13:01:18 <claudiub|2> #startmeeting hyper-v 13:01:18 <openstack> Meeting started Wed Mar 23 13:01:18 2016 UTC and is due to finish in 60 minutes. The chair is claudiub|2. Information about MeetBot at http://wiki.debian.org/MeetBot. 13:01:19 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 13:01:21 <openstack> The meeting name has been set to 'hyper_v' 13:01:27 <sagar_nikam> Hi 13:01:38 <c64cosmin> hello all 13:01:59 <claudiub|2> ok, I'll try to keep it short. 13:02:08 <claudiub|2> #topic mitaka releases 13:02:46 <claudiub|2> so, mitaka is going to be released in early april, meaning that any and all dependencies will have to be released until then too. 13:02:59 <claudiub|2> that includes networking_hyperv, compute_hyperv, and os-win 13:03:21 <claudiub|2> I'm going to cut the stable/mitaka branches on friday / saturday 13:03:32 <claudiub|2> #action claudiub to cut the stable/mitaka branches 13:04:15 <claudiub|2> which means, that if there are any more issues or bugs, this week is going to be a good chance to fix them. :) 13:04:58 <claudiub|2> if there's anything, let me know 13:05:08 <claudiub|2> ok, moving on. 13:05:23 <claudiub|2> #topic hyper-v neutron vif plug event listener on spawn 13:06:19 <claudiub|2> so, this was suggested some time ago by the HP folks. The main idea was that the Hyper-V Driver was not waiting for the vNICs to be bound, which was leading to missed DHCP requests. 13:06:52 <claudiub|2> we've addressed that issue in compute_hyperv, and it will have to be included in nova as well 13:06:56 <claudiub|2> #link https://review.openstack.org/#/c/292615/ 13:07:27 <claudiub|2> #action propose the hyper-v neutron vif plug event listerner commit on nova 13:08:25 <claudiub|2> from experiments, it seemed to be behaving properly, with little to no decrease in performance, but a rally would be nice to confirm this. 13:08:58 <claudiub|2> ok, moving on. 13:09:10 <claudiub|2> #topic Rally performance tests 13:09:27 <claudiub|2> abalutoiu: hello. do you have any good news for us? :) 13:09:33 <abalutoiu> hello 13:10:50 <abalutoiu> so last week we had some results for the spawn / destroy scenario, here are the results for the spawn / destroy scenario including ssh guest access http://46.101.88.127:8888/PyMI_KVM_ssh_Mitaka.html 13:11:25 <abalutoiu> Liberty ones, for comparison: http://46.101.88.127:8888/PyMI_KVM_ssh_Liberty.html 13:12:40 <claudiub|2> #link Mitaka Hyper-V vs KVM http://46.101.88.127:8888/PyMI_KVM_ssh_Mitaka.html 13:12:57 <sagar_nikam> avg time is more in HyperV 13:13:03 <claudiub|2> #link Liberty Hyper-V vs KVM http://46.101.88.127:8888/PyMI_KVM_ssh_Liberty.html 13:13:19 <sagar_nikam> kvm - nova.boot_server 19.502 34.971 66.859 74.398 78.741 42.074 100.0% 100 13:13:33 <sagar_nikam> hyperv - nova.boot_server 32.596 120.705 147.194 155.103 177.713 109.565 98.0% 100 13:14:01 <abalutoiu> sagar_nikam: those are the results for Liberty, the old ones, have a look over the Mitaka results 13:14:11 <sagar_nikam> has anything changed ? last week we saw almost same result for hyperv and kvm 13:14:32 <sagar_nikam> ok 13:14:53 <abalutoiu> the results from last week didn't include ssh guest access 13:15:01 <sagar_nikam> ok, this looks good 13:15:07 <sagar_nikam> kvm and hyperv almost same 13:15:25 <sagar_nikam> the avg time includes ssh access ? 13:15:31 <claudiub|2> yep. 13:15:35 <sagar_nikam> ok 13:15:47 <sagar_nikam> and how many NICs ? 