link but wanted to bring this up again
13:23:55 <sagar_nikam> kvinod: thala: testing, can you add these patches
13:24:15 <sagar_nikam> in that way we can check the perf with pyMI and the patches
13:24:34 <kvinod> sagar_nikam: sure will ask thala he is on leave till Thursday
13:25:02 <sagar_nikam> kvinod:ok, please let him know that the patches need to be included
13:25:22 <sagar_nikam> his tests are on liberty, so the patches needs to be liberty based
13:25:43 <sagar_nikam> kvinod: some work needs to be done before it is given to thala
13:25:47 <alexpilotti> alinb tested it, found some issues but most probably due to the that fact that sonu's neutron patch was not there on the neutron side
13:26:00 <kvinod> my comment on native thread was about using the existing patches which were already committed rather then creating new one
13:26:25 <alexpilotti> so we'll provide a complete update ASAP as soon as we test w the updated neutron master
13:26:56 <alexpilotti> kvinod: which patches?
13:27:30 <sagar_nikam> alexpilotti: thala will be testing liberty + pyMI, since that is what we requested him last meeting. does the same plan hold good today as well ?
13:28:13 <alexpilotti> claudiub: we can use networking-hyperv master on liberty, correct?
13:28:15 <kvinod> alexpilotti:https://review.openstack.org/#/c/235793/
13:28:42 <kvinod> we could have used this one rather than creating a new one
13:29:00 <claudiub> alexpilotti: Don't really think so, requirements are different, so it could conflict with nova requirements.
13:30:03 <alexpilotti> kvinod: the implementation differs
13:30:39 <kvinod> yes agreed but the concept and few file contents were same
13:30:43 <alexpilotti> claudiub: since they are testing on liberty, does it make sense to prepare a backport now?
13:31:20 <kvinod> we could have uploaded the new patches on top of existing one with new implementation approach
13:31:39 <claudiub> backport what?
13:31:56 <claudiub> sorry, my attention is split in 3 directions. :(
13:31:57 <kvinod> the changes were already in master
13:32:04 <claudiub> the native threads ones?
13:32:14 <alexpilotti> the liberty native threads patches
13:32:54 <claudiub> well, the native threads implementation is fairly straightforward and simple, we could.
13:33:40 <kvinod> claudiub: agreed, the implementation is straight and good
13:33:43 <alexpilotti> since HP has already a scheduled test run, we can add it on top, w/o having to wait for another test run
13:34:38 <alexpilotti> sagar_nikam: when is thala going to do run the tests?
13:35:58 <sagar_nikam> alexpilotti: probably on friday, we should be able to delay by one or two days
13:36:19 <sagar_nikam> is monday a good day to start the tests ? india monday morning
13:36:21 <kvinod> alexpilotti: Thala scheduled the test and went on leave, we will login to his setup and collect the results
13:37:08 <alexpilotti> kvinod: can we add some bits to networking-hyperv or it's "sealed"?
13:37:13 <sagar_nikam> kvinod: the next tests-- with the patches, can it be re-run on monday ? setup available on monday
13:37:22 <kvinod> we did not had his setup details so was not able to collect the results
13:38:42 <kvinod> alexpilotti: Anyways the test might have completed by now, if required we can bring in the changes and schedule the test again
13:39:02 <kvinod> Probably on Friday or Monday
13:39:05 <sagar_nikam> kvinod: that will be good
13:39:28 <alexpilotti> kvinod: some Hyper-V Windows updates came in very recently that apparently fixed the VLAN issue
13:39:38 <sagar_nikam> alexpilotti: are there any patches which you want to apply for the next run of tests ?
13:40:01 <alexpilotti> alinb has to come back with details, in case it'd be very important to have it in your setup
13:40:21 <alexpilotti> sagar_nikam: the networking-hyperv liberty backport if we have enough time to prepare them :-)
13:40:27 <kvinod> alexpilotti: so apart from native thread patch you want to pull windows recent update?
13:40:49 <alexpilotti> kvinod: yes, the goal is to run always on 100% udpated OSs
13:40:58 <kvinod> ok
13:41:06 <sagar_nikam> alexpilotti: we should be able to wait for some time, by when will those patches be available ?
13:41:07 <alexpilotti> there were a few prereq that I sent to thala
13:41:35 <kvinod> alexpilotti:ok
13:41:59 <alexpilotti> claudiub or alinb will backport them, by early next week we can have them
13:42:35 <kvinod> alexpilotti:so thala already knows which and how to pull the windows update
13:42:35 <sagar_nikam> kvinod: i think we should delay thala: next tests till early next week
13:42:38 <alexpilotti> depends really on when you want to run the tests
13:43:04 <alexpilotti> we run scenario tests very often, so it does not really matter
13:43:27 <sagar_nikam> i think the next tests should have the latest windows updates and the patches
13:43:33 <kvinod> sagar_nikam: then in that case let us know when to trigger the test
13:43:34 <alexpilotti> on the other side since your planning takes more time, we prefer to be flexible around your scheduling
13:44:09 <alexpilotti> kvinod: and the high perf powercfg scheme
13:44:18 <sagar_nikam> kvinod: lets wait to hear from alexpilotti: when the patches are ready and then use them
13:44:49 <kvinod> alexpilotti: sagar_nikam: fine
13:44:53 <alexpilotti> sagar_nikam cool!
