| *** mriedem_away has quit IRC | 00:57 | |
| *** amodi has quit IRC | 05:08 | |
| *** takashin has quit IRC | 05:29 | |
| *** e0ne has joined #openstack-placement | 06:36 | |
| openstackgerrit | Pawel Baclawski proposed openstack/osc-placement master: Add support for 1.22 microversion https://review.opendev.org/651783 | 06:58 |
|---|---|---|
| openstackgerrit | Pawel Baclawski proposed openstack/osc-placement master: Add support for 1.22 microversion https://review.opendev.org/651783 | 07:00 |
| gibi | cdent: logo works for me | 07:09 |
| openstackgerrit | Pawel Baclawski proposed openstack/osc-placement master: Add support for 1.22 microversion https://review.opendev.org/651783 | 07:47 |
| *** takashin has joined #openstack-placement | 08:05 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: Spec for nested magic 1 https://review.opendev.org/662191 | 08:26 |
| openstackgerrit | Balazs Gibizer proposed openstack/os-traits master: Add COMPUTE_NET_VF_PARENT_NAME_REPORTING trait https://review.opendev.org/658852 | 09:21 |
| *** takashin has left #openstack-placement | 09:31 | |
| *** cdent has joined #openstack-placement | 09:33 | |
| *** cdent has quit IRC | 10:05 | |
| *** cdent has joined #openstack-placement | 10:39 | |
| cdent | efried: thanks for the reminder on the map: https://review.opendev.org/#/c/662439/ | 12:03 |
| *** cdent has quit IRC | 12:13 | |
| *** openstack has joined #openstack-placement | 12:30 | |
| *** ChanServ sets mode: +o openstack | 12:30 | |
| *** mriedem has joined #openstack-placement | 13:05 | |
| cdent | brief but sincere moments of joy: https://anticdent.org/placement-update-19-21.html | 13:52 |
| efried | cdent: what's our current policy on deleting consumers? | 14:32 |
| efried | Do they go away as soon as all their allocations go away? | 14:32 |
| cdent | pretty much | 14:33 |
| cdent | do you mean their db row, or their conceptual existence? | 14:33 |
| efried | we have a consumers table. So I mean their row in that table I guess | 14:33 |
| *** openstackstatus has joined #openstack-placement | 14:34 | |
| *** ChanServ sets mode: +v openstackstatus | 14:34 | |
| cdent | at the end of 'set_allocations' we call a method to delete the consumer if they have no allocations | 14:34 |
| cdent | 'delete_consumers_if_no_allocations' | 14:34 |
| efried | okay, I see it now, thank you. | 14:35 |
| cdent | there was lot of argument about that (and consumers in general) but the upshot was that from placement's standpoint (for now) a consumer is something it is only aware of in an ephemeral sense | 14:35 |
| efried | yeah, I recall all of that, just couldn't remember where we landed. Leaving consumers around forever, once they had their own row in a table, would have made no sense, because there will be a lot of those in a long-lived placement db, but I wanted to make sure. | 14:36 |
| cdent | ✔ | 14:37 |
| efried | cdent: Do we still have anything that we "leave around forever"? | 14:37 |
| cdent | aggregate uuids, maybe | 14:37 |
| efried | k | 14:37 |
| cdent | and CUSTOM traits and resource classes (which makes sense) | 14:37 |
| cdent | while I do imagine us wanting to become "million resource provider capable" I don't know that orphaned aggregates, even at that scale, will matter much | 14:38 |
| efried | CUSTOM traits, that's totally going to be a problem once people start doing CUSTOM_CREATED_AT_20190531_0940 | 14:40 |
| cdent | you spin that kind of rope into your life, you're welcome to hang? | 14:42 |
| cdent | people are welcome to create their own clean up tools in that case? | 14:43 |
| mriedem | the delete consumer once there are no more allocatoins was a thing that bit nova with a race in the gate, | 14:43 |
| mriedem | i believe the issue was shelve offload and then immiatedly unshelve, scheduling would fail b/c the consumer generatoin was gone or something, | 14:43 |
| mriedem | because we changed the vm status before removing the allocatoins | 14:43 |
| mriedem | it was fun | 14:43 |
| efried | good times | 14:43 |
| * cdent checks "fun" in the dictionary | 14:44 | |
| mriedem | agree that if people starting doing "openstack resource provider trait create CUSTOM_$(uuidgen)" then it's their own fault | 14:45 |
| cdent | especially since we don't create traits on the fly: an explicit creation step is required | 14:47 |
| cdent | cleaning them up out from under people would be a problem | 14:47 |
| efried | meh, as long as we don't allow deletion when in use, it doesn't matter to the client that the next time they use CUSTOM_FOO it has a different internal ID. | 14:50 |
| efried | but ^ assumes we went to creating traits automatically when used. | 14:50 |
