| *** fried_rolls1 has joined #openstack-meeting-alt | 00:48 | |
| *** fried_rolls has quit IRC | 00:50 | |
| *** fried_rolls1 is now known as fried_rolls | 00:50 | |
| *** fried_rolls is now known as efried | 01:07 | |
| *** rcernin has quit IRC | 01:40 | |
| *** rcernin has joined #openstack-meeting-alt | 01:40 | |
| *** hongbin has joined #openstack-meeting-alt | 01:49 | |
| *** cloudrancher has joined #openstack-meeting-alt | 02:22 | |
| *** yamamoto has quit IRC | 02:42 | |
| *** ijw has joined #openstack-meeting-alt | 03:28 | |
| *** hongbin has quit IRC | 03:39 | |
| *** yamamoto has joined #openstack-meeting-alt | 04:08 | |
| *** cloudrancher has quit IRC | 04:20 | |
| *** dtrainor has quit IRC | 04:51 | |
| *** cloudrancher has joined #openstack-meeting-alt | 05:06 | |
| *** cloudrancher has quit IRC | 05:21 | |
| *** yamamoto has quit IRC | 05:30 | |
| *** dave-mccowan has joined #openstack-meeting-alt | 05:36 | |
| *** jtomasek has joined #openstack-meeting-alt | 05:37 | |
| *** yamamoto has joined #openstack-meeting-alt | 05:39 | |
| *** cloudrancher has joined #openstack-meeting-alt | 06:01 | |
| *** dpawlik has joined #openstack-meeting-alt | 06:03 | |
| *** dtrainor has joined #openstack-meeting-alt | 06:05 | |
| *** apetrich has joined #openstack-meeting-alt | 06:05 | |
| *** e0ne has joined #openstack-meeting-alt | 06:23 | |
| *** cloudrancher has quit IRC | 06:26 | |
| *** ijw has quit IRC | 06:36 | |
| *** belmoreira has joined #openstack-meeting-alt | 06:36 | |
| *** e0ne_ has joined #openstack-meeting-alt | 06:41 | |
| *** e0ne has quit IRC | 06:44 | |
| *** belmoreira has quit IRC | 06:45 | |
| *** belmoreira has joined #openstack-meeting-alt | 06:47 | |
| *** giblet is now known as gibi | 06:53 | |
| *** rdopiera has joined #openstack-meeting-alt | 07:04 | |
| *** ijw has joined #openstack-meeting-alt | 07:06 | |
| *** rcernin has quit IRC | 07:06 | |
| *** ijw has quit IRC | 07:10 | |
| *** ircuser-1 has joined #openstack-meeting-alt | 07:20 | |
| *** alexchadin has joined #openstack-meeting-alt | 07:51 | |
| *** jcoufal has joined #openstack-meeting-alt | 08:07 | |
| *** alexchadin has quit IRC | 08:14 | |
| *** janki has joined #openstack-meeting-alt | 08:33 | |
| *** janki has quit IRC | 08:34 | |
| *** janki has joined #openstack-meeting-alt | 08:34 | |
| *** finucannot is now known as stephenfin | 08:38 | |
| *** rossella_s has quit IRC | 08:41 | |
| *** alexchadin has joined #openstack-meeting-alt | 08:42 | |
| *** derekh has joined #openstack-meeting-alt | 08:44 | |
| *** yamamoto has quit IRC | 08:45 | |
| *** jcoufal has quit IRC | 09:07 | |
| *** pbourke has joined #openstack-meeting-alt | 09:20 | |
| *** ianychoi_ has joined #openstack-meeting-alt | 09:30 | |
| *** ianychoi has quit IRC | 09:34 | |
| *** e0ne_ has quit IRC | 09:38 | |
| *** e0ne has joined #openstack-meeting-alt | 09:38 | |
| *** yamamoto has joined #openstack-meeting-alt | 09:40 | |
| *** e0ne has quit IRC | 10:08 | |
| *** ganso has joined #openstack-meeting-alt | 10:08 | |
| *** janki is now known as janki|lunch | 10:14 | |
| *** e0ne has joined #openstack-meeting-alt | 10:27 | |
| *** e0ne has quit IRC | 10:40 | |
| *** e0ne has joined #openstack-meeting-alt | 10:49 | |
| *** erlon has joined #openstack-meeting-alt | 10:50 | |
| *** jcoufal has joined #openstack-meeting-alt | 10:56 | |
| *** e0ne has quit IRC | 10:57 | |
| *** janki|lunch is now known as janki | 11:10 | |
| *** alexchadin has quit IRC | 11:20 | |
| *** alexchadin has joined #openstack-meeting-alt | 11:59 | |
| *** alexchadin has quit IRC | 12:05 | |
| *** alexchadin has joined #openstack-meeting-alt | 12:06 | |
