Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2020-07-02
10:58:50 lyarwood sean-k-mooney: sorry yeah us isn't branchless, devstack-gate is.
10:58:56 lyarwood uc*
11:00:15 sean-k-mooney ah ok
11:00:32 sean-k-mooney if you set it in the stable/ version of requirements
11:00:45 sean-k-mooney then it will be applied to devstack gate too right
11:00:53 gokhani hi nova folks, I wonder if we can use gpu virtualization on ubuntu kvm compute nodes? In nvidia docs, I can't see support of ubuntu kvm hypervisor ( https://docs.nvidia.com/grid/7.0/product-support-matrix/index.html ). Is there any suggestions for gpu virtualization in an OpenStack environment which works on ubuntu 18.04?
11:00:55 sean-k-mooney unless this is in unit tests
11:01:12 sean-k-mooney even then you can set it via an env var
11:01:48 sean-k-mooney gokhani: if you can install the driver then proably but they dont package for debian
11:02:42 sean-k-mooney if they are using dkms then you could proably use ailen too install the rpm
11:02:56 sean-k-mooney not sure if it would work however
11:03:15 sean-k-mooney gokhani: form a nova point of view we dont care
11:03:30 sean-k-mooney nvida driver support is the limiting factor
11:04:50 sean-k-mooney gokhani: https://www.teradici.com/web-help/ter1702003/2.8/Content/_Graphics%20Agent/Installing/Linux/installing_nvidia_linux_ubuntu.htm
11:05:50 sean-k-mooney that might just be the guest driver
11:11:55 gokhani sean-k-mooney, I have a supermicro server which has 4 nvidia tesla v100-SXM2 gpu cards and nvidia has tesla driver for ubuntu 18.04 (https://www.nvidia.com/Download/driverResults.aspx/160638/en-us). but for virtual gpu software nvidia doesn't give debian package. may be I need to also consult nvidia.
11:12:27 sean-k-mooney gokhani: you need a different driver for grid
11:12:45 sean-k-mooney teslsa cards are supported on ubunut but i dont think grid is
11:12:52 sean-k-mooney on the host at least for vgpu
11:13:11 sean-k-mooney that does not mean you cant get it to work but yes you should contact nvidia
11:13:49 sean-k-mooney you will also need to buy the approriate licenses and set up the server to delegate the licneses to your compute nodes
11:18:22 gokhani sean-k-mooney, ok ı see. initially I wanna test vgpu on openstack and then if everything goes well, we are planning to buy appropriate licenses. If I can't get it to work on ubuntu, I will try centos/rhel.
11:19:23 sean-k-mooney you cna use a low res gpu without cuda without a license i think
11:20:10 sean-k-mooney so you can test you can boot a vm but not use if for useful work beyond a slighly better vnc/spice console experinece
11:22:09 gokhani sorry what do you mean with *res ?
11:22:32 sean-k-mooney resolution e.g i think its capped at 720p
11:24:47 lyarwood sean-k-mooney: sorry back, https://github.com/openstack/devstack-gate/blob/master/devstack-vm-gate-wrap.sh#L519-L520 is the issue, I don't think we actually have requirements pulled down for the correct branch at this point
11:25:11 sean-k-mooney lyarwood: zuul should have cloned it
11:25:33 sean-k-mooney it does that before devstack-gate runs as far as i know
11:27:47 sean-k-mooney so i think you can jsut pass -c /opt/git/openstack/requirements/upper-constriats.txt
11:28:07 lyarwood hmmm I think this might actually be fixed, I can't reproduce locally
11:28:23 gokhani sean-k-mooney, ok if I see correctly, for testing ı can test gpu virtualization with no license.
11:30:20 sean-k-mooney gokhani: i belvie so just with reduced capablities
11:30:27 sean-k-mooney bug you can confirm the hardware works
11:35:01 gokhani sean-k-mooney, I am confused. from nova point for using nvidia vgpu capabilities, do I need also nvidia vgpu software on my server or it is built in function in nova ? ı mean tesla or grid driver is enough for nova except licenses ?
11:41:50 sean-k-mooney gokhani: you need the vgpu software on your server
11:42:05 sean-k-mooney nova just has the ablity to tell libvirt to configure it
11:42:13 sean-k-mooney well vms that use vgpus
11:43:58 sean-k-mooney gokhani: the lices are ativly issued by a server like windows linces so you need to run an instance of that licensing serverr somewhere and you need to configre the hosts to phone home ot that server everytime a vms is booted with a vgpu to claim a lisence
11:44:18 sean-k-mooney gokhani: this should be all coverd in the nvida docs
11:44:31 sean-k-mooney its not specific to nova
11:57:05 gibi stephenfin: left a bunch of notes in https://review.opendev.org/#/c/631363 . I can be convinced to do the functional test coverage in a follow up
12:06:09 openstackgerrit Huaqiang Wang proposed openstack/nova master: objects: Introduce the 'CPUAllocationPolicy.MIXED' enum https://review.opendev.org/716267
12:06:10 openstackgerrit Huaqiang Wang proposed openstack/nova master: libvirt: apply mixed instance CPU policy https://review.opendev.org/728481
12:06:10 openstackgerrit Huaqiang Wang proposed openstack/nova master: Validate CPU pinning configuration for mixed instance https://review.opendev.org/716268
12:06:10 openstackgerrit Huaqiang Wang proposed openstack/nova master: Calculate the CPU usage for mixed instance https://review.opendev.org/716269
12:06:11 openstackgerrit Huaqiang Wang proposed openstack/nova master: scheduler: mixed instance CPU resource translating https://review.opendev.org/728482
12:06:12 openstackgerrit Huaqiang Wang proposed openstack/nova master: flavor: Enable 'hw:cpu_dedicated_mask' for creating a mixed instance https://review.opendev.org/714706
12:13:10 jsuchome gibi: Hi, I've created https://review.opendev.org/#/c/738738/ for you
12:13:45 gibi jsuchome: thans!