13:45:09 <alexpilotti> next topic
13:45:18 <sagar_nikam> alexpilotti: as soon the patches are ready, let us know
13:45:22 <alexpilotti> #topic Hyper-V cluster
13:45:40 <kvinod> alexpilotti: wanted to know when you are planning to merge the Enhanced RPC Changes done by Sonu
13:46:05 <alexpilotti> kvinod: as soon as we have tests results from alinb
13:46:15 <kvinod> Blueprint spec is created and commit message is modified
13:46:31 <kvinod> ok, in the mean time I will upload the .rst file
13:46:36 <alexpilotti> I wrote you above that we had issues but most probably due to the missing neutron patch that just merged
13:46:47 <kvinod> ok
13:46:52 <alexpilotti> kvinod: no need for an RST for this one
13:47:15 <alexpilotti> it's quite straight forward
13:47:16 <kvinod> alexpilotti: fine, will not upload it
13:47:24 <kvinod> i mean the rst file
13:47:36 <alexpilotti> thanks
13:47:47 <claudiub> kvinod: also, unit tests should be added on the enhanced rpc patch.
13:47:47 <alexpilotti> back to the topic
13:48:04 <alexpilotti> sonu wrote that he's adding them
13:48:40 <kvinod> claudiub: I added one to test the newly added function and committed
13:49:00 <alexpilotti> folks: we have a different topic now :-)
13:49:17 <kvinod> claudiub: please review and let me know anything additional required
13:49:35 <sagar_nikam> alexpilotti: how is the cluster driver progressing ?
13:49:40 <alexpilotti> here we go
13:49:40 <kvinod> sorry, please continue with topic
13:49:46 <alexpilotti> thanks
13:50:13 <alexpilotti> atuvenie is porting it from the original kilo implementation to mitaka
13:50:23 <alexpilotti> some important areas
13:51:02 <alexpilotti> volumes: ensuring that passthrough volumes are logged in the target host during a failover is quite tricky
13:51:22 <alexpilotti> the first implementation will support SMB3 volumes
13:51:46 <alexpilotti> which is also MSFT's recommended approach
13:52:29 <alexpilotti> for iSCSI / FC, we're investigating how to trigger an event on the target and block the VM start until all LUNs are mounted
13:52:57 <alexpilotti> in short, replicating what happens for live migration
13:53:50 <alexpilotti> on the other side, the interaction with Nova became easier since we dont have to explicitly call into the conductor
13:54:11 <alexpilotti> thanks to the new versioned objects, this becomes more transparent
13:54:46 <alexpilotti> the patches will be ready in the short term (February)
13:55:23 <sagar_nikam_> i got disconnected
13:55:26 <sagar_nikam_> i am back
13:55:35 <primeministerp> welcome back
13:55:37 <primeministerp> ;)
13:55:46 <sagar_nikam_> alexilotti: what about iscsi and FC volumes ?
13:56:02 <alexpilotti> sagar_nikam_: talked about it while you were disconnected :-)
13:56:04 <sagar_nikam_> for cluster driver
13:56:08 <primeministerp> sagar_nikam_, :<alexpilotti> for iSCSI / FC, we're investigating how to trigger an event on the target and block the VM start until all LUNs are mounted
13:56:08 <sagar_nikam_> ok
13:56:28 <alexpilotti> sagar_nikam_: in short: SMB3 now, iSCSI / FC later
13:56:42 <sagar_nikam_> ok
13:57:03 <sagar_nikam_> later as in Mitaka or N ?
13:57:11 <sagar_nikam_> in compute-hyperv ?
13:57:13 <alexpilotti> we need also to contain the patches if we want to hope for merging them in N or O
13:57:35 <alexpilotti> they will go straight in compute-hyperv in Mitaka of course
13:57:58 <alexpilotti> and we might even think about a Liberty backport if customers ask, but that's a separate topic
13:58:19 <sagar_nikam_> alexpilotti: in Mitaka, compute-hyperv will have support for iscsi and FC for cluster driver ?
13:58:28 <alexpilotti> we're alsmost out of time
13:58:44 <alexpilotti> sagar_nikam_: depends if we find a feasible solution
13:58:50 <sagar_nikam_> ok
13:58:57 <alexpilotti> sagar_nikam_: the issue is with how the clustering works
13:59:14 <sagar_nikam_> ok
13:59:21 <alexpilotti> the VM gets migrated before we can trigger an event and mount the LUNs on the target
13:59:25 <sagar_nikam_> lets discuss in detail in next meeting
13:59:31 <alexpilotti> resulting in I/O issues
13:59:37 <sagar_nikam_> we are almost out of time now
13:59:43 <alexpilotti> we have a few ideas in an embryonal state
13:59:52 <sagar_nikam_> can we have this topic as first topic next meeting ?
13:59:58 <alexpilotti> which require some C/C++ native components on the cluster side
14:00:11 <alexpilotti> sagar_nikam_: sure
14:00:24 <alexpilotti> ok guys time's up!
14:00:31 <alexpilotti> thanks for joining!
14:00:33 <kvinod> thanks, bye
14:00:34 <sagar_nikam_> bye
14:00:37 <alexpilotti> #endmeeting