13:15:51 <sagar_nikam> per VM 13:16:05 <abalutoiu> only one NIC 13:16:06 <sagar_nikam> ? 13:16:11 <sagar_nikam> and are we trying attaching volumes 13:17:06 <abalutoiu> it's a simple scenario, spin up a VM, add floating ip to it, wait until it becomes active via ssh, then destroy the VM 13:17:33 <sagar_nikam> ok 13:17:49 <sagar_nikam> these results are good 13:17:53 <sagar_nikam> nice work 13:18:18 <sonu> can we continue our tests without destroying the VM? 13:18:32 <abalutoiu> we also have some results for Hadoop workloads 13:18:42 <claudiub|2> sonu: any reason why? 13:18:46 <abalutoiu> #link Mitaka Hyper-V vs KVM Hadoop cluster tests http://46.101.88.127:8888/Cluster_Results_Mitaka.html 13:19:19 <sonu> The only reason being, the larger the number of ACLs in MI, the more load on the neutron agent. 13:19:43 <sonu> and more time it takes. And we can measure the times it varies with the number of ACLs. 13:20:11 <abalutoiu> Flavor for VMs: 2 vCPUs, 6GB RAM, 25GB disk, number of parallel clusters for multi cluster test: 3, input data set for tera jobs: 5.000.000, clusters consists of 4 nodes (1 master + 3 slaves) 13:20:53 <claudiub|2> sonu: That also depends on the rally scenario. For example, if the scenario says 100 VMs, with 100 VMs per round, for 1 round, it will spawn all 100 VMs. Then destroy them. 13:21:15 <sonu> sure. then we are good. 13:21:37 <claudiub|2> anyways, looking at abalutoiu's hadoop results, it seems hyper-v is more favorable in this scenario 13:21:50 <sonu> and we use default security group rules for our tests. 13:22:14 <claudiub|2> sonu: default + ssh. :) 13:22:25 <sonu> Thank you. We will consume this coming next week and run some scale tests. 13:22:43 <claudiub|2> sure, sounds good. :) 13:23:04 <claudiub|2> anything else on this topic? 13:23:20 <claudiub|2> ok, moving on. 13:23:20 <sonu> Any chance we can get these improvements backported to L 13:23:29 <sonu> we are on L release 13:23:55 <sagar_nikam> it would be nice if these are backported to L 13:24:01 <sonu> we are trying to backport on our local branch - native threading and PYMI and ERPC 13:24:11 <claudiub|2> sonu: any improvement related to spawning / creating / destroy the vm etc. has been included in compute_hyperv on stable/liberty 13:24:24 <sonu> thank you 13:24:59 <claudiub|2> for the networking-hyperv side, I'm afraid not. networking-hyperv is an official project under neutron's governance. The improvements we've done in Mitaka were marked as blueprints, meaning that they can't be backported. 13:25:52 <claudiub|2> ok, moving on. 13:26:04 <claudiub|2> #topic OVS agent on Hyper-V 13:26:28 <claudiub|2> atuvenie: hello. I suppose the agent works fine for Mitaka, right? 13:26:52 <atuvenie> yeah, we just merged the last patch that was causing some issues, all is fine now 13:27:28 <claudiub|2> cool. 13:27:47 <claudiub|2> seems that we're almost done with the work for Mitaka. :) 13:27:52 <sonu> atuvenie : Do we know the feature parity between OVS for Linux and OVS Windows. Are all features supported as is in OVS. 13:28:45 <atuvenie> sonu: I'm not quite sure what you mean. Are you talking about the OVS Agent or OVS itself? 13:28:51 <sonu> Who participates in OVS forums from within us, who can bring us this info? 13:28:55 <sonu> OVS itself. 13:29:40 <atuvenie> sonu: I do not work on OVS directly so I'm not up to speed on the topic 13:30:11 <claudiub|2> AFAIK, no, there shouldn't be any disparity. all network types work the same on Linux and Windows. 