| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-09-25 | |||
| 16:33:19 | mriedem | nope https://blueprints.launchpad.net/nova/+spec/hide-hypervisor-id-flavor-extra-spec was rocky | |
| 16:33:39 | gregwork | and there is no way to pass libvirt domain customizations | |
| 16:33:52 | gregwork | even kludgy ones :) | |
| 16:33:57 | gregwork | or even qemu execution lines | |
| 16:34:10 | mriedem | not through the compute api no | |
| 16:34:29 | mriedem | that's not really...cloud | |
| 16:35:29 | artom | Don't we have image props or something to make Windows happy? | |
| 16:35:54 | mriedem | artom: read scrollback | |
| 16:36:03 | gregwork | ever since they started planning to go to core based licensing instead of socket they have implemented some draconian things to prevent the potential for nested virtualization | |
| 16:36:11 | openstackgerrit | Merged openstack/nova master: docs: Scrub available quotas https://review.opendev.org/670125 | |
| 16:36:31 | gregwork | especially since you can do performant nested virt with kvm on intel xeon e3-v4 processors using vmcs shadowing and device passthrough of network/storage | |
| 16:38:05 | gregwork | in our lab using those chips and passing through a virtual function off our nic to the L1 guest (hypervisor) we were only seeing a 5-10% difference in perf | |
| 16:38:10 | gregwork | it was really interesting | |
| 16:38:27 | gregwork | this is how hitachi does LPAR's on their x86_64 platform | |
| 16:39:04 | gregwork | hardware logical partitions which can run performant guests using intel xeon chips | |
| 16:39:17 | gregwork | anyhow it sucks i cant fiddle with this on queens | |
| 16:39:19 | gregwork | :/ | |
| 16:40:51 | mriedem | i can't believe sean-k-mooney isn't around to chat about this | |
| 16:41:07 | mriedem | gregwork: well you could if you $$$ your vendor to backport a feature | |
| 16:43:32 | artom | mriedem, next time dinner first | |
| 16:47:00 | mriedem | dinner? you mean lunch? | |
| 16:47:05 | mriedem | what are you 80? | |
| 16:47:45 | artom | In my mind :( | |
| 16:53:35 | mriedem | stephenfin: i'm not sure we need this https://review.opendev.org/#/c/684781/ - it appears to already be fixed in master, though i guess we might want it just for backports | |
| 16:53:37 | mriedem | i left some notes inline | |
| 17:02:59 | gmann | mriedem: ohk. did you push the fix to tag tempest on pike gate ? | |
| 17:03:15 | mriedem | gmann: yup | |
| 17:03:41 | gmann | thanks. got it | |
| 17:09:17 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Ignore sqla-migrate inspect.getargspec deprecation warnings on py36 https://review.opendev.org/684781 | |
| 17:09:18 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Ignore warning from sqlalchemy-migrate https://review.opendev.org/684772 | |
| 17:09:18 | openstackgerrit | Matt Riedemann proposed openstack/nova master: tox: Use common 'command' definition for unit tests https://review.opendev.org/684774 | |
| 17:09:19 | openstackgerrit | Matt Riedemann proposed openstack/nova master: tox: Stop overriding the 'install_command' https://review.opendev.org/684775 | |
| 17:09:19 | openstackgerrit | Matt Riedemann proposed openstack/nova master: tests: Nuke OS_STDOUT_CAPTURE, OS_STDERR_CAPTURE https://review.opendev.org/684773 | |
| 17:26:08 | gregwork | mriedem: what about sean-k-mooney ? | |
| 17:33:08 | sean-k-mooney | gregwork: mriedem sriov testing? | |
| 17:34:02 | gregwork | sean-k-mooney: hi, no actually trying to fool a windows instance on openstack into not thinking it is a guest. either modifying the domain in libvirt or possibly passing args to qemu-kvm | |
| 17:34:08 | gregwork | i have a working libvirt domain that does what i need | |
| 17:34:17 | gregwork | trying to figure out how to do this with nova/kvm | |
| 17:34:27 | sean-k-mooney | i tested doing nested sriov like 3 weeks ago and it does work if you enable a vIOMMU in the l1 guest and use q35 chipset but you need to play with the pci layout to fix the iommu groups | |
| 17:34:30 | sean-k-mooney | oh | |
| 17:34:32 | sean-k-mooney | ok | |
| 17:34:44 | sean-k-mooney | i think we fixed that | |
| 17:34:55 | sean-k-mooney | so you can enable hypervior hideing | |
| 17:34:56 | gregwork | yeah the nested virt works great if you have a chip that can do vmcs shadowing, and do passthrough devices for the L1 guest | |
| 17:35:10 | gregwork | if you dont pass through network/storage it will suck real bad for the L2 guests | |
| 17:35:23 | gregwork | at least then you remove 1 layer of nested virt (storage/network) | |
| 17:35:27 | sean-k-mooney | and we not hardcode a fake vendor id in the hyperv secotion so it thinks its running on a phyical host | |
| 17:35:45 | gregwork | part of it is disabling the hypervisor cpu flag in the guest | |