12:13:47 gibi thanks
12:17:11 gokhani sean-k-mooney, thanks for your useful information. After consulting to nvidia, I will decide.
12:54:43 openstackgerrit Stephen Finucane proposed openstack/nova master: virt: Add 'context', drop 'network_info' parameters for 'unrescue' https://review.opendev.org/730382
12:54:43 openstackgerrit Stephen Finucane proposed openstack/nova master: Add emulated TPM support to Nova https://review.opendev.org/631363
12:54:43 openstackgerrit Stephen Finucane proposed openstack/nova master: Add support for resize and cold migration of emulated TPM files https://review.opendev.org/639934
12:56:58 stephenfin dansmith: Can you look at https://review.opendev.org/#/c/683043/ again today?
13:16:44 openstackgerrit Lee Yarwood proposed openstack/nova stable/rocky: Make greande jobs n-v for EM and oldest stable https://review.opendev.org/739020
13:35:39 openstackgerrit Lee Yarwood proposed openstack/nova stable/rocky: fix scsi disk unit number of the attaching volume when cdrom bus is scsi https://review.opendev.org/730312
13:41:39 isakgicu Hi guys, is it possible to convert a file from /var/lib/nova/instances/_base to an image for glance ?
13:44:59 isakgicu for example, accidentally an image from glance was deleted after instances where created, but I know which file from /var/lib/nova/instances/_base/ is an actual copy of that image, is it possible to make an image from /var/lib/nova/instances/_base/ files ?
13:46:15 lyarwood isakgicu: yeah assuming it's just a RAW file you should be able to copy it and upload it as a fresh image into glance
13:47:20 isakgicu thank you for quick answer !
13:47:24 isakgicu gonna try
15:04:51 gmann lyarwood: np! now we have devstack is up for stable/rocky - https://review.opendev.org/#/c/735615/9
15:05:13 gmann lyarwood: we should be good to do for nova now, grenade jobs as n-v
15:12:30 lyarwood gmann: ack yeah thanks, we are seeing an issue with g-api on the subnodes being unable to run as uwsgi is installed under a different path on 16.04
15:12:46 lyarwood just working through it now
15:16:26 gmann lyarwood: yeah, we actually do not run uwsgi on subnode as no api service is supposed to run there - https://review.opendev.org/#/c/738780/
15:17:16 gmann i remember we talked to disable g-api on subnode long back too but let's see in 738780
15:18:11 lyarwood yeah it eventually got nack'd as the glance team were using it during their grenade tests
15:18:19 lyarwood https://review.opendev.org/#/q/I02bff11495f29822f1b5d0123d26df7c9bb76b30
15:18:55 elod gmann lyarwood : actually I'm surprised that g-api is installed on subnode, because this patch removed it (at least I thought that): https://review.opendev.org/#/c/546765/
15:19:33 sean-k-mooney1 elod: i think glance wanted to test multi store backends or soemthing like that
15:20:27 sean-k-mooney1 elod: that is for zuulv3 jobs
15:20:36 sean-k-mooney1 the legacy jobs still install it
15:21:02 lyarwood https://review.opendev.org/#/q/312517d5101206b33d3c395d27ec93d385b7fd24 ah ha!
15:21:08 lyarwood of course Ian has already fixed this
15:22:30 lyarwood gmann: FWIW without this in stable/rocky all of the services are still using the pip installed uwsgi version
15:24:58 gmann elod: sean-k-mooney1 lyarwood if it needed in mutlistore then we also need uwsgi setup in subnode too but that can be from Victoria and all stable can remove g-api may be
15:25:28 sean-k-mooney1 gmann: its not strictly required fro multistore you can do that single node
15:25:57 sean-k-mooney1 gmann: apparently it was actully for grenade
15:26:05 sean-k-mooney1 not multistore
15:26:16 gmann grenade?
15:27:16 sean-k-mooney gmann: apparently glance were doing some testing in grenade that wanted to glance-api on the subnode
15:27:31 gmann ohk
15:27:42 sean-k-mooney but to be clear if the use the v3 grenade job they can contol that themselves
15:27:45 elod i found it on multinode only, but if glance team needs g-api on subnode, then maybe the easiest fix is just to use '$(which uwsgi)' in glance (and others)
15:28:08 gmann yeah and we have grenade zuulv3 till stable/train
15:28:53 lyarwood elod: yeah posting that now
15:28:56 sean-k-mooney well id devstack need to find the correct uwsgi to use we still might need to do the $(which...)
15:28:58 lyarwood https://review.opendev.org/739072
15:29:17 sean-k-mooney but they can at least contol if its enabled on the subnode trivially so we can disable it by defualt if we want too
15:30:42 elod lyarwood: thanks for the patch, looks good :]
15:30:50 openstackgerrit Lee Yarwood proposed openstack/nova stable/rocky: DNM - Testing g-api on subnode fix https://review.opendev.org/739074
15:30:52 lyarwood ^ testing there
15:35:13 gmann lyarwood: thanks
15:40:45 openstackgerrit Lucio Seki proposed openstack/nova master: Implement extend_volume for libvirt NFS volume driver https://review.opendev.org/739077
15:56:58 gibi nova meeting starts in 3 minutes on #openstack-meeting-3
16:58:51 stephenfin We've got driver capabilities and those can optionally be reported as traits. Why would someone *not* want to report the trait?
17:00:39 elod melwitt: about subunit.parser: i've seen that only on stein (so far). do you think subunit 1.4.0 would make that dissappear?

Earlier   Later