Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-02-10
23:32:41 melwitt sean-k-mooney: afaik request spec has always been in the api db i.e. it didn't exist prior to that. I didn't move anything related to it :)
23:33:42 sean-k-mooney its existed since like juno it was made an object back in 2014 https://github.com/openstack/nova/commit/2d998c8df2048167c2db3694d3aae4605ff2625c
23:34:38 melwitt I mean request spec
23:34:57 sean-k-mooney oh sorry i said request spec
23:35:07 sean-k-mooney i ment instace_pci_request
23:35:34 sean-k-mooney sorry im currently trying to figure out why the numa policy i am setting is not makeing it to the db
23:36:43 sean-k-mooney when i grab it out of the db {"count": 1, "spec": [{"physical_network": "public"}], "alias_name": null, "is_new": false, "numa_policy": null, "request_id": "6ca3ee2f-59d7-4828-b95a-f503f7eb261d", ...
23:37:11 melwitt are you doing like request_spec.pci_requests = <stuff> and then request_spec.save()?
23:37:29 sean-k-mooney the instance_extra table to be exact its reset to null
23:37:34 melwitt how are you setting them
23:37:40 sean-k-mooney am im not doing the save but ill show you one sec
23:38:38 sean-k-mooney https://review.opendev.org/c/openstack/nova/+/773792/1/nova/network/neutron.py#2151
23:39:12 sean-k-mooney we are modifying it just above too spec[pci_request.PCI_TRUSTED_TAG] = str(trusted)
23:39:44 sean-k-mooney no not quite
23:39:51 sean-k-mooney the oter values are all passed in
23:40:03 melwitt oh, I can see this object doesn't work like the typical
23:40:28 sean-k-mooney well it should get commits with the rest of the requests
23:40:40 sean-k-mooney create_resource_requests
23:41:01 sean-k-mooney is invoked during the build process and we pass back a set of resouce requests
23:41:41 sean-k-mooney technically i gues there realy is no reason i could not pass the affinity policy to the consturor i guess
23:41:53 sean-k-mooney but what im doing should work
23:42:56 melwitt generally, the pci tracker is the thing that writes all pci related things to the db, so my first thought is maybe it doesn't handle numa_policy yet?
23:45:18 sean-k-mooney i really hope https://github.com/openstack/nova/commit/59d94633518e6f6272e9f0654bb908e332f97a96 added that
23:45:38 sean-k-mooney or else i have to talk to stephenfin about his commit naming
23:45:42 melwitt oh, hm
23:45:46 melwitt lol
23:47:20 sean-k-mooney i mean i know im missing unit/functional tests in that change and when i add them i might find out why its not working but still
23:47:22 sean-k-mooney numa_policy = port_numa_policy or affinity_policy
23:47:24 sean-k-mooney if numa_policy:
23:47:26 sean-k-mooney request.numa_policy = numa_policy
23:47:29 sean-k-mooney that should work
23:48:31 melwitt yeah.. I don't see anything either
23:49:45 sean-k-mooney i think tomorrow ill just start writing a functional test. i need to anyway but i was hoping to test it manually once first to make sure it worked
23:50:07 sean-k-mooney because its being lost before it gets to the db it does not.
23:50:08 melwitt seems like it would have to mean port_numa_policy or affinity_policy are both None?
23:50:19 sean-k-mooney yep
23:50:26 sean-k-mooney affinity_policy is None
23:50:28 melwitt or empty in some way
23:50:37 sean-k-mooney well none or empty ya
23:50:47 sean-k-mooney port_numa_policy in my testing is required
23:50:58 sean-k-mooney well
23:51:04 sean-k-mooney the policy on the port is required
23:51:22 sean-k-mooney that does not mean i dont have a different bug
23:52:10 sean-k-mooney im reading the value form the port here https://review.opendev.org/c/openstack/nova/+/773792/1/nova/network/neutron.py#2053
23:52:21 sean-k-mooney i fixed the spelling localy
23:52:29 melwitt is it possible is null or empty from neutron itself?
23:53:38 sean-k-mooney not any more alos https://review.opendev.org/c/openstack/nova/+/773792/1/nova/network/neutron.py#3039
23:53:43 melwitt or the constant is wrong?