| *** e0ne has joined #openstack-meeting-alt | 12:10 | |
| *** tpsilva has joined #openstack-meeting-alt | 12:16 | |
| *** leakypipes is now known as jaypipes | 12:24 | |
| *** sambetts_ is now known as sambetts | 12:35 | |
| *** jcoufal has quit IRC | 12:43 | |
| *** alexchadin has quit IRC | 12:43 | |
| *** alexchadin has joined #openstack-meeting-alt | 12:53 | |
| *** yamamoto has quit IRC | 13:07 | |
| *** yamamoto has joined #openstack-meeting-alt | 13:08 | |
| *** lbragstad has joined #openstack-meeting-alt | 13:17 | |
| *** belmorei_ has joined #openstack-meeting-alt | 13:30 | |
| *** belmoreira has quit IRC | 13:32 | |
| *** SteelyDan is now known as dansmith | 13:34 | |
| *** e0ne has quit IRC | 13:38 | |
| *** e0ne has joined #openstack-meeting-alt | 13:47 | |
| *** alex_xu has joined #openstack-meeting-alt | 13:48 | |
| *** helenafm has joined #openstack-meeting-alt | 13:51 | |
| *** yamamoto has quit IRC | 13:54 | |
| *** yamamoto has joined #openstack-meeting-alt | 13:55 | |
| *** yamamoto has quit IRC | 13:55 | |
| *** yamamoto has joined #openstack-meeting-alt | 13:55 | |
| *** yamamoto has quit IRC | 13:55 | |
| *** mriedem has joined #openstack-meeting-alt | 13:55 | |
| *** dustins has joined #openstack-meeting-alt | 13:55 | |
| *** beekneemech is now known as bnemec | 13:57 | |
| efried | #startmeeting nova_scheduler | 14:00 |
|---|---|---|
| openstack | Meeting started Mon Sep 24 14:00:13 2018 UTC and is due to finish in 60 minutes. The chair is efried. Information about MeetBot at http://wiki.debian.org/MeetBot. | 14:00 |
| openstack | Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. | 14:00 |
| *** openstack changes topic to " (Meeting topic: nova_scheduler)" | 14:00 | |
| openstack | The meeting name has been set to 'nova_scheduler' | 14:00 |
| mriedem | o/ | 14:00 |
| alex_xu | o/ | 14:00 |
| gibi | o/ | 14:00 |
| * bauzas waves | 14:00 | |
| edleafe | \o | 14:01 |
| efried | alrightythen | 14:01 |
| efried | #link agenda https://wiki.openstack.org/wiki/Meetings/NovaScheduler#Agenda_for_next_meeting | 14:01 |
| *** raildo has joined #openstack-meeting-alt | 14:02 | |
| efried | #topic last meeting | 14:02 |
| efried | #link last minutes: http://eavesdrop.openstack.org/meetings/scheduler/2018/scheduler.2018-09-17-13.59.html | 14:02 |
| efried | Thanks jaypipes for covering ^ while I was having Day From Hell. | 14:02 |
| *** openstack changes topic to "last meeting (Meeting topic: nova_scheduler)" | 14:02 | |
| efried | Any old business to discuss? | 14:02 |
| mriedem | grenade changes are ongoing | 14:02 |
| mriedem | https://review.openstack.org/#/c/604454/ | 14:03 |
| *** cloudrancher has joined #openstack-meeting-alt | 14:03 | |
| mriedem | it's a mess as most can probably imagine | 14:03 |
| efried | excellent. | 14:03 |
| mriedem | ^ currently depends on dansmith's db migration script, | 14:03 |
| mriedem | and a change to the neutron-grenade job so that we can clone openstack/placement into the CI run | 14:03 |
| mriedem | i'm not sure yet if grenade will actually have to do the placement install/config stuff that devstack normally does | 14:04 |
| mriedem | b/c we've got a weird chicken/egg situation where cdent's devstack change that does the new style placement setup depends on the grenade change | 14:04 |
| mriedem | otherwise ideally new devstack would have placement cloned and setup via the separate repo before the upgrade runs | 14:04 |
| *** hongbin has joined #openstack-meeting-alt | 14:05 | |
| mriedem | tl;dr i'm whacking the moles as i hit them | 14:05 |
| mriedem | also, dan is out the rest of the week after today so if we need updates to his script we'll have to make them | 14:05 |
