Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2024-05-07
16:47:46 gibi have a nice rest of the day
16:47:50 gibi #endmeeting
16:47:50 opendevmeet Meeting ended Tue May 7 16:47:50 2024 UTC. Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4)
16:47:50 opendevmeet Minutes: https://meetings.opendev.org/meetings/nova/2024/nova.2024-05-07-16.00.html
16:47:50 opendevmeet Minutes (text): https://meetings.opendev.org/meetings/nova/2024/nova.2024-05-07-16.00.txt
16:47:50 opendevmeet Log: https://meetings.opendev.org/meetings/nova/2024/nova.2024-05-07-16.00.log.html
16:48:01 elodilles thanks o/
16:48:03 fwiesel Thanks
16:48:57 opendevreview sean mooney proposed openstack/nova master: tweak emulation job to avoid OOM errors https://review.opendev.org/c/openstack/nova/+/918464
16:49:08 sean-k-mooney gibi: i think ^ will fix the emulation job
16:49:09 gibi ganso: as far as I see in ussuri the functional env called functional
16:49:18 gibi so tox -e functional should work
16:49:51 tkajinam I wonder if we have any core time concept for review spec days. I'll try to be available around the irc meeting time anyway.
16:50:27 gibi tkajinam: I think there are no core time for it
16:50:45 opendevreview sean mooney proposed openstack/nova master: [DNM] test nova-emulation https://review.opendev.org/c/openstack/nova/+/918465
16:52:27 tkajinam gibi, ok, thanks. that's what I guessed.
16:52:38 sean-k-mooney dansmith: by the way as far as im aware for the libvirt driver virtio-scsi should be the default scsi modle when you enable it with hw_disk_bus=scsi
16:52:57 dansmith sean-k-mooney: it is most definitely not
16:53:08 dansmith defaults to lsilogic
16:53:10 sean-k-mooney looking at the code paths it looks like there might be a second way to get a contoler
16:53:36 dansmith took me a while while debugging to realize it was working, but cirros didn't have drivers for lsi
16:53:51 sean-k-mooney so i dont knwo fi we regressed it then
16:54:00 sean-k-mooney beucase at one point it was virtio-scisi
16:54:48 sean-k-mooney we doucmented it as vrito-scsi here https://github.com/openstack/glance/blob/master/etc/metadefs/compute-libvirt-image.json#L65-L69
16:55:17 sean-k-mooney in libverty
16:55:59 dansmith just sayin' it doesn't work for me locally unless I set that
16:56:12 sean-k-mooney what do you have hw_disk_bus set too
16:57:27 sean-k-mooney it looks like we have 2 ways to get a scsi contoler either we get fone form seting hw_scsi_modle here https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L5794
16:57:45 sean-k-mooney or we get one by setign hw_disk_bus or hw_cdrom_bus to scsi
16:58:02 sean-k-mooney i was aware of the latter two options but i didnt knwo about https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L5923
16:58:20 dansmith I don't have disk_bus set on the image or flavor, but it _is_ set in the BDM of courtse
16:59:33 opendevreview Dan Smith proposed openstack/nova master: Enable virtio-scsi in nova-next https://review.opendev.org/c/openstack/nova/+/918458
16:59:35 dansmith forgot the zuul bit ^
16:59:55 sean-k-mooney oh including the playbook ya
17:00:09 dansmith yeah, duh
17:00:29 sean-k-mooney so realted quetion
17:00:46 sean-k-mooney if we are trying to test this with BFV or even a data volume
17:00:52 sean-k-mooney to we really need to set this on the image
17:01:08 sean-k-mooney or can it be set on the voulume createf from the image
17:01:12 dansmith it only works on the image, FWIW.. doesn't work on the flavor
17:01:24 sean-k-mooney well it should not work on the flaovr
17:01:26 dansmith also, BFV in this case doesn't work.. there's some other bug that we hit when it tries
17:01:43 sean-k-mooney ok
17:01:52 dansmith ack, well, it doesn't, but I expected it to because it's a hardware sort of thing
17:02:06 sean-k-mooney then that answer my quetion regrading data voluem with a data volume we take the info form the root disk
17:03:01 sean-k-mooney basically i was wondering if in teh tempest test we coudl create a volmue form the glance image
17:03:07 gmann gibi: sean-k-mooney: RE on SG BP. I cannot join the Nova meeting as it conflicts with the time I drop my son to school. I checked the BP/code change https://blueprints.launchpad.net/nova/+spec/shared-security-groups
17:03:09 sean-k-mooney then update the image properites on teh volume to add it
17:03:20 sean-k-mooney and then boot form the volume
17:03:38 sean-k-mooney to avoid needing to configure this in the job
17:03:51 sean-k-mooney gmann: no worries
17:04:06 gmann gibi: sean-k-mooney: it seems ok to approve specless, there is no interface or API change here, SecurityGroupNotFound error can occur in any case even no SG in tenant or so and adding to get shared SG does not change it even improve to avoid notFound cases
17:04:44 opendevreview Vasyl Saienko proposed openstack/nova master: [ironic] Fix rebooting instance https://review.opendev.org/c/openstack/nova/+/918195
