16:00:30 <primeministerp> #startmeeting hyper-v
16:00:31 <openstack> Meeting started Tue Jul  9 16:00:30 2013 UTC.  The chair is primeministerp. Information about MeetBot at http://wiki.debian.org/MeetBot.
16:00:32 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
16:00:35 <openstack> The meeting name has been set to 'hyper_v'
16:00:41 <primeministerp> hi everyone
16:00:43 <luis_fdez> hi
16:00:44 <primeministerp> it's been a while
16:00:48 <primeministerp> hey luis
16:01:21 <primeministerp> let's wait for the others
16:01:26 <primeministerp> I don't see alex here yet
16:01:58 <luis_fdez> ok
16:02:04 <primeministerp> luis_fdez: so let's start w/ the puppet modules while waiting
16:02:08 <luis_fdez> ok
16:02:11 <primeministerp> #topic hyper-v puppet modules
16:02:18 <schwicht> primeministerp: hi everyone
16:02:33 <luis_fdez> have you looked at the last pull requrest?
16:02:33 <primeministerp> i renamed the module and merged your pull requests last night
16:02:49 <primeministerp> luis_fdez: I haven't looked since last night
16:02:56 <primeministerp> luis_fdez: you had 2
16:03:02 <primeministerp> luis_fdez: i merged them both
16:03:09 <primeministerp> the changes to the vswitch and the rename
16:03:14 <luis_fdez> yes
16:03:15 <primeministerp> for module conformance
16:03:21 <primeministerp> I also renamed the project
16:03:23 <primeministerp> to match
16:03:25 <luis_fdez> perfect
16:03:36 <primeministerp> luis_fdez: so go ahead and add your other changes
16:03:51 <primeministerp> schwicht: Hi frank
16:04:04 <primeministerp> ociuhandu: hi tavi
16:04:04 <luis_fdez> ok, one of the pull request is just a refactoring creating a new hyper-v.pp class
16:04:08 <alexpilotti> hi guys
16:04:13 <primeministerp> luis_fdez: perfect
16:04:22 <luis_fdez> to extract the installation of Hyper-V and configuration .... it can be the future init.pp of the hyper-v module
16:04:24 <primeministerp> luis_fdez: i was thinking we need to move some into general windows
16:04:30 <primeministerp> class as well
16:04:42 <alexpilotti> sorry me and Tavi were in a meeting
16:04:49 <primeministerp> alexpilotti: no problem
16:04:51 <luis_fdez> the other one is a proofofconcetp? hehe of how the nova_config type looks as in nova puppetlabs... but it adds one dependency
16:04:57 <zehicle> Hello!
16:05:01 <ociuhandu> hi all
16:05:16 <primeministerp> luis_fdez: I need to add a puppetfile for dependencies
16:05:21 <luis_fdez> ((last time we talked about it I was wrong saying that it wont add new dependency))
16:05:40 <primeministerp> luis_fdez: I don't mind adding dependencies
16:05:49 <luis_fdez> also.. I extracted windows_feature as a separate define
16:05:53 <primeministerp> then we can reuse
16:06:08 <luis_fdez> to make the code cleaner and also follow the style rules
16:06:15 <primeministerp> luis_fdez: that works too, because if using hyper-v server the role is already there
16:06:36 <primeministerp> luis_fdez: perfect
16:06:58 <primeministerp> luis_fdez: is there anything else we need to discuss?
16:07:05 <luis_fdez> now... I'm looking to the download question and.... opened to new proposals... can i give you a hand on something related with python?
16:07:51 <alexpilotti> luis_fdez: any BP you'd like to work on? :-)
16:08:15 <primeministerp> luis_fdez: it needs to be there
16:08:15 <luis_fdez> I'll take a look... now... im a puppet man... i have to revisit my compute driver profile :)
16:08:19 <primeministerp> at least for the start
16:08:21 <primeministerp> or
16:08:25 <primeministerp> we have it as a seed
16:08:27 <primeministerp> also
16:08:33 <primeministerp> once we move to all source
16:08:41 <primeministerp> we won't necessarily need it
16:08:45 <primeministerp> bc we can move to a git pull
16:08:50 <primeministerp> for all that
16:08:51 <primeministerp> however
16:08:59 <primeministerp> if we're going to use the public bins
16:09:19 <primeministerp> we need to have a way to pull them
16:10:10 <primeministerp> that's my 2c
16:10:19 <luis_fdez> yeps, I think for the start the public bins is the best option
16:10:34 <primeministerp> ok
16:10:45 <primeministerp> let's move on topic wise
16:10:49 <luis_fdez> do you have changes to be committed or can I changes thinks on the python part?