| * efried gets ready to swap in getopt | 14:06 | |
| *** munimeha1 has joined #openstack-meeting-alt | 14:06 | |
| efried | I'd like to see the pg script commonized in some sane way, since it's 90% duplicated. | 14:06 |
| efried | I commented accordingly. | 14:07 |
| dansmith | my script is easily broken out into common bits and the mysql bits. | 14:07 |
| efried | yup, that'd be one way, lib-ify the common bits | 14:08 |
| dansmith | it would be trivial I think to just take --dbtype on one script | 14:08 |
| dansmith | and just call the right function I would think | 14:08 |
| efried | yup, that was what I was thinking | 14:08 |
| dansmith | or commonize it, either one | 14:08 |
| dansmith | I just wish we wouldn't, but.. | 14:08 |
| efried | whyzat? | 14:08 |
| dansmith | because we hardly even test pg and have said we're not making an effort to make sure it works | 14:09 |
| dansmith | all this script does is dump a set of tables but with some safeguards.. if you're running pg against community advice, I'm sure you're capable of doing that part yourself | 14:09 |
| efried | Fair enough, but there's clearly a want for the pg version since someone went through the trouble to propose it; and if we're going to have it at all, it might as well not be duplicated code. | 14:09 |
| dansmith | I don't disagree that if we're going to have it it should be commonized | 14:10 |
| efried | If we want to take a harder line on not supporting pg, we should downvote that patch I guess. | 14:10 |
| mriedem | we've already merged a grenade-postgresql job | 14:11 |
| mriedem | so we can test it | 14:11 |
| efried | that's nice | 14:11 |
| efried | Moving on? | 14:11 |
| efried | #topic specs and review | 14:12 |
| *** openstack changes topic to "specs and review (Meeting topic: nova_scheduler)" | 14:12 | |
| efried | #link Consumer generation & nrp use in nova: Series starting at https://review.openstack.org/#/c/591597 | 14:12 |
| efried | In a runway (ends Oct 4) | 14:12 |
| efried | Any discussion on this? | 14:12 |
| gibi | yep | 14:12 |
| bauzas | I'll sponsor it | 14:13 |
| gibi | efried: you have some concerns about the first patch | 14:13 |
| bauzas | s/sponsor/shepherd | 14:13 |
| mriedem | i had questions in it as well | 14:13 |
| gibi | efried: I tried to answer it inline | 14:13 |
| efried | Yeah, I haven't processed your response yet. Does what I'm saying at least make sense? | 14:14 |
| gibi | mriedem: yours are more like comments I have to fix (and I'm fixing right now) | 14:14 |
| gibi | efried: I'm not sure what you eventually suggests about the patch? shall we not do the delete patch at all? | 14:14 |
| efried | That was my thought, yes. | 14:14 |
| efried | Switching from DELETE to PUT {} doesn't buy us anything, and it makes it *look* like we're covering races when we're really really not. | 14:15 |
| gibi | efried: for delete allocation we have DELETE /allocations/<uuid> to ignore generation | 14:15 |
| efried | right | 14:15 |
| gibi | efried: for the rest of the allocation manipulate we don't have such a workaround | 14:15 |
| efried | Right | 14:15 |
| efried | But | 14:15 |
| gibi | efried: I really don't like ignorig generation in delete but handling it in every other case | 14:16 |
| efried | I think the scenarios are different | 14:16 |
| efried | like for initial allocs in spawn | 14:16 |
| efried | we ought to be able to count on the generation being None | 14:16 |
| efried | If it ain't, we should blow up righteously | 14:17 |
| gibi | right | 14:17 |
| efried | There's no corollary to that in the delete case. | 14:17 |
| gibi | there are places where nova needs to read the allocation from placement and then manipualte it and then PUT it back | 14:18 |
