| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2022-01-27 | |||
| 10:10:03 | opendevreview | Frode Nordahl proposed openstack/nova master: Ensure binding profile keys added by Nova are cleared https://review.opendev.org/c/openstack/nova/+/826601 | |
| 10:41:42 | gibi | bauzas: can we land this? https://review.opendev.org/q/topic:bug%252F1952941 I'm getting pressure from downstream... | |
| 11:14:34 | opendevreview | Dmitrii Shcherbakov proposed openstack/nova master: [yoga] Include pf mac and vf num in port updates https://review.opendev.org/c/openstack/nova/+/824833 | |
| 11:31:13 | guesswhat | is xen ( xcp-ng ) supported in openstack as nova driver? seems its deprecated, right ? | |
| 11:44:05 | dmitriis | the same config for various PCI devices which may or may not be present in all servers. So I am trying to decide whether implementing runtime checks during pci_passthrough_whitelist parsing is a good idea or not. | |
| 11:44:05 | dmitriis | sean-k-mooney: o/ question about https://review.opendev.org/c/openstack/nova/+/824834/1/nova/pci/stats.py#530. Looking at pci_passthrough_whitelist alone without runtime checks we can't tell if a device is a PF or not. There may be a case where a device is present in the config but not yet hot-plugged into the host or deployment tooling may apply | |
| 12:51:05 | sean-k-mooney | dmitriis: its not that tere wont be any runtime checks its that when we first inistalise the pci tracker and iterate over the pci devcied on the host we can check once and either raise an error that prevent the agent form starting or strip out the pf depending on the configurtion | |
| 12:53:51 | sean-k-mooney | dmitriis: you do not use the same config on all servers | |
| 12:53:59 | sean-k-mooney | the pci whitelist is differnt per host | |
| 12:54:44 | sean-k-mooney | i dont enforce this today but its logcially an error to list a device in the pci whitelist if its not present on the host. | |
| 12:55:01 | sean-k-mooney | the only reason we dont enforce it is it could alredy be pass through to a vm | |
| 12:55:16 | dmitriis | sean-k-mooney: What about the hot-plug case? I know it's a niche one and I haven't actually seen this to be used | |
| 12:55:42 | sean-k-mooney | dmitriis: we do not support hotplug of pci device on the host | |
| 12:56:10 | sean-k-mooney | libvirt caches the pci device list and you woudl have ot restart it for libvirt to work and then restart the nova agent | |
| 12:56:21 | dmitriis | sean-k-mooney: ah, ok, so then it's not a problem and I can make a runtime check when whitelist is parsed | |
| 12:56:30 | sean-k-mooney | so for the lifetime of the nova agent hot-plug on the host is not supported | |
| 12:56:37 | sean-k-mooney | yes | |
| 12:56:53 | sean-k-mooney | we can ensure the PFs never actully make it to the db in the first plance | |
| 12:57:05 | sean-k-mooney | and therefore can never be a candiate for selection | |
| 12:57:22 | dmitriis | sean-k-mooney: right, presumably the compute node will filter out PCI devices obtained from libvirt based on the whitelist | |
| 12:58:05 | dmitriis | sean-k-mooney: what about the "if PF is specified in a request with a remote_managed tag"? Do I need to support this? | |
| 12:59:33 | sean-k-mooney | well there wont be any pf with remote managed in the db because we filtered them at start up so it will be handeled automaticaly | |
| 12:59:54 | sean-k-mooney | vnic-type direct-phsyical will only match PFs without remote-managed=true | |
| 13:00:02 | sean-k-mooney | because that is all tha will exist | |
| 13:00:15 | sean-k-mooney | and vnic-type smartnic will only match VFs | |
| 13:00:20 | sean-k-mooney | so i think that is covered | |
