| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2021-03-26 | |||
| 09:25:00 | swp20 | bauzas: placemet has support PCI filter such as GPU/FPGA devices, isn't it? | |
| 09:27:09 | gibi | vgpu mdevs are modelled in placement but PCI devices not yet | |
| 09:27:49 | gibi | due to qos min bandwidth scheduling support some sriov PFs are modelled in placement but that only used for the bandwdith resource at the moment | |
| 09:28:00 | gibi | the PCI schduling still happens in the PCIFilter in nova | |
| 09:30:59 | bauzas | swp20: we only model vGPUs | |
| 09:31:11 | bauzas | swp20: but we don't really model PCI devices | |
| 09:31:36 | bauzas | and given we also have NUMA PCI affinity, we would also need to support this | |
| 09:31:45 | bauzas | a long story here. | |
| 09:33:49 | swp20 | gibi, buazas: got, thanks a lot. | |
| 09:35:41 | openstackgerrit | OpenStack Release Bot proposed openstack/placement stable/wallaby: Update .gitreview for stable/wallaby https://review.opendev.org/c/openstack/placement/+/783232 | |
| 09:35:45 | openstackgerrit | OpenStack Release Bot proposed openstack/placement stable/wallaby: Update TOX_CONSTRAINTS_FILE for stable/wallaby https://review.opendev.org/c/openstack/placement/+/783233 | |
| 09:35:50 | openstackgerrit | OpenStack Release Bot proposed openstack/placement master: Update master for stable/wallaby https://review.opendev.org/c/openstack/placement/+/783234 | |
| 09:36:13 | openstackgerrit | OpenStack Release Bot proposed openstack/placement master: Add Python3 xena unit tests https://review.opendev.org/c/openstack/placement/+/783235 | |
| 11:25:26 | kashyap | gibi: or anyone else: Can I borrow your eyes: In this method here _add_rng_device() — do you see a way where 'hw_rng:allowed = False' on image property might *not* take effect? -- https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L5665,#L5682 | |
| 11:31:16 | openstackgerrit | Slawek Kaplonski proposed openstack/nova master: [neutron] Get only ID and name of the SGs from Neutron https://review.opendev.org/c/openstack/nova/+/783275 | |
| 12:24:17 | gibi | kashyap: the image_meta is unused in that function | |
| 12:24:41 | gibi | only the flavor extra_spec is used | |
| 12:24:53 | gibi | so that function does not care about any image property | |
| 12:25:23 | kashyap | gibi: Oh, so the "hw_rng:allowed" image meta is useless in this case then | |
| 12:28:50 | gibi | seem so | |
| 12:28:55 | gibi | it might be a bug | |
| 12:29:16 | kashyap | Hmm, OSP QE reported it to me... | |
| 12:29:47 | gibi | as the function gets an image_meta parameter but the it does not use it | |
| 12:31:11 | kashyap | gibi: Right; I guess we should: | |
| 12:31:19 | kashyap | flavor.extra_specs.get() or image_imeta.properties.get() ... | |
| 12:32:31 | gibi | yeah that could be a solution | |
| 12:34:37 | sean-k-mooney | we have a function for that | |
| 12:34:50 | sean-k-mooney | at lesate wehre we named then the same | |
| 12:35:27 | sean-k-mooney | hw_rng:allowed | |
| 12:35:37 | sean-k-mooney | shoudl not be use in image meta | |
| 12:35:47 | sean-k-mooney | only the flavor can allow it | |
| 12:37:05 | sean-k-mooney | https://github.com/openstack/glance/blob/master/etc/metadefs/compute-libvirt-image.json#L54-L59 is the only image metadata value | |
| 12:38:15 | kashyap | sean-k-mooney: You mean "hw_rng:allowed" can only be used by extra_spec and not image_meta? | |
| 12:38:21 | sean-k-mooney | yes | |
| 12:38:42 | sean-k-mooney | that was a desgin chocie | |
| 12:39:09 | sean-k-mooney | because we used to use /dev/random not /dev/urandom as you know | |
| 12:39:17 | sean-k-mooney | there were concerns about running out of entropy | |
| 12:39:26 | kashyap | sean-k-mooney: Wait. What you're pointing to is different: "hw_rng_model" --> this is useless now. | |
| 12:39:30 | sean-k-mooney | so it was an admin only extra spec | |
| 12:39:38 | sean-k-mooney | which defaut to false | |
| 12:39:59 | kashyap | sean-k-mooney: Okay ... so it's not a "bug" to not have it via image_meta | |
| 12:40:25 | sean-k-mooney | right it would be a bug to use it in image meta but the fact its not there is not | |
| 12:40:47 | sean-k-mooney | hw_rng_model is pretty useless not too yes | |
| 12:41:03 | sean-k-mooney | https://github.com/openstack/nova/blob/master/nova/objects/image_meta.py#L387-L394 | |
| 12:41:21 | sean-k-mooney | kashyap: but as you said we should not remove/deprecate it yet ^ | |
| 12:41:38 | kashyap | sean-k-mooney: You mean "as I said" in the commit message? :) | |
| 12:41:43 | sean-k-mooney | unless that comment you worte is also incorect | |
| 12:42:18 | kashyap | sean-k-mooney: No; it's not incorrect. I also wrote this in the commit message of the patch that introduced it: | |
| 12:42:21 | kashyap | - The 'hw_rng_model' image metadata property is now rendered | |
| 12:42:24 | kashyap | (temporarily) useless -- as it's not used anywhere outside the | |
| 12:42:26 | kashyap | _add_rng_device() method. But we don't want to deprecate it yet, as | |