13:30:44 <sonu> Thanks. May be we start attending OVS IRC chats to get more info. 13:31:06 <claudiub|2> but as far as OVS agent is concerned, it works the same on Windows and Linux. 13:31:24 <sonu> And is there a CI for Microsoft HyperV when a change is done in neutron-openvswitch-agent? 13:31:47 <sonu> because it is shared b/w Linux and Windows now :) 13:32:03 <claudiub|2> not yet, but we're working on it. :) 13:32:17 <sonu> Great. Thank you. 13:32:40 <claudiub|2> ok, moving on 13:32:56 <sagar_nikam> claudiub|2: i have a topic 13:33:07 <claudiub|2> #topic OpenStack Summit 13:33:07 <sagar_nikam> on certs 13:33:49 <claudiub|2> sagar_nikam: sure, I'll get to that soon. :) 13:34:21 <claudiub|2> anyways, we've requested a worksession at the next OpenStack summit. 13:35:14 <claudiub|2> in which we can discuss any further improvement and development of os-win and other Hyper-V / Windows related workloads. 13:35:15 <sonu> Will OVN be part of design discussion? 13:35:21 <claudiub|2> including new features and so on. 13:36:04 <claudiub|2> sonu: sure, we are looking into it as well. 13:36:18 <claudiub|2> so, question: who is going to attend the summit? 13:37:11 <sonu> I will have one representative from networking team. 13:38:05 <claudiub|2> ok, cool. 13:38:47 <claudiub|2> ok, moving on 13:38:51 <claudiub|2> # topic certificates 13:38:57 <claudiub|2> #topic certificates 13:39:16 <sagar_nikam> based on the meeting discussion last week 13:39:29 <sagar_nikam> we tried using certs on the hyperv hosts 13:39:39 <sagar_nikam> we copied the .crt file from controller 13:39:48 <sagar_nikam> to the hyperv host 13:40:04 <sagar_nikam> and had the correct https and cafile entry in nova.conf 13:40:28 <sagar_nikam> we are using python 2.7.10 64 bit on the hyperv host 13:40:32 <sagar_nikam> we hit a issue 13:41:00 <sagar_nikam> which is exactly same as described here 13:41:10 <sagar_nikam> http://stackoverflow.com/questions/33140382/troubleshooting-ssl-certificate-verify-failed-error 13:41:39 <sagar_nikam> the issue was urllib2.URLError: <urlopen error [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed (_ssl.c:590) 13:41:55 <sagar_nikam> this looks like a issue in python 2.7.10 13:42:15 <claudiub|2> interesting 13:42:22 <sagar_nikam> since it uses openssl 1.0.2a 13:42:35 <sagar_nikam> and we need 1.0.2b or greater 13:42:44 <sagar_nikam> which is available in python 2.7.11 13:42:49 <sagar_nikam> now my question 13:42:59 <sagar_nikam> how was your tests done using certs 13:43:07 <sagar_nikam> did you not hit this issue 13:43:09 <sagar_nikam> ? 13:43:22 <sagar_nikam> what version of python do you use ? 13:43:33 <sagar_nikam> on hyperv host 13:45:06 <claudiub|2> the hyper-v compute installers come with python 2.7.9 13:45:20 <claudiub|2> so, we're using that. so, the person that answered your previous email is not here at the moment, but he'll be back later and he'll be able to send you an answer. 13:45:46 <sagar_nikam> ok 13:46:01 <sagar_nikam> thanks 13:46:14 <sagar_nikam> not sure if python 2.7.9 can solve the issue 13:46:22 <sagar_nikam> but i will wait for the mail 13:46:37 <claudiub|2> sagar_nikam: although, i am a bit curious, can you check what python version you use on your Openstack controller? 13:46:48 <sagar_nikam> python 2.7.9 13:47:00 <sagar_nikam> we had some issues with python 2.7.9 on windows 13:47:08 <sagar_nikam> and hence used 27.10 13:47:26 <sagar_nikam> nova code was not getting compiled on 2.7.9 13:47:33 <claudiub|2> what kind of issues did you have with python 2.7.9 on windows? 