| 17:35:49 | gregwork | the other is the vendor string thing | |
| 17:36:07 | sean-k-mooney | yes if you set teh image property i think it does both | |
| 17:36:42 | sean-k-mooney | gregwork: have you tried setting img_hide_hypervisor_id | |
| 17:36:47 | sean-k-mooney | in the image to yes | |
| 17:37:05 | gregwork | do you know if a certain hat wearing vendor backported that to their queens release | |
| 17:37:26 | sean-k-mooney | i could check | |
| 17:37:35 | sean-k-mooney | but not off the top of my head | |
| 17:37:51 | sean-k-mooney | the image property woudl not be backportable due to object changes | |
| 17:38:03 | sean-k-mooney | but i belive you can also contol the behavior via the flavor | |
| 17:38:14 | sean-k-mooney | the flavor extra spec would be backportable | |
| 17:38:21 | sean-k-mooney | downstream | |
| 17:40:34 | sean-k-mooney | it looks like the kvm supprot is ther for linux guests | |
| 17:40:44 | sean-k-mooney | which means the image property should be there | |
| 17:41:13 | sean-k-mooney | but the support for windows guest has not been backported | |
| 17:41:38 | sean-k-mooney | http://paste.openstack.org/show/779253/ | |
| 17:41:54 | sean-k-mooney | gregwork: ^ that is the downstream code for generating the elements | |
| 17:49:49 | sean-k-mooney | gregwork: assuming you or one of your customers are a customer of said company have you filed a Bugzilla ro case to request the feature backport. | |
| 17:50:13 | gregwork | im initially trying to figure out if this was a thing accomplished another way | |
| 17:50:30 | mriedem | sean-k-mooney: if you're around i think we're waiting on an ack from you for https://review.opendev.org/#/c/683437/ | |
| 17:50:32 | sean-k-mooney | gregwork: im not committing to it being granted but in principal i belive this would be a minimal risk | |
| 17:51:24 | sean-k-mooney | mriedem: strictly speaking its my birthday and im off but yes im waiting for windows updates to install so if have a few minutes | |
| 17:51:40 | gregwork | sean-k-mooney: happy bday sean :) | |
| 17:51:59 | sean-k-mooney | thanks you :) | |
| 17:52:18 | gregwork | i think i spoke to you here a few years about getting dpdk working with mellanox cards | |
| 17:52:21 | gregwork | your nic is familiar | |
| 17:52:27 | gregwork | *nick | |
| 17:52:43 | sean-k-mooney | yes we have spoken before | |
| 17:52:57 | sean-k-mooney | an ya i used to work with dpdk alot | |
| 17:53:08 | sean-k-mooney | im hopeing to get a dpdk gate job running soon | |
| 17:53:34 | gregwork | are you still with intel | |
| 17:53:57 | sean-k-mooney | no i moved to redhat about 14 months ago | |
| 17:54:14 | sean-k-mooney | hence why i was able to check the OSP downstream code | |
| 17:54:24 | gregwork | oh i just figured you were more in the know | |
| 17:54:46 | gregwork | that link you posted, was that for osp 15 ? | |
| 17:55:01 | sean-k-mooney | no it was osp 13 which is queens | |
| 17:55:40 | sean-k-mooney | so queens has support for hideing the hypervior signature for linux guest | |
| 17:55:53 | gregwork | not not the required bits for a windows guest | |
| 17:55:55 | sean-k-mooney | all its missing is the hyperv vendor signiture bit | |
| 17:55:58 | gregwork | *but | |
| 17:56:27 | sean-k-mooney | so a backport of the fix would just be the xml generation change the rest of the code is alreay in place | |
| 17:57:29 | sean-k-mooney | gregwork mriedem can correct me if im wrong but i think we fixed windwos guest in stien so 15 should have the support. | |
| 17:57:53 | gregwork | we are doing a prod rollout so the business decision was to stick with LTS releases | |
| 17:58:04 | gregwork | so 14 and 15 dont exist as options for us sadly | |
| 17:58:18 | mriedem | sean-k-mooney: train https://review.opendev.org/#/c/579897/ | |
| 17:58:25 | sean-k-mooney | unfortunetly you are not the only customer with that view | |
| 17:59:02 | sean-k-mooney | oh so it was. it felt recent but i did not think it was train | |
| 17:59:42 | sean-k-mooney | mriedem: we ended up treating this as a bug. is it a candiate for upstream backport? generally i would assume not but it does not hurt to ask | |
| 17:59:54 | gregwork | im debating trying to patch this myself and splice it in /var/lib/config-data/puppet-generated/nova_libvirt and see if i can get the hyperv processor bits to die | |
| 18:00:12 | gregwork | then maybe circle back with an rfe | |
| 18:04:33 | mriedem | sean-k-mooney: exhibit A https://review.opendev.org/#/c/663616/ | |
| 18:05:09 | mriedem | either way the hide_hypervisor_id stuff upstream was in rocky so gregwork doesn't have it anyway | |
| 18:05:32 | mriedem | unless you guys in rhosp are going to backport a feature | |