16:10:49 <primeministerp> alexpilotti: ready
16:11:00 <luis_fdez> pasdfadf
16:11:00 <primeministerp> luis_fdez: no I've been working on the provisioning side
16:11:04 <primeministerp> had some stuff to clean up
16:11:04 <alexpilotti> sure
16:11:12 <primeministerp> alexpilotti: where do we want to start?
16:11:19 <primeministerp> wmiv2?
16:11:29 <alexpilotti> yep
16:11:37 <primeministerp> #topic wmiv2
16:11:53 <alexpilotti> so 2012R2 Preview is publicly avlaible
16:11:59 <primeministerp> haha
16:12:04 <primeministerp> why yes it is
16:12:10 <alexpilotti> and the V1 namespace is gone :-)
16:12:20 <primeministerp> why yes it is
16:12:22 <primeministerp> ;)
16:12:32 <alexpilotti> so it's time to do some testing of Havana on 2012R2
16:12:58 <alexpilotti> to get there, the first BP that we are releasing this week is teh WMIV2 one
16:13:01 <primeministerp> is the wmiv2 code upstream
16:13:08 <primeministerp> perfect
16:13:09 <alexpilotti> primeministerp: ?
16:13:13 <alexpilotti> ok
16:13:17 <primeministerp> you answered my question
16:13:43 <alexpilotti> thanks to the Grizzly refactoring it was a fairly contained work
16:14:13 <alexpilotti> a factory is instantiating the relevant utils classes based on the OS: < 2012: V1, >= 2012 V2
16:14:37 <alexpilotti> a separate pachset will be released for the Neutron agent
16:14:42 <primeministerp> awesome
16:14:52 <schwicht> nice work
16:14:55 <primeministerp> in the same timeframe?
16:14:58 <alexpilotti> one that's in place, we'll need also to update the installer
16:15:16 <alexpilotti> I'd love to get both (Nova, Quantum) in for H2
16:15:39 <primeministerp> ok
16:15:47 <primeministerp> that would be great
16:15:49 <alexpilotti> we'll need a lot of help in testing
16:15:57 <primeministerp> schwicht: .....
16:16:23 <primeministerp> schwicht: will your folks be able to take a look
16:16:25 <schwicht> primeministerp: we'll need to chat off line
16:16:30 <primeministerp> schwicht: np
16:16:32 <alexpilotti> one of the reasons I want to have those bits out early is that they replace the layer that interacts with the OS
16:16:55 <alexpilotti> which cannot be tested with unit testing, only system / integration testing
16:17:12 <primeministerp> alexpilotti: i'll be able to try them tomorrow
16:17:14 <alexpilotti> and sicne we need a CI for that, the only alternative is having some hands doing it ;-)
16:17:19 <schwicht> is the testing the WMIv2 testing?
16:17:27 <primeministerp> alexpilotti: yes I know
16:17:28 <alexpilotti> schwicht: yes
16:17:50 <alexpilotti> schwicht: basically doing a lot of regression testing
16:17:55 <schwicht> we will have both funtional (IBM speak IVT) and SVT testing for the base stuff
16:18:14 <alexpilotti> schwicht: great tx!
16:18:43 <schwicht> however we are investigating the scope of the testing atm
16:18:57 <primeministerp> gotcha
16:19:39 <primeministerp> so next topic?
16:19:47 <alexpilotti> Dynamic memory
16:19:50 <primeministerp> sure
16:19:56 <primeministerp> #topic dynamic memory
16:20:23 <alexpilotti> We have this patch ready as well, I'm reviewing it internally, but is basically ready for public review
16:20:25 <primeministerp> alexpilotti: i'm assuming you've completed it as well?
16:20:43 <alexpilotti> we have 2 new options:
16:20:57 <alexpilotti> one to enable / disable the feature
16:21:06 <alexpilotti> and one for the overcomit ratio
16:21:28 <alexpilotti> for teh second one, we just reused the option coming from teh scheduler filter
16:21:45 <schwicht> is the ratio part of the flavor ?