| cdent | which, let's not | 14:52 |
| cdent | because then there's no protection against typos | 14:52 |
| efried | agreed. cause then we might need generations somewhere we don't have them yet, can't think through exactly where. | 14:53 |
| mriedem | someone can write a script to purge old unused traits after a given created_at time if they want | 14:57 |
| mriedem | in one of the 50 ops repos | 14:57 |
| cdent | 50+1 | 14:57 |
| openstackgerrit | Kashyap Chamarthy proposed openstack/os-traits master: hw: cpu: Rework the directory layout; add missing traits https://review.opendev.org/655193 | 14:58 |
| efried | cdent: 1:1 allocation:consumer? | 15:07 |
| efried | that's probably the wrong question | 15:08 |
| cdent | it depends on what you are calling an "allocation" | 15:08 |
| efried | yeah | 15:08 |
| -openstackstatus- NOTICE: Gerrit is now entering its maintenance window. Expect Gerrit outages in the near future. We will notify when it is back up and running. | 15:08 | |
| *** ChanServ changes topic to "Gerrit is now entering its maintenance window. Expect Gerrit outages in the near future. We will notify when it is back up and running." | 15:08 | |
| cdent | there are multiple rows in the allocation table per consumer | 15:08 |
| cdent | s/are/can be/ | 15:08 |
| cdent | efried: is there something you are thinking about that's leading to these questions? | 15:08 |
| efried | and can be against different resource providers | 15:08 |
| cdent | yes | 15:09 |
| efried | because sharing | 15:09 |
| efried | but we wouldn't enforce even if multiple nonsharing | 15:09 |
| efried | because tree | 15:09 |
| efried | but we wouldn't enforce even if multiple nonsharing in different trees | 15:09 |
| efried | because we don't. | 15:09 |
| cdent | i'm not parsing any of what you just said | 15:09 |
| efried | yeah, I'm reviewing the consumer types spec and bikeshedding on the name of the new 'count' field. | 15:10 |
| efried | It needs to say what it's a count of | 15:10 |
| efried | and I thought 'allocation_count', but it's really 'consumer_count' | 15:10 |
| efried | because "one allocation" doesn't really make sense, as noted above. | 15:11 |
| efried | "the group of one consumer's allocations" or "one row in the allocations table", and here the intent is the former, so we should just say "consumer" instead. | 15:11 |
| * cdent needs to re-review that spec | 15:11 | |
| * cdent usually does such things on a monday | 15:13 | |
| efried | cdent: wait for a) gerrit to be back up, b) me to post my blast of comments. | 15:13 |
| efried | Some of them are actually going to be good. | 15:13 |
| efried | So Monday might be best :) | 15:13 |
| * cdent withholds judgement | 15:13 | |
| cdent | If you suggest that consumers persist, I will get out a large knife and brandish it like a crazy person, throw it, and run away into the hills | 15:14 |
| *** e0ne has quit IRC | 15:14 | |
| edleafe | cdent: so just another Friday | 15:15 |
| cdent | pretty much | 15:16 |
| edleafe | Consumers that don't consume - what a concept! | 15:16 |
| mriedem | well i'd post my comments on the nested magic 1 spec but can't | 15:18 |
| mriedem | i'm not qualified to vote on it either way | 15:18 |
| edleafe | mriedem: why not? Can't you pull a rabbit out of a hat? | 15:19 |
| mriedem | not sure which comment that question is directed at | 15:19 |
| mriedem | assuming nested magic | 15:19 |
| mriedem | hence the rabbit, i get it | 15:19 |
| cdent | mriedem: if it makes insufficient sense for you to be able to vote or comment effectively, perhaps it needs to be improved and you should say so? | 15:20 |
| cdent | that said, it seems to take about 27 years of indoctrination to participate in a topic that includes numa or nested, so something like 98 for both... | 15:21 |
| mriedem | if i actually had a thing i needed to implement then i could stand it up against the proposed change and see if it would solve my problem, but i don't, besides the resourceless request group stuff | 15:21 |
| mriedem | which is fairly straight-forward | 15:21 |
| mriedem | i just have a general feeling of dread for both performance in general (to account for handling these types of requests), the complexity in maintaining it, and weird edge cases we haven't thought of or hit yet, but won't until we actually have something using this, which is likely nova/cyborg | 15:22 |
| mriedem | and once we do, | 15:23 |
| mriedem | we'll need even more query params to be like, "yeah so i want all of this stuff as before, but also now a new magical fairy unicorn knob b/c you're doing part of my thing wrong" | 15:23 |