| efried | For other paths like migrate or resize, I agree we have a similar issue, in that we're basically retrieving the allocations *right* before we mess with them. So we're not closing much of a gap race-wise. | 14:19 |
| *** belmorei_ has quit IRC | 14:19 | |
| gibi | https://review.openstack.org/#/c/583667/24/nova/scheduler/utils.py | 14:19 |
| gibi | yeah | 14:19 |
| * jaypipes here now, sorry for lateness | 14:19 | |
| gibi | efried: what to do with those? there we cannot ignore the generation | 14:20 |
| efried | Though again for migration, if we attempt to claim for the migration UUID and the generation isn't None, that's the same as spawn, which is definitely a thing to guard against. | 14:20 |
| efried | idk, I just think the cure is worse than the disease for deletion. | 14:20 |
| gibi | efried: so when we read - manipulate - and put (like in force live migrate and in force evacuate) then we can still have to prepare for conflict | 14:21 |
| efried | yes | 14:22 |
| gibi | even if this is not something that happens frequently | 14:22 |
| gibi | I can drop the delete patch if others like jaypipes and mriedem agrees | 14:22 |
| mriedem | no comment | 14:23 |
| gibi | but in the rest of the cases we will have conflicts to prepare for | 14:23 |
| efried | yes | 14:23 |
| efried | I'm not suggesting getting rid of any of the others | 14:23 |
| gibi | efried: OK | 14:23 |
| efried | jaypipes: thoughts? | 14:23 |
| efried | or dansmith? | 14:23 |
| jaypipes | reading back still | 14:24 |
| gibi | jaypipes: how do you feel about still using DELETE /allocations/{uuid} to delete instance allocations instead of PUTting allocations: {} with generation | 14:24 |
| dansmith | if we don't use PUT, | 14:24 |
| dansmith | then we don't know if they have changed since we examined them right? | 14:24 |
| gibi | dansmith: right | 14:25 |
| jaypipes | gibi: I don't get why efried thinks your patch will not avoid races. | 14:25 |
| dansmith | like, maybe we shouldn't be deleting them if we thought they should be gone but they changed? | 14:25 |
| efried | dansmith: Point is that we only know something changed in the teeny window *within* this one report client method between the GET and the PUT. | 14:25 |
| dansmith | I thought we discussed in dublin even that it should be a put because delete with a body is weird, for exactly that reason | 14:25 |
| efried | which is not useful | 14:25 |
| jaypipes | gibi: if someone changes the allocations in between the initial read and the call to PUT {} then we will fail, which will prevent the race. | 14:25 |
| dansmith | it's not? | 14:25 |
| efried | because the window-of-change we care about strats when the delete is initiated | 14:26 |
| efried | and the vast majority of that window is *before* we hit this method. | 14:26 |
| efried | So switching from DELETE to PUT {} in this method makes it *look* like we're closing a race when we're really really not. | 14:26 |
| gibi | jaypipes: yes, this is what the patch does, but the window is small as efried pointed out | 14:26 |
| dansmith | isn't efried arguing that the window is so small that we shouldn't close it? | 14:27 |
| gibi | efried: we close one really small race. But if nova does not store the allocation then there is no way to close the others | 14:27 |
| dansmith | because that does not resonate with me | 14:27 |
| efried | dansmith: We're not closing it. | 14:27 |
| dansmith | it's a small window now, but in the future it might be larger if we change the way the workflow is arranged | 14:27 |
| efried | We're narrowing it a fraction. It's still wide open. | 14:27 |
