Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2025-02-27
22:57:35 sean-k-mooney kind of so we had a converation with the ironic folk about the fact that they wanted simialr metadat for ther own usage.
22:58:18 sean-k-mooney so i did a poc of factoring out the metadat functions https://review.opendev.org/c/openstack/nova/+/905406 which we combiend with there work
22:58:20 sean-k-mooney https://review.opendev.org/c/openstack/nova/+/923910/5/nova/virt/driver.py#49
22:58:27 sean-k-mooney and ^ were the calsess we ended up with
22:59:09 sean-k-mooney what i didnt end of pocign was basiclly the changes to print all fo the files in the xml
22:59:38 sean-k-mooney so for image you see it has id name and propereis
23:00:12 sean-k-mooney in my orginal poc i also had id name and "extra_specs" for flavor https://review.opendev.org/c/openstack/nova/+/905406/2/nova/virt/driver.py
23:00:33 sean-k-mooney so some of those files were doped in the version we merged because ironic didnt need them
23:03:00 sean-k-mooney Callum027: so lookign at my local git repo i dont seem to have my poc for extending the xml and i do not see it on gerrit either
23:03:19 sean-k-mooney so either i didnt commit it or i did but didnt push it anywaher
23:04:30 sean-k-mooney Callum027: lookign at your patch
23:04:43 sean-k-mooney you seam to be extendign the ground work we created 6 months ago
23:05:25 sean-k-mooney Callum027: its kidn fo funny because the value your addign are once i intetally was not going to incldue for image
23:06:02 sean-k-mooney to me the image prorpees were the most valueabel rather then min_(ram/disk) or the format
23:06:23 opendevreview Merged openstack/placement master: doc: Use dnf instead of yum https://review.opendev.org/c/openstack/placement/+/938499
23:06:27 opendevreview Merged openstack/nova master: doc: Use dnf instead of yum https://review.opendev.org/c/openstack/nova/+/938496
23:07:05 Callum027 seak-k-mooney: Yeah they are, the properties are what I actually want, but since they're included in image_meta by Nova notifications and Ceilometer uses them I thought it was a good idea to include them
23:07:10 sean-k-mooney Callum027: whiel at the saem time your not including the name or id
23:07:38 Callum027 Actually, I don't think min_ram etc are actually used by Ceilometer out of the box but why not I thought
23:08:11 sean-k-mooney so part of the reason i didnt spend too much time on this before beyond the fact i was busy with other things
23:08:24 sean-k-mooney is i knwo some of the nova core are concerned with not makign the metadta too log
23:08:58 sean-k-mooney on the ohter hand i have spend enough days of my life in extra debuging time trying to diagnose customer issues
23:09:08 sean-k-mooney where having this info would have made it seconds
23:09:25 sean-k-mooney so i think ther is value in provide at least some of it in the xml
23:11:24 sean-k-mooney Callum027: you have impleted it very close to how i woudl have done it so im mostly supprotive of this propsoal
23:11:44 sean-k-mooney Callum027: are you famialr with our process for proposing new features?
23:11:53 Callum027 I'm afraid not no
23:12:35 sean-k-mooney well the first step is generally either ot propose a blueprint or ask about adding the enhancement on irc or the mailing list
23:13:01 sean-k-mooney today is actully Feature Implemation Freeze for the 2025.1 release
23:13:38 sean-k-mooney so we are entring a code freeze period for about 2 week until the first release candiate for 2025.1 is created and then master reopens for development
23:14:19 sean-k-mooney Callum027: when a blueprint is proposed we useally ask the propsoer to bring it up in our next irc meeting to ask the wider core team to approve it as a specless bluepirnt
23:14:42 sean-k-mooney or if we think it need more detailed deicsion we may ask for a spec docuemnt
23:15:33 sean-k-mooney in this case i think it could be a specless blueprint but we will have to see how other feel
23:16:04 sean-k-mooney Callum027: also as we are at the end of one release syscycle and will be starting another shortly
23:16:27 sean-k-mooney we will have a virtual comunity event call the PTG (project teams gattering) april 7-11th
23:16:59 sean-k-mooney Callum027: so if there was not agreement on this and you wanted a higher bandwith venue to disucss it it could be dicussed there
23:17:40 sean-k-mooney Callum027: ah i just check you have contibuted to ceilometer and adjusat before
23:17:56 sean-k-mooney Callum027: so you are proably familar with the general flow.
23:19:29 Callum027 I have made contributions to Ceilometer and Adjutant before but mainly smaller patches to add features or fix issues that didn't require more fundamental changes to things like file formats
23:19:36 Callum027 So this explanation was very helpful, thank you
23:19:57 sean-k-mooney Callum027: based on https://review.opendev.org/c/openstack/ceilometer/+/942879 i take it yoru usecase for this is to consuem it in ceilometer and then publish the metadat to a telemetery backend of some kind?
23:21:13 sean-k-mooney Callum027: may i ask what is the end goal that this enables
23:21:19 Callum027 can provide this information as well.
23:21:19 Callum027 Yes, in particular we're working on an OpenStack billing service that takes telemetry and metadata submitted to Gnocchi by Ceilometer, one of the things we're billing for requires image metadata from the instance. *Just* relying on Nova notifications to supply this is not good enough, we want it in the libvirt XML metadata so Ceilometer's pollsters
23:21:38 sean-k-mooney ah ok
23:22:00 sean-k-mooney part of the reason i ask is i ahve recnetly been workign a bit on watcher
23:22:34 sean-k-mooney we ahve been addign the ablity to pull data form prometeous (and we have celomiter metrics feeding into promethous)
23:22:56 sean-k-mooney Callum027: so i was trying to decied if this would be useful in that context
23:23:42 sean-k-mooney watcher also has the ablity to consume the novaificaton but i was trying to decied if there was any benifit in consuming this info via the metrics form ceilometer
23:25:35 Callum027 published to Gnocchi.