| mriedem | "wrong" | 15:23 |
| cdent | I know that feeling well | 15:24 |
| cdent | I said much the same thing somewhere in the pre-ptg discussions | 15:24 |
| mriedem | i also tend to think that if the query param structure gets any more complicated than this, because of the need for more magical fairy unicorn knobs, we'll need to consider a json request payload | 15:24 |
| mriedem | just to visualize the damn thing | 15:24 |
| efried | ^ comes up pretty much every release | 15:25 |
| edleafe | mriedem: agree. And I think that is a pretty strong smell | 15:25 |
| edleafe | which is why I've generally pushed back againt it | 15:25 |
| edleafe | *against | 15:25 |
| mriedem | it's just an impl detail | 15:25 |
| * efried chauffeurs while gerrit is down | 15:25 | |
| mriedem | given my current state of corporate affairs i'm in a limbo of how much i need to care about numa or cyborg modeling in placement for that matter, | 15:26 |
| cdent | mriedem: speaking of visualisation, did you see this little toy a friend and I have been playing with: https://cdent.github.io/placeview/ | 15:26 |
| mriedem | so i'm not overly concerned on this stuff just yet :) | 15:26 |
| cdent | mriedem: we need you to reivew and vote anyway, unless we want me or eric to be self voting | 15:27 |
| mriedem | i have not | 15:27 |
| mriedem | regarding the spec, i'm assuming tetsuro is much more on top of this stuff than i am | 15:27 |
| cdent | yes, that's true | 15:27 |
| cdent | but you have a unique attention to detail skill that's very valuable, even when you don't "get it" | 15:28 |
| mriedem | ask my wife for her opinion about that particular skill of mine | 15:29 |
| cdent | ha! I'm sure she suffers greatly. | 15:31 |
| openstackgerrit | Merged openstack/placement master: Bump openstackdocstheme to 1.30.0 https://review.opendev.org/662356 | 15:41 |
| cdent | bbs | 15:45 |
| *** cdent has quit IRC | 15:46 | |
| *** cdent has joined #openstack-placement | 15:51 | |
| *** ChanServ changes topic to "See https://docs.openstack.org/placement/latest/ and https://developer.openstack.org/api-ref/placement/" | 16:00 | |
| openstackgerrit | Kashyap Chamarthy proposed openstack/os-traits master: hw: cpu: Rework the directory layout; add missing traits https://review.opendev.org/655193 | 16:02 |
| mriedem | gerrit is back yay, ok nested magic spec comments posted | 16:03 |
| cdent | thanks | 16:05 |
| efried | cdent: done with review | 16:07 |
| efried | of consumer types | 16:07 |
| cdent | skimming I see that I'm going to have to not skim before making a legit response as there's at least one thing that made me squirm | 16:09 |
| cdent | will do that monday | 16:10 |
| efried | cdent: Feel like approving https://review.opendev.org/#/c/655193/ before you head out? | 16:11 |
| cdent | i haven't been following that one closely, so I'd need to look properly given it had so much dancing | 16:11 |
| efried | couple separate efforts waiting on that, and depends-on doesn't work for reasons, so it would be nice to get that in a release. | 16:12 |
| cdent | well, we wouldn't likely do a release on friday, so waiting on my +W for monday morn is probably safe | 16:13 |
| *** mriedem is now known as mriedem_hangry | 16:14 | |
| cdent | efried: I'm timed out, if some of the other os-traits stuff is close and you think so, point it out and I'll get what I can early monday and see about starting the release process | 16:21 |
| efried | ight | 16:21 |
| * cdent waves | 16:25 | |
| *** cdent has quit IRC | 16:25 | |
| -openstackstatus- NOTICE: Gerrit is back up and running again. Thank you for your patience and sorry for the delay in this notification (we thought the statusbot was still busy updating channel topics). | 16:47 | |
| efried | stephenfin, mriedem_hangry: What problem are you seeing with that footnote? It wffm | 16:52 |
| stephenfin | efried: I was just going on what mriedem_hangry said but didn't check it myself | 16:52 |
| *** amodi has joined #openstack-placement | 17:18 | |
| *** mriedem_hangry is now known as mriedem | 18:44 | |
| *** openstackgerrit has quit IRC | 19:01 | |
| *** amodi has quit IRC | 19:47 | |
| *** e0ne has joined #openstack-placement | 20:07 | |
| *** e0ne has quit IRC | 20:12 | |
| *** e0ne has joined #openstack-placement | 20:13 | |
| *** e0ne has quit IRC | 20:30 | |
| *** openstackgerrit has joined #openstack-placement | 20:55 | |
| openstackgerrit | Eric Fried proposed openstack/placement master: Spec for nested magic 1 https://review.opendev.org/662191 | 20:55 |
| *** dklyle has quit IRC | 20:58 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!