23:54:07 sean-k-mooney i also extended the nova VIF object
23:54:20 sean-k-mooney to store it and if i look in the network info cache its set to required
23:54:32 sean-k-mooney and i create the port before i booted the vm
23:54:40 sean-k-mooney so i could check it was set
23:54:57 sean-k-mooney so the VIF object makes it to the db fine
23:55:01 melwitt oh ok. hrm
23:55:12 sean-k-mooney ya its weird
23:55:36 sean-k-mooney i did spell consttants wong but thats what i fixed locally
23:55:56 sean-k-mooney oh....
23:56:05 sean-k-mooney that would do it https://review.opendev.org/c/openstack/nova/+/773792/1/nova/network/neutron.py#2039
23:57:04 sean-k-mooney if i dont ask for the field... then my numa_policy = port.get(consttants.NUMA_POLICY, None) is always going to be None
23:57:47 sean-k-mooney we normlly dont filter like that so i never check for it
23:58:10 melwitt oh oops
23:58:42 melwitt at least you found it :)
23:58:46 sean-k-mooney yep
23:59:07 sean-k-mooney i did have a probalem before where i also forgot to trun the extention on in neutorn fully
23:59:38 sean-k-mooney so i would create the port, osc would show the policy in the port in the respocne but neutorn never saved it to the db on its end
#openstack-nova - 2021-02-11
00:00:30 melwitt sounds like good times all around
00:00:49 sean-k-mooney neutron list all exteions that are avaiabel not just the ones that are enabled which is __tottaly__ inutitive and not confusing at all....
00:01:32 sean-k-mooney im not sure what the best way to express sarcasium is on irc
00:01:38 melwitt yeah.. whenever I wander into other projects code where they still have extensions, it confuses me
00:02:07 melwitt that was clear sarcasm :) some people use /s though
00:02:08 sean-k-mooney i was originally affrait i forgot a mircoversion and it silent droped it or something
00:02:27 sean-k-mooney *affraid
00:02:52 sean-k-mooney ok ill fix that try it again and if it doesnt work its tomrrows problem
00:03:14 melwitt sounds like a plan
00:03:18 sean-k-mooney although its technially now 00:03 here so i guess todays problem
00:06:31 melwitt 🤞😬🤞
00:15:21 sean-k-mooney hum ok
00:15:27 sean-k-mooney so that didnt work
00:15:37 melwitt :(
00:15:47 sean-k-mooney and its not in the pci request in either the api db or instance_extra table
00:15:54 sean-k-mooney but both of those are not the same either
00:16:30 sean-k-mooney the instance_extra table version has "alias_name": null, "is_new": false, "numa_policy": null,
00:16:40 sean-k-mooney which are not in the request spec version at all
00:17:13 sean-k-mooney so i think this is not geting saved back into the request spec
00:17:44 sean-k-mooney and then on the compute node i guess we load it form the db which save the default for numa_policy back
00:17:59 sean-k-mooney and well the other fields
00:19:00 sean-k-mooney although no
00:19:10 sean-k-mooney if we loaded it from the db teh default would be different
00:19:12 sean-k-mooney https://github.com/openstack/nova/blob/5b52db409b00e6257e83598efa0d8528ebe380e0/nova/objects/instance_pci_requests.py#L99-L100
00:20:09 sean-k-mooney thats just weird but ya not nows problem
02:08:11 openstackgerrit Merged openstack/placement master: Implement secure RBAC for reshaper https://review.opendev.org/c/openstack/placement/+/774827
02:08:20 openstackgerrit Merged openstack/placement master: policy: Add releasenote for RBAC work https://review.opendev.org/c/openstack/placement/+/774654
09:12:34 bauzas and type hints :p
09:12:40 bauzas stephenfin: :p
09:13:26 stephenfin I don't get the dislike, tbh. They're really just structured comments that are (usually) checked for correctness
09:13:33 stephenfin who doesn't want self-checking comments :)
09:13:39 stephenfin *self-validating?
09:13:59 stephenfin I can stick in a follow-up patch to add them though, seeing as I have them worked out already :D
09:16:26 kashyap stephenfin: FWIW, yeah, I definitely see the value of function annotations
09:20:13 kashyap Although, IIRC, they annotations aren't enforced at runtime

Earlier   Later