16:22:00 <alexpilotti> no, it's a configuration in the nova scheduler
16:22:06 <schwicht> oh
16:22:07 <alexpilotti> default's to 1.5
16:22:17 <alexpilotti> so if the host that 100GB memory
16:22:22 <schwicht> understood
16:22:23 <alexpilotti> it overcommits to 150GB
16:22:33 <schwicht> I was hoping we can do that for each instance
16:22:33 <alexpilotti> on HyperV this was not possible before
16:22:51 <schwicht> and that I thought we would drive with a flavor attribute
16:23:06 <alexpilotti> as w/o dynamic memory there's no balooning and the mem is allocated 1:1
16:23:12 <schwicht> QoS (and overcommit are instance attributes)
16:23:28 <schwicht> alexpilotti: understood
16:23:35 <schwicht> we do that on other hypervisors ...
16:24:14 <primeministerp> alexpilotti: anything else on the topic?
16:24:16 <alexpilotti> the reason for reusing the same option is that they have anyway to match
16:24:26 <alexpilotti> not really
16:24:37 <alexpilotti> only one question
16:24:48 <schwicht> same page sharing is not a feature supported, right?
16:24:58 <alexpilotti> should we default to true or false on enabling this feature?
16:25:08 <alexpilotti> schwicht: not on Hyper-V
16:25:11 <schwicht> k
16:25:21 <alexpilotti> schwicht: it's not efficient with SLAT
16:25:41 <alexpilotti> schwicht: unlike e.g. ESXi
16:25:45 <schwicht> yep
16:25:53 <schwicht> I was thinking at KSM ..
16:25:55 <schwicht> but that is ok
16:26:12 <schwicht> openstack can not offer more than the base hypervisor
16:26:17 <alexpilotti> can you guys please vote on the above question? :-)
16:26:28 <alexpilotti> > should we default to true or false on enabling this feature?
16:26:37 <primeministerp> I would say false
16:26:50 <alexpilotti> schwicht: ?
16:27:10 <alexpilotti> pnavarro is not joining us today?
16:27:10 <schwicht> VMware uses false ...
16:27:34 <alexpilotti> schwicht: cool, consistency is important
16:27:43 <alexpilotti> I vote for false as well
16:27:54 <alexpilotti> It's a risky feature
16:28:05 <primeministerp> agreed
16:28:08 <schwicht> what happens on overcommit isues?
16:28:10 <alexpilotti> I prefer the deoployer to know what he / she is doing
16:28:12 <schwicht> halt the VM ?
16:28:24 <alexpilotti> schwicht: yes
16:28:38 <alexpilotti> schwicht: which usually means that a lot of VMs get stopped
16:28:46 <schwicht> oh ok
16:28:51 <alexpilotti> there's also a "safety buffer" option
16:28:52 <schwicht> I thought the iggest gets supsended
16:28:56 <schwicht> biggest
16:29:08 <alexpilotti> the one requesting for the memory gets suspended
16:29:35 <alexpilotti> the safety buffer makes sure that an extra memory percentage is available (e.g. 20%)
16:30:07 <alexpilotti> so a new VM cannot be started if there's less than the expected memory
16:30:40 <alexpilotti> ok, that's it on the topic on my side
16:31:02 <primeministerp> ok
16:31:10 <primeministerp> anything on the RDP side of things?
16:31:38 <alexpilotti> not yet, we put the WMIV2 into priority
16:31:51 <alexpilotti> I'll probably send in a patch for RDP as well for H2
16:32:10 <primeministerp> ok
16:32:13 <alexpilotti> knowing that the review will take some time
16:32:58 <alexpilotti> I'd like to discuss if we should support or not the serial console
16:33:23 <primeministerp> o yes
16:33:56 <primeministerp> #topic serial console
16:34:00 <schwicht> alexpilotti: I like the serial console a lot
16:34:02 <alexpilotti> OpenStack at the moment has only the option of getting the output
16:34:13 <primeministerp> alexpilotti: you want to describe
16:34:13 <schwicht> r/o ?