| 12:42:29 | kashyap | we may extend it (see code comment for details); docucment that. | |
| 12:42:30 | kashyap | So we wan to keep it. | |
| 12:42:47 | sean-k-mooney | well there too but i was not reading the comit message | |
| 12:43:13 | sean-k-mooney | i rember this form when the rng supprot was first beign added | |
| 12:43:15 | kashyap | Righ; I wrote the code comment in image_meta.py | |
| 12:58:53 | openstackgerrit | Balazs Gibizer proposed openstack/osc-placement master: Update master for stable/wallaby https://review.opendev.org/c/openstack/osc-placement/+/781894 | |
| 13:02:22 | openstackgerrit | Balazs Gibizer proposed openstack/osc-placement master: Update master for stable/wallaby https://review.opendev.org/c/openstack/osc-placement/+/781894 | |
| 13:14:05 | dansmith | gibi: ah, based on the time, it must have been the last one before I went to bed, cool | |
| 13:19:27 | gibi | dansmith: yepp you almost saw the patch landing | |
| 13:19:30 | gibi | :) | |
| 13:21:36 | dansmith | well, sorry | |
| 13:22:54 | gibi | no worries. I'm glad you pushed the patch forwatrd | |
| 13:40:19 | openstackgerrit | Merged openstack/placement master: Update master for stable/wallaby https://review.opendev.org/c/openstack/placement/+/783234 | |
| 14:04:43 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: Add Python3 xena unit tests https://review.opendev.org/c/openstack/nova/+/783218 | |
| 14:10:28 | gibi | sean-k-mooney: could you please hit this two automatic os-vif patches https://review.opendev.org/c/openstack/os-vif/+/781891 ? | |
| 14:17:50 | openstackgerrit | Merged openstack/os-vif master: Update master for stable/wallaby https://review.opendev.org/c/openstack/os-vif/+/781890 | |
| 14:22:34 | openstackgerrit | Merged openstack/placement stable/wallaby: Update .gitreview for stable/wallaby https://review.opendev.org/c/openstack/placement/+/783232 | |
| 14:22:57 | openstackgerrit | Merged openstack/placement stable/wallaby: Update TOX_CONSTRAINTS_FILE for stable/wallaby https://review.opendev.org/c/openstack/placement/+/783233 | |
| 14:23:18 | openstackgerrit | Merged openstack/placement master: Add Python3 xena unit tests https://review.opendev.org/c/openstack/placement/+/783235 | |
| 14:23:54 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: Add Python3 xena unit tests https://review.opendev.org/c/openstack/nova/+/783218 | |
| 14:34:04 | sean-k-mooney | gibi: ill take a look now | |
| 14:34:26 | sean-k-mooney | gibi: a gmann has don it alredy | |
| 14:34:28 | gibi | sean-k-mooney: thanks | |
| 14:34:51 | gibi | ohh | |
| 14:35:03 | sean-k-mooney | gmann++ | |
| 14:35:15 | gibi | thanks then | |
| 15:08:12 | artom | sean-k-mooney, hey, in https://review.opendev.org/c/openstack/nova/+/783084/1/nova/tests/functional/libvirt/test_pci_sriov_servers.py#382 when you say it's a side effect of our Neutron fixture, what did you mean? | |
| 15:08:36 | artom | Btw, that should have been WIP :P | |
| 15:08:48 | artom | I'm just starting to explore wth's going on | |
| 15:08:52 | sean-k-mooney | i mean that in reality you will not get that key error | |
| 15:09:06 | sean-k-mooney | the pci_slot key will be set in the profile | |
| 15:09:12 | sean-k-mooney | it just will have the wrong value | |
| 15:09:22 | sean-k-mooney | so your test is failing but not for the correct reason | |
| 15:09:49 | sean-k-mooney | did you see my top level comment | |
| 15:09:57 | artom | sean-k-mooney, right, so... 1. what bit in the fixture isn't like reality? and 2. dvd seems to have gotten the same keyerror here: https://bugzilla.redhat.com/show_bug.cgi?id=1767797#c17 | |
| 15:09:59 | sean-k-mooney | i suggested how to write the test diffeerntly | |
| 15:09:59 | openstack | bugzilla.redhat.com bug 1767797 in openstack-nova "When unshelving an SR-IOV instance, the binding profile isn't reclaimed or rescheduled, and this might cause PCI-PT conflicts" [High,Assigned] - Assigned to alifshit | |
| 15:10:49 | artom | sean-k-mooney, yeah, I actually started like that, and then went back because... well, I wasn't thinking straight. Lemme do it "properly", and upload that I find | |
| 15:11:12 | sean-k-mooney | i think that was self inflicted | |
| 15:11:19 | sean-k-mooney | you can actully unshleve i hvae tested that | |
| 15:11:47 | sean-k-mooney | so the nasty part of this bug is it appears to work but its subtely broken | |
| 15:11:54 | valleedelisle | sean-k-mooney, can we unshelve on a host with the same pci device in use? | |
| 15:12:07 | artom | valleedelisle, oh that's your upstream nick | |
| 15:12:09 | valleedelisle | valleedelisle, I think that the point of that bz is the pci device not being re-calculated | |
| 15:12:16 | valleedelisle | yeah dvd is already in use hehe | |
| 15:12:17 | sean-k-mooney | valleedelisle: no that will cause the libvirt error in the upstream bug | |
| 15:12:30 | sean-k-mooney | libvirt/qemu will cay the device is already in use | |
| 15:12:38 | sean-k-mooney | but if the pci device is not in use then yes | |
| 15:12:46 | artom | valleedelisle, 'dvd :No such nick/channel' | |