| 13:00:48 | sean-k-mooney | you do not need to specificlyu request remote_mannaged=false in the driect-physical case | |
| 13:00:51 | dmitriis | sean-k-mooney: ok, so I can just drop the check here https://review.opendev.org/c/openstack/nova/+/824834/1/nova/pci/stats.py#505 | |
| 13:01:19 | dmitriis | basically we assume that those devices won't get into the DB so no additional filtering will be needed | |
| 13:01:52 | sean-k-mooney | yes that is more or less why im suggesting. you jst need to move the logic to the inital startup | |
| 13:02:07 | sean-k-mooney | so its done once per host rather then once per request | |
| 13:02:23 | sean-k-mooney | * once per host start up | |
| 13:02:58 | dmitriis | sean-k-mooney: ok, makes sense, I'll have a look at how/where to put those and resubmit | |
| 13:04:43 | dmitriis | sean-k-mooney: regarding os-traits, looks like the 2.7.0 package got built and pushed out but 2.6.0 is still referenced in upper-constraits https://github.com/openstack/requirements/blob/master/upper-constraints.txt#L382 | |
| 13:05:05 | dmitriis | I've seen a change to the release repo and the requirements repo | |
| 13:06:07 | dmitriis | ah, looks like this hasn't been merged yet https://review.opendev.org/c/openstack/requirements/+/826447 while it's approved | |
| 13:06:33 | sean-k-mooney | right i had to make a change to placment and rebase that to depend on it | |
| 13:06:51 | sean-k-mooney | https://review.opendev.org/c/openstack/placement/+/826486 | |
| 13:06:58 | sean-k-mooney | still need to be approved. | |
| 13:07:13 | sean-k-mooney | bauzas: can you quickly hit ^ | |
| 13:07:41 | dmitriis | sean-k-mooney: ack, I see | |
| 13:08:16 | sean-k-mooney | dmitriis: we have some packaging issue with os-traits and placment that dont really work with how our ci is | |
| 13:08:33 | sean-k-mooney | we may need to merge os-traits into placment to fix it | |
| 13:08:40 | sean-k-mooney | so i had to hack around it for now | |
| 13:08:57 | sean-k-mooney | until recently we did not have ci in the requirements repo that tested placement | |
| 13:09:36 | sean-k-mooney | so this is the first tiem we are seeign the cirular depency problem | |
| 13:10:24 | sean-k-mooney | dmitriis: by the way you and chateaulav both need the os-trait bump | |
| 13:11:18 | sean-k-mooney | the best way to handel that is to have a singel patch to nova that increases our lower-constirati and requiremets.txt to 2.7.0 and then have both of ye rebase your series ontop of that patch | |
| 13:11:53 | sean-k-mooney | i can prably go file that but if you or chateaulav want to file that patch to nova feel free too | |
| 13:12:16 | dmitriis | sean-k-mooney: ack, I'll propose a patch and make my change depend on it | |
| 13:12:56 | sean-k-mooney | cool can you ping chateaulav when you do or add them as a reviewer so they see it | |
| 13:13:37 | dmitriis | oslo_config.cfg.RequiredOptError: value required for option lock_path in group [DEFAULT] | |
| 13:13:37 | dmitriis | raise cfg.RequiredOptError('lock_path') | |
| 13:13:37 | dmitriis | File "/home/<user>/src/opendev/nova/.tox/py39/lib/python3.9/site-packages/oslo_concurrency/lockutils.py", line 183, in _get_lock_path | |
| 13:13:37 | dmitriis | Just trying to figure out why my local unit test runs are failing with | |
| 13:13:37 | dmitriis | sean-k-mooney: will do. | |
| 13:15:49 | sean-k-mooney | your not adding any locks are you | |
| 13:15:57 | sean-k-mooney | that seam unrelated | |
| 13:16:04 | sean-k-mooney | try reruning with tox -r | |
| 13:16:09 | sean-k-mooney | to regenerate teh tox venv | |
| 13:16:21 | sean-k-mooney | incase you have some out of date pacakges | |
| 13:16:52 | sean-k-mooney | or just rm the tox enve manually without -r and let it regenerate it | |