16:34:20 <alexpilotti> schwicht: yep
16:34:32 <alexpilotti> we us it r/w a lot for troubleshooting
16:34:40 <schwicht> yep
16:34:42 <alexpilotti> both on Linux and on Windows
16:34:59 <alexpilotti> on Windows we created a service that provides a command prompt
16:34:59 <schwicht> logging boot issues is cool, messing with network etc
16:35:28 <schwicht> primeministerp: the OVS issue we usually debug with a serial console
16:35:32 <alexpilotti> schwicht: yep
16:35:43 <primeministerp> schwicht: ;)
16:35:54 <alexpilotti> on Windows we use it to set the user name, reboot, setting network details
16:36:07 <alexpilotti> diskpart for resizing the partitions, etc
16:36:29 <alexpilotti> enabling RDP if it was not enabled
16:36:53 <schwicht> so as long openstack does not support r/w it is merly a debugging, logging tool
16:37:03 <schwicht> and for that not as important for my use cases
16:37:06 <alexpilotti> It'd be great to have in in OpenStack with an HTML5 websocked based VT100 emulator :-)
16:37:15 <schwicht> since I can attach a console on the hyperV host
16:37:49 <alexpilotti> schwicht: yes, we a simple powershell script that attaches it and starts putty
16:38:33 <alexpilotti> schwicht: but the logging part is already cool to detect issues with cloud-init, to detect why e.g. the instance does not get SSH connections
16:38:43 <schwicht> yes it is
16:38:55 <schwicht> (we SOL enable all physical hosts for the same reason)
16:39:20 <alexpilotti> ok, I wanted to introduce the topic, I have a working proto, just need to file up a BP and polish up the patch
16:39:50 <primeministerp> nice
16:39:52 <primeministerp> well good work
16:40:42 <alexpilotti> cool, tx
16:41:11 <alexpilotti> I'm running a bit late, we organized the second Python event in Timisoara
16:41:14 <primeministerp> ok
16:41:18 <primeministerp> let's finish up
16:41:27 <primeministerp> is there anything else we're missing?
16:42:07 <alexpilotti> We have issues with OVS on RDO
16:42:25 <alexpilotti> but let's discuss about this next time once we have a solution :-)
16:42:26 <primeministerp> schwicht: this is what we were discussing earlier
16:42:33 <primeministerp> alexpilotti: can you discribe it
16:42:40 <alexpilotti> very simple:
16:42:44 <schwicht> alexpilotti: do you have a pointer with more details?
16:42:58 <alexpilotti> not yet, it came out this sunday
16:43:06 <alexpilotti> in brief:
16:43:13 <alexpilotti> multinode setup:
16:43:33 <alexpilotti> controller, network, kvm compute and hyperv compute
16:43:52 <alexpilotti> based on CentOS 6.4
16:44:14 <alexpilotti> everything works except networking
16:44:16 <schwicht> 802.1Q ?
16:44:29 <alexpilotti> anything, VLAN or flat
16:44:55 <schwicht> what physical ethernet adapter? what switch?
16:45:00 <alexpilotti> looks like either the datapaths or the flows are not working
16:45:04 <alexpilotti> OVS
16:45:17 <alexpilotti> no phisycal switches
16:45:19 <schwicht> the physical switch between the KVM and and the hyperV node
16:45:56 <alexpilotti> cloud be virtual switches on ESXi / Fusion / workstation (testing)
16:46:03 <schwicht> ah ok
16:46:08 <alexpilotti> or any L2 switch when testing a flat network config
16:46:09 <schwicht> stacked ...
16:46:12 <alexpilotti> yep
16:46:38 <schwicht> do you use portgroup 4095 on ESXi ?
16:46:40 <alexpilotti> at the beginning I was thinking about an issue due to the stacking
16:47:01 <alexpilotti> nope
16:47:20 <alexpilotti> as I was saying the issue can be reproduced even with flat networking
16:47:25 <schwicht> we probably should move that off the meeting discussion
16:47:46 <alexpilotti> I can tcpdump the traffic
16:47:54 <alexpilotti> at  the ethx level
16:48:17 <alexpilotti> it's simply not getting forwarded to the proper bridge / tap device
16:48:30 <alexpilotti> (in OpenVSwitch)
16:48:58 <alexpilotti> I'm gonna get in touch today with the RDO guys and see
16:49:15 <alexpilotti> oki guys, I really have to run unfortunately
16:49:19 <schwicht> are your br-eth and br-int bridges up?
16:49:19 <primeministerp> ok
16:49:20 <alexpilotti> Ah, one more thing
16:49:27 <alexpilotti> schwicht: yep
16:49:45 <alexpilotti> This week we start committing the Crowbar upport for Hyper-V :-)
16:50:26 <primeministerp> awesome!
16:50:59 <primeministerp> so
16:51:46 <primeministerp> anyone have anything else
16:51:49 <primeministerp> to add?
16:52:10 <primeministerp> if not, I'm going to end it.
16:52:29 <primeministerp> thanks alexpilotti for the updates
16:52:37 <primeministerp> #endmeeting