13:47:33 <sagar_nikam> some dependent packages 13:47:44 <sagar_nikam> when we run setup.py 13:47:49 <sagar_nikam> of nova 13:47:59 <sagar_nikam> packages were not getting complied 13:48:05 <claudiub|2> sure, nova and a lot of other projects have linux specific dependencies, which cannot be installed on windows. 13:48:21 <claudiub|2> which is why we recommend the installer, as we package those dependencies as well 13:48:59 <sagar_nikam> the compiled packages --- are they available for others to use 13:49:11 <sagar_nikam> whatever you got it compiled ? 13:49:32 <claudiub|2> plus, there are other dependencies like numpy that has to be compiled. windows doesn't typically come with a compiler and you wouldn't want one on nano or bare hyper-v hosts. 13:50:03 <sagar_nikam> we use mingw for compiling 13:50:30 <sagar_nikam> and that solved lot of complilation issues 13:50:36 <sagar_nikam> in 2.7.10 13:51:31 <claudiub|2> yeah, ofc, but there is no need to compile anything when using the installer, everything is already packaged in the Python folder it installs. 13:51:31 <sagar_nikam> one question -- all the packages which you have complied 13:51:45 <sagar_nikam> is that available for others to use 13:52:13 <sagar_nikam> we are not using that installer 13:52:21 <sagar_nikam> hence the question 13:52:34 <claudiub|2> i'm not against it. :) 13:53:26 <sagar_nikam> if the compiled files are available, we can use it while we run setup.py 13:53:47 <sagar_nikam> and that would solve lot of the compilation issues we faced in 2.7.9 13:54:06 <claudiub|2> but we still recomend using the installer. it can be installed in unnattended mode, which is perfect for automations. 13:54:13 <slogan> #join #openstack-monasca 13:54:56 <claudiub|2> #topic open discussion 13:55:34 <claudiub|2> so, the Newton branch is open, which means we will start merging patches on master again 13:55:52 <claudiub|2> one pending patch that we want is the os-win in os-brick patch 13:56:48 <claudiub|2> that is based on henma's patch 13:57:05 <sagar_nikam> upstream branch open ? 13:57:09 <claudiub|2> sagar_nikam: hist patch is still WIP. Any news from him? 13:57:16 <sagar_nikam> for newton 13:57:23 <claudiub|2> sagar_nikam: yeah, for most projects. 13:57:24 <sagar_nikam> i can check 13:57:32 <sagar_nikam> can you let me know which patch set 13:57:58 <sagar_nikam> can we start submitting FC and cluster BPs and patches in N 13:58:00 <claudiub|2> #link https://review.openstack.org/#/c/275943/5 13:58:11 <sagar_nikam> we can get review time 13:58:13 <sagar_nikam> now 13:58:17 <claudiub|2> the blueprints have been reapproved for N. 13:58:32 <sagar_nikam> can we re-submit code as well 13:58:40 <sagar_nikam> both for cluster driver and FC 13:58:57 <sagar_nikam> it would be nice to get it merged upstream 13:58:59 <sagar_nikam> in N 13:59:04 <claudiub|2> anyways, we'll also have to recreate the famous 3 hyper-v patches queue that needs to be reviewed and ready to merge on the etherpad. 13:59:32 <claudiub|2> ofc. having os-win in os-brick will greatly help on this subject. 13:59:53 <claudiub|2> #action claudiub to create the nova patches queue on etherpad. 14:00:06 <claudiub|2> Seems that our time is over 14:00:16 <claudiub|2> thanks all for attending! 14:00:23 <claudiub|2> #endmeeting