| 13:18:22 | dmitriis | sean-k-mooney: no, and the test cases aren't related to what I've changed. Tried nuking .tox before. Let's see if -r helps. | |
| 13:18:57 | sean-k-mooney | if you nuked . tox it wont | |
| 13:19:09 | sean-k-mooney | i can try running it locally what patch have you checked out | |
| 13:20:13 | opendevreview | Dmitrii Shcherbakov proposed openstack/nova master: [yoga] Include pf mac and vf num in port updates https://review.opendev.org/c/openstack/nova/+/824833 | |
| 13:20:26 | dmitriis | sean-k-mooney: ^ https://review.opendev.org/c/openstack/nova/+/824833 the latest here | |
| 13:21:33 | sean-k-mooney | funcitonal or unit | |
| 13:21:47 | dmitriis | sean-k-mooney: unit: `tox -e py39` | |
| 13:22:08 | sean-k-mooney | ack just running them now under py38 | |
| 13:22:18 | sean-k-mooney | if they pass ill install py39 and test again | |
| 13:23:08 | dmitriis | sean-k-mooney: ack, I'll spin up a container and do a clean check as well. | |
| 13:23:54 | sean-k-mooney | we do ocationally get unit test breakages when a new lib is released | |
| 13:24:36 | sean-k-mooney | although the last release was Oct 21, 2021 | |
| 13:24:41 | sean-k-mooney | so that is likely not the issue here | |
| 13:26:00 | dmitriis | ack | |
| 13:28:12 | sean-k-mooney | there was an oslo.limits release and olslo utils release in the last few days but also does not look like the woudl be at fault | |
| 13:28:25 | sean-k-mooney | so far i think things are pass ing but im gong to grab a drink | |
| 13:28:27 | sean-k-mooney | brb | |
| 13:28:48 | dmitriis | sean-k-mooney: thanks for checking, will try to figure out what's on my env that's causing it | |
| 13:38:13 | sean-k-mooney | got the same error here | |
| 13:38:42 | bauzas | sean-k-mooney: gibi: sorry folks, was on discussion with someone you know | |
| 13:38:46 | bauzas | looking at both your changes | |
| 13:38:46 | sean-k-mooney | https://paste.opendev.org/show/812399/ | |
| 13:39:16 | sean-k-mooney | dmitriis: so its in the block device tests so not related to your patch | |
| 13:39:31 | gibi | bauzas: no worries | |
| 13:40:19 | sean-k-mooney | this is failing here https://github.com/openstack/oslo.concurrency/blob/master/oslo_concurrency/lockutils.py#L172-L185 | |
| 13:42:14 | dmitriis | sean-k-mooney: yeah, just confusing. Got 1 failure only with py39 in a clean 21.10 container https://paste.opendev.org/show/812400/ | |
| 13:42:40 | dmitriis | hopefully it will just work in CI :^) | |
| 13:43:47 | sean-k-mooney | ah there was a fasterners release 2 days ago | |
| 13:43:54 | sean-k-mooney | https://pypi.org/project/fasteners/ | |
| 13:44:30 | sean-k-mooney | which we picked up | |
| 13:44:32 | sean-k-mooney | https://github.com/openstack/requirements/commit/c086fff066e40f3a87ff81fe72c8c16cdefe7c64 | |
| 13:54:19 | opendevreview | Dmitrii Shcherbakov proposed openstack/nova master: [yoga] Introduce remote_managed tag for PCI devices https://review.opendev.org/c/openstack/nova/+/824834 | |
| 13:54:20 | opendevreview | Dmitrii Shcherbakov proposed openstack/nova master: Filter computes without remote-managed ports early https://review.opendev.org/c/openstack/nova/+/812111 | |
| 13:54:20 | opendevreview | Dmitrii Shcherbakov proposed openstack/nova master: [yoga] Add support for VNIC_TYPE_SMARTNIC https://review.opendev.org/c/openstack/nova/+/824835 | |
| 13:54:21 | opendevreview | Dmitrii Shcherbakov proposed openstack/nova master: Bump os-traits to 2.7.0 https://review.opendev.org/c/openstack/nova/+/826675 | |
| 13:56:58 | dmitriis | sean-k-mooney: looking | |
| 13:57:20 | dmitriis | sean-k-mooney, chateaulav: os-traits version bump change: https://review.opendev.org/c/openstack/nova/+/826675 | |
| 13:57:32 | chateaulav | thanks | |