23:25:35 Callum027 For us anyway there are benefits. The main issue is the reliability of Nova's notifications. For existing instances that are not changed frequently, Nova only sends out a compute.instance.exists notification once per hour or so. If Ceilometer fails to consume that notification for one reason or another from RabbitMQ, the sample metadata is not
23:25:56 sean-k-mooney well notificatoin are really ment to be fire and forget
23:26:11 sean-k-mooney they are not inteded as a reliable message deliveray mechanium
23:26:21 sean-k-mooney so ya i can see why that would nto be ideal for billing
23:26:30 Callum027 Yes, and for billing we want to accurately know the state of the resource at the time it was polled
23:26:37 Callum027 Otherwise we're not billing correctly
23:27:50 sean-k-mooney ya makes sesne. so next steps, RC1 is planed for 2 weeks time
23:28:22 sean-k-mooney the week after that will be the first irc meetign of the 2025.2 development cycle
23:28:47 sean-k-mooney that would be yoru first real opertunity to ask for the spec to be approved
23:28:58 sean-k-mooney *bluepirnt to be approved
23:29:37 Callum027 Alright, sounds good, so the meeting for the week of 17-23 March?
23:30:25 sean-k-mooney yes im not sure if this time https://meetings.opendev.org/#Nova_Team_Meeting suits you
23:30:40 sean-k-mooney if not you can ask for the blueprint to be approved on the mailing list
23:31:13 Callum027 Tuesday 1600 UTC is Wednesday 0700 NZDT so I should be okay for that
23:31:38 Callum027 In the meantime I'm happy to make any suggestions/additions you think would make the proposal fit your use case better
23:31:47 sean-k-mooney if its too early you can also just add the bluepirint to the adjeda and we will discss it without you
23:32:19 sean-k-mooney basifclly there is an open dicusstion section here https://wiki.openstack.org/wiki/Meetings/Nova
23:32:52 sean-k-mooney Callum027: the process to ask for it to be approved is too just add a secion with yoru name and link to it and a line or two about what it is and whay you want to enable it
23:33:14 sean-k-mooney Callum027: you have already done a pretty good job of descibing it in https://blueprints.launchpad.net/nova/+spec/xml-image-meta
23:33:54 sean-k-mooney Callum027: in term fo feed back i am not sure when ill have time to review the patch next but ill try and leave some comments on the review
23:34:42 Callum027 Awesome, thank you. Should I add the topic to the agent on the week of the first meeting for 2025.2 development, or is now fine?
23:34:47 Callum027 agenda*
23:35:15 Callum027 I should be able to be there at the meeting time as well
23:35:19 sean-k-mooney so you could add it now but it will come up next tuseday and we will proably be too busy with feature feeze stuff
23:35:41 sean-k-mooney so you coudl see but we might say lets wait until we are doen with 2025.1 first
23:36:24 Callum027 Sounds good!
23:37:49 sean-k-mooney just looking at this in the ci
23:37:51 sean-k-mooney https://zuul.opendev.org/t/openstack/build/aeae5e9ef4c042fb89555623e3be9466/log/controller/logs/screen-n-cpu.txt#9620
23:38:30 sean-k-mooney it loosk pretty reasonabel but if you look at the falvor vs the image you see the flavor has naem and the image has the uuid but in a sub element
23:41:06 sean-k-mooney that the relevent part https://paste.opendev.org/show/bbHRV4uzP2yZr8j7yPEW/
23:41:19 sean-k-mooney if i was to update it i would modify it slightly
23:41:59 sean-k-mooney we currently store the image uuid in <nova:root type="image" uuid="993fddc9-8bd4-48a4-b894-c71c2298ddf3"/>
23:44:42 sean-k-mooney i think i woudl want the xml to look like this https://paste.opendev.org/show/826974/
23:44:42 Callum027 Yes it is, so nova:baseImageRef is technically not required. The main reason I have that in there is because from notifications that value is set as image_meta.base_image_ref, and that is what Ceilometer uses to set image_ref in its sample metadata
23:45:03 Callum027 Ah, I see
23:45:17 sean-k-mooney so i added uuid to flavor
23:45:26 sean-k-mooney and add name and uuid to the image tag
23:46:00 sean-k-mooney put the top level image atribut at the top at the same level as proeprteis adn nested each propety in the properties arry
23:46:43 sean-k-mooney that ust makes the xml a little bit more self consitent
23:47:05 Callum027 That does look good, yeah. Just one more case this needs to cover, though: Instances can be booted from volumes that were originally created from image. In this case there are image_meta details we want, but the UUID is NOT set. As long as Nova, Ceilometer and anything else downstream can handle uuid being "", that should be fine.
23:47:37 sean-k-mooney yep that shoudl be fine
23:47:51 sean-k-mooney we coudl aslo encode if it BFV or not if that was helpfuly
23:48:13 sean-k-mooney but i think haveign empty sting woudl be fine too
23:48:43 Callum027 When implementing it I would probably check both for existence and empty string of the uuid field, yeah
23:49:04 Callum027 And in that case just set base_image_ref on the Ceilometer side to empty string to match the Nova notifications
23:49:37 sean-k-mooney if i was doign thei chagne i woudl also add the flavor extra spec too for my usecase
23:49:49 sean-k-mooney that a bit beyond what you orgnally needed
23:49:50 Callum027 Yep, I can do that
23:50:06 sean-k-mooney but my orgianl motivation was to aide in debugging
23:50:10 Callum027 It's beyond what I needed but it would address another major issue with Ceilometer because it needs to do a Nova API to get that info

Earlier   Later