Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2023-07-06
10:27:35 sean-k-mooney so simply updating os-vif is not enough the vms need tobe moved ro rebooted to have ovs port recreated
10:27:56 sean-k-mooney so i didnt want operators to think cool i patch it so my performance will imporve
10:28:21 sean-k-mooney it will for new instances or old ones once recreated via a hard reboot or move op
10:29:32 sean-k-mooney i could have just called that out in the fixes section but my thinking was when you are upgrading you would wnat to know that you shoudl live migratie the instace or comunicate to your custoemres that a reboot will improve the perfermance if they are expirince low network througput
10:46:32 elodilles sean-k-mooney: it's not a problem you added it in the upgrade section, it attracts more attention :)
10:46:42 elodilles and thanks for the explanation
10:46:56 elodilles i've +2'd the patch
10:46:57 sean-k-mooney hehe thats the point of release notes
10:47:09 elodilles yepp :)
10:47:34 sean-k-mooney in terms fo the backprot woudl you prefer to remove the default for the config option
10:48:04 sean-k-mooney we woudl have to slightly modify the test to set it and add a new test to assert its not set by default
10:48:18 sean-k-mooney but i think that would be pretty simple
10:48:25 elodilles well, it's not about my preference, really :) the thing is that a change should not change default behaviour when an operator upgrades to a new release
10:49:07 elodilles especially if it's a change that could cause issues for operators
10:49:30 sean-k-mooney ya the tricky part here is because a bug in ovs we did not see that theree was actully a change in behavior when we swapped form libvirt plugging to os-vif pluging the interface
10:49:51 sean-k-mooney well it cant as far as i am aware
10:50:20 sean-k-mooney the patch only sets teh linux-noop qos policy on the port if the port does not exist on teh bridge
10:50:56 sean-k-mooney so this woudl only cause an issue if that was not supported by your ovs version
10:51:17 elodilles and we don't always realize what can cause issues ;) so to be on the safe side, should chose opt-in config ;)
10:51:18 sean-k-mooney to my knoldage this has been supproted for many many years in ovs
10:51:33 elodilles (on stable branches)
10:51:38 sean-k-mooney yep
10:51:50 sean-k-mooney that is why i normally do this as two patches on master
10:52:01 sean-k-mooney but i forgot this time since we had some heat to fix this downstream
10:52:51 sean-k-mooney as i said im find with making it opt in in the backport if thats what other think we should do
10:52:54 elodilles i see. yepp, it's better when it is separated into 2 patches
10:54:05 sean-k-mooney there is no opt out in the master version because when libvirt was doing this it alwasy set the noop qdisc on the port
10:54:06 elodilles sean-k-mooney: i mean, generally it is the best way of working to follow. of course if a team decides that this is fine like this, i'm not against then.
10:55:09 sean-k-mooney of couse that was not documented anywhere which is why we missed it orgianlly
10:55:51 sean-k-mooney i dropped my +2 anyway until we come to a concensus
10:56:04 elodilles the opt-in version is always easy to backport, the opt-out ones need consideration by the team every time though. to put it this way. o:)
10:58:04 sean-k-mooney correct but there is no opt out in master becasue we did not think opting out was a reasonable thing to do. that said i coudl see addign that option to master as a sperate patch and then backporting that as the stabel default. i.e. add "disable" to the enum and make that the default in stable
10:59:44 sean-k-mooney basicaly the master version alwasy creates the port with linux-noop by default (you can enables other QOS by default too) and then all QOS config should then be doen via neutron
11:00:04 sean-k-mooney instead of setting the default via /sys/...
11:04:11 sean-k-mooney what changed in the kernel was in older kernels settign a default qos in /sys/... did not get applied to tap devices
11:04:46 sean-k-mooney at somepoitn that change so any operator using that to set the default qos for there physical nic suddenly got that qos applied to the vm port on upgrading there kernel
11:19:18 elodilles ACK, thanks for the explanation :)
12:08:02 sean-k-mooney bauzas: gibi pushed my comments with a +1 on https://review.opendev.org/c/openstack/nova-specs/+/887011
12:08:34 sean-k-mooney can you take a look and let me knwo if this is somethign we are ok defering to the impelmetaion and fixing up after spec freeze
12:10:15 sean-k-mooney i agree with the overall content and think we should proceed with this im just not convince by the storage of the keymanger uuid isn a flavor extra spec or image poroperty
12:10:29 sean-k-mooney and i would like to disucss that with melwitt when she is back form PTO
12:10:37 sean-k-mooney but i dont want to really hold the spec on that
12:11:55 bauzas sean-k-mooney: okay, so you could accept the spec and just tell you would like to discuss about some implementation question after the spec freeze
12:12:17 opendevreview Merged openstack/nova-specs master: Re-propose "Policy service role spec" https://review.opendev.org/c/openstack/nova-specs/+/881880
12:12:36 sean-k-mooney ya so if peopel are ok with that and potially updating the spec to refelct what we decied in the implemation reivew then i can approve as is
12:13:15 sean-k-mooney i just dont think we should be providing a way for end users to specify the keymangaer secret uuid
12:13:29 sean-k-mooney espically when the end user cant know if its barbican or something else
12:14:00 sean-k-mooney we also dont want to force a image/flavor per instance
12:14:15 sean-k-mooney and we obvioulys done want to share keeys between instnaces
12:14:37 sean-k-mooney so recording the uuid somehwere in our db sure (ideally instance_system_metadta)
12:14:51 sean-k-mooney but using flavor/image proeprty for that is a no in my book
12:18:11 bauzas ++
12:32:52 sean-k-mooney ok i have +2w'd the spec and -2'd the implemenation patch that added the image property https://review.opendev.org/c/openstack/nova/+/870935/4
12:33:06 sean-k-mooney ill drop that once we chat about it with melwitt
12:39:56 opendevreview Merged openstack/nova-specs master: Re-propose spec for ephemeral storage encryption https://review.opendev.org/c/openstack/nova-specs/+/887011
14:56:46 bauzas gibi: sean-k-mooney: other cores, we have 2 already +2'd specs, can you try to look at them today given the spec freeze ? https://review.opendev.org/q/project:openstack/nova-specs+status:open+label:Code-Review%253E%253D%252B2
14:57:51 bauzas also, I'm definitely not an expert of SmartNics, so could you please look at https://review.opendev.org/c/openstack/nova-specs/+/859290 ?