17:05:15 sean-k-mooney gmann: ya i had a simialr feeling. the secuirty gorup could also have been delete between the vm create request being accpeted by the api and the swapn on the compute
17:05:18 gmann and as it check the neutron extension enable or not, we have the discovery path also for inteop point of view
17:05:27 gmann sean-k-mooney: yeah
17:05:27 sean-k-mooney gmann: so its always possibel that this would fail the way it is now
17:09:31 sean-k-mooney dansmith: so i think there is some funky interaction between specifying block_device_mapping_v2.disk_bus vs haveign hw_disk_bus=scsi on the image
17:10:22 sean-k-mooney as far as im waware we dont offically supprot having multiple diffent disk buses in a nova inastance at once
17:10:32 sean-k-mooney however using bdms you could achive that
17:12:18 sean-k-mooney i tought on of the geneal probelms with that was that we did nto have the correct code to create teh requried contolers
17:12:43 sean-k-mooney i.e. scsi contoler in this case but i think your hiiting that other code path
17:13:52 dansmith sean-k-mooney: I don't know that's true.. without the hw_scsi_model I can still request the disk_bus=scsi, and I get a lsilogic controller, I just don't have the drivers
17:14:08 sean-k-mooney dansmith: so i think that a bug
17:14:18 sean-k-mooney or libvirt defaulting is happeing
17:14:24 sean-k-mooney if you try doing that for stata
17:14:29 sean-k-mooney i dont think it will work
17:14:54 dansmith the docs for hw_scsi_model also don't say that you can request lsi, it only says that if it's set, you can set it to virtio, which kinda implies it's not virtio by default
17:15:55 sean-k-mooney my understanding was hw_scsi_model was only ment to be coniderd if hw_disk_bu, hw_cdrom_bus or hw_rescuse bus was set to scsi
17:16:33 sean-k-mooney i.e. i did not exepct it to be used just bey settign disk_bus=scsi in the bdms
17:16:38 sean-k-mooney evedently it can be
17:16:44 dansmith I guess it doesn't seem reasonable to force all block devices to scsi in order to be able to have one as scsi
17:16:57 dansmith meaning, sata for everything and scsi for this special direct lun volume makes sense, no/
17:17:09 sean-k-mooney well we dont support having diffent disk on diffent busses in general
17:17:35 sean-k-mooney i.e. if you have root swap and multiple ephmeral volumes
17:17:51 sean-k-mooney then by default they are all specified vai hw_disk_bus
17:18:12 sean-k-mooney you could use the bdm api with destion_type=local
17:18:20 sean-k-mooney at leat for epmeral disk i think
17:18:30 sean-k-mooney but i have never seen that documented or recommended before
17:19:29 dansmith well, we do for ide cd and sata/scsi root disk right?
17:19:33 sean-k-mooney i always found block_device_mapping_v2.disk_bus (Optional) odd because i didnt knwo fo a supproted usecase wehre it could ever be diffent then the root disk
17:19:47 dansmith given the BDM takes a disk_bus, I think it's sort of implied that you should be able to ... you know, select it
17:19:48 sean-k-mooney for q35 ya
17:20:04 sean-k-mooney its implied but i dotn think we docuement or test it
17:20:31 sean-k-mooney like this has been somethign that has bugged me for years because i tied to do that in the past and it didnt appear to work
17:20:50 sean-k-mooney so i have alwasy wondered why its there if it does nto work
17:21:17 sean-k-mooney anyway you made it work but to me this still feels like a grey area
17:23:43 sean-k-mooney our docs https://docs.openstack.org/nova/latest/user/block-device-mapping.html#block-device-mapping-v2 still say "Leaving these empty is the most common thing to do" with regards to disk_bus and device_type
17:25:19 sean-k-mooney so i alwasy tough these we hold overs form bdm v1 with no reall usecase now outside fo device_type=cdrom for cinder vomues created form install isos
17:25:29 dansmith IMHO it shouldn't be grey.. it might be because of other problems we have, but I see no reason not to allow picking the bus per device
17:25:47 sean-k-mooney we can my issue is not with the feature
17:25:51 sean-k-mooney its with the testing fo that
17:26:11 sean-k-mooney i.e. i dont think we have tempest coverage of this in general
17:28:26 sean-k-mooney as far as im aware the only tempest test we hav efor attaching a device to a speicic disk bus is https://github.com/openstack/tempest/blob/ccd034eb3a916ce4d3955fa11fe905aa50cbc15d/tempest/api/compute/admin/test_volume.py#L93
17:28:37 sean-k-mooney and that does it via the image properties
17:29:31 sean-k-mooney dansmith: with that said your partly adressing that now
17:32:21 opendevreview Dan Smith proposed openstack/nova master: Fix device_type=lun with boot_index https://review.opendev.org/c/openstack/nova/+/918470
17:32:28 dansmith sean-k-mooney: this fixes the other problem ^ with BFV
17:32:44 dansmith I have a tempest change to test BFV with multi-attach type=lun, which seems to work fine
17:32:52 dansmith I'll get them all re-swizzled after this run finishes

Earlier   Later