15:00:58 melwitt sean-k-mooney, bauzas: the image property is for the case of the image created from a snapshot, that has no instance associated with it. it's associated only with the image. that's the problem I need to solve
15:32:34 opendevreview Merged openstack/nova-specs master: Adds cleanup to remove dangling volumes https://review.opendev.org/c/openstack/nova-specs/+/878757
16:00:22 bauzas ralonsoh: I know it's pretty late in your day and I'm pinging you late but could you get a quick pass on https://review.opendev.org/c/openstack/nova-specs/+/859290 ?
16:00:49 bauzas today is Nova spec freeze and I don't know whether the Neutron community accepts this
16:00:52 ralonsoh let me check
16:00:58 bauzas thanks a lot
16:01:39 ralonsoh bauzas, ah yes, there was a missing question but is replied
16:01:42 ralonsoh I'll +1 it
18:40:17 opendevreview Merged openstack/nova-specs master: Propose tooling and docs for unified limits https://review.opendev.org/c/openstack/nova-specs/+/887014
21:20:49 opendevreview melanie witt proposed openstack/nova-specs master: Amend spec to add more details around encryption secrets https://review.opendev.org/c/openstack/nova-specs/+/887905
21:22:54 melwitt fyi sean-k-mooney ^
23:02:04 opendevreview melanie witt proposed openstack/nova-specs master: Amend spec to add more details around encryption secrets https://review.opendev.org/c/openstack/nova-specs/+/887905
#openstack-nova - 2023-07-07
07:28:57 opendevreview Danylo Vodopianov proposed openstack/nova master: Packed virtqueue support was added. https://review.opendev.org/c/openstack/nova/+/876075
10:31:05 bauzas elodilles: when you're around (it's lunch time), can you help me on https://review.opendev.org/c/openstack/releases/+/885365 ?
10:36:12 bauzas context : https://meetings.opendev.org/meetings/nova/2023/nova.2023-06-27-16.01.log.html#l-145
10:36:41 bauzas when gibi said 'someone needs to handle the fallout' :)
10:59:45 elodilles bauzas: o/
10:59:57 elodilles bauzas: so what is the question? :)
11:01:54 elodilles do you want to find the CI jobs that can be deleted OR do you want to prepare to the fallout in case we delete all nova deliverables in train? :)
11:09:19 opendevreview Elod Illes proposed openstack/nova stable/train: DNM: Experimental Fallout https://review.opendev.org/c/openstack/nova/+/887943
11:09:42 elodilles bauzas: let's see if this shows something regarding the 'fallout' :) ^^^
11:13:59 bauzas elodilles: ok, I'll look
11:18:47 elodilles bauzas: strangely zuul did not scream that some job definition is missing, so maybe this does not prove anything :/
11:24:38 elodilles bauzas: anyway, for the 'CI jobs that can be deleted': i've compared .zuul.yaml in ussuri and in train, and the following jobs were removed in ussuri AND are not defined in nova's .zuul.yaml: https://paste.opendev.org/show/bGpxq2Ww1bqGWQ7wlSHa/
11:25:01 bauzas ok, so should we cut stable/train ?
11:25:08 elodilles bauzas: some of the jobs are probably still used in other repos (of not all) ^^^
11:25:39 elodilles bauzas: so we have to figure out where the definitions are and IFF not used anywhere else, then propose patches to delete them
11:25:45 bauzas elodilles: ok so I'll modify https://review.opendev.org/c/openstack/releases/+/885365 to EOL all our repos
11:25:58 elodilles bauzas: ACK
11:26:25 elodilles and thanks in advance
11:51:22 sean-k-mooney elodilles: i think as long as the jobs are still on the branch we wont get a warnign form zuul but i could be wrong
11:51:44 sean-k-mooney i think we will only get that after its merged unless you try to do a depend on patch
11:52:05 sean-k-mooney if you co t oa repo that is using a delete job and depend on that patch then it should give you an error
11:52:35 sean-k-mooney something along the lines of "unable to sablaise job graph" or similar
12:28:14 elodilles sean-k-mooney: yepp, apparently it works like that :S
12:29:03 elodilles at least this works: https://review.opendev.org/c/openstack/openstack-zuul-jobs/+/887950
12:29:16 elodilles (the other way around)
12:30:36 elodilles bauzas: I've found only these 2 jobs that can be deleted after nova's stable/train goes to EOL ^^^
12:36:02 sean-k-mooney hopefully the dvsm job is one fo them...
12:36:27 sean-k-mooney it would be nice to finally get rid of all the zuul legacy jobs
12:38:23 elodilles yes. though there are other repos that uses some. and of course until old stable branches are not deleted these are likely 'used'
12:40:06 sean-k-mooney elodilles: fun https://bugs.launchpad.net/nova/+bug/2025813 apprenatly that affect nova-ceph-multistore too
12:41:22 sean-k-mooney validations are not disabled in nova

Earlier   Later