| dansmith | efried: you say that because why.. because we're able to detect it and we'll just re-do it right? | 14:28 |
| efried | well, if we do that (retry) then the change is truly pointless. But we agreed we would make 409 a hard fail here. | 14:28 |
| gibi | dansmith: we are not even retrying the delete in this case | 14:28 |
| dansmith | I totally don't understand | 14:28 |
| gibi | dansmith: the end user can retry the delete | 14:28 |
| dansmith | gibi: ack, just trying to understand what efried is talking about | 14:29 |
| efried | The race window starts when the conductor receives the delete request. | 14:29 |
| efried | Lots of stuff happens, lots of stuff happens, then finally down in the compute service we hit the report client method. | 14:29 |
| jaypipes | dansmith: efried is saying that because the window the race condition exists in is super-tiny, he doesn't feel this patch is important. | 14:29 |
| efried | Then *within* that method, we do a GET followed by an immediate PUT. | 14:29 |
| dansmith | jaypipes: I thought he was arguing that it doesn't solve the race | 14:30 |
| efried | Right. It doesn't. | 14:30 |
| *** belmoreira has joined #openstack-meeting-alt | 14:30 | |
| efried | If allocations are changed anywhere in that "lots of stuff happens" timeframe, we'll miss it, ignore it, delete anyway. | 14:30 |
| dansmith | efried: I totally don't understand what the "conductor does lots of stuff" part has to do it | 14:30 |
| dansmith | okay, but in the grand scheme of things, | 14:31 |
| efried | So I'm saying this patch gives us a false sense of security that we've actually done something | 14:31 |
| efried | when we really haven't. | 14:31 |
| dansmith | the get followed by the put should include some "does the data I got in the GET make sense? Yes, okay, delete if it hasn't changed" | 14:31 |
| dansmith | which is the pattern we should be following, | 14:31 |
| dansmith | even if right now we just PUT $(GET ..) | 14:32 |
| dansmith | IMHO | 14:32 |
| dansmith | jaypipes: agree? | 14:32 |
| gibi | efried: without nova storing the instance allocation there is nothing to compare to so we cannot detect the race in a the big window | 14:32 |
| efried | You mean e.g. comparing the GET data with some understanding of what we think the allocation should look like? | 14:32 |
| efried | gibi: Precisely. | 14:32 |
| dansmith | efried: correct | 14:32 |
| dansmith | just because the code plays fast and loose on the client side right now doesn't mean it will be that way forever | 14:32 |
| jaypipes | dansmith: agree. but this particular code path doesn't use the reportclient's cached provider tree and so cannot check a known consumer generation. | 14:32 |
| efried | Okay, but the only thing we have that would allow us to do that is... the allocation in placement, isn't it? | 14:32 |
| dansmith | jaypipes: that's a transient and unfortunate state of the code at this moment though right? | 14:33 |
| jaypipes | dansmith: not sure? | 14:33 |
| efried | We've discussed and dismissed the idea of caching allocations in the past, I thought, but I suppose we could revisit that. | 14:33 |
| jaypipes | efried, gibi: where is delete_allocation_for_instance() called from at this pint? | 14:33 |
| jaypipes | point... | 14:33 |
| jaypipes | efried: it's not allocations we need/want to cache. it's consumer generation. | 14:34 |
| jaypipes | just another reason why *we should have a separate consumer endpoint*. | 14:34 |
| * jaypipes goes back into hole. | 14:34 | |
| efried | jaypipes: To answer your question, it's called from all over the place. | 14:34 |
| gibi | jaypipes: normal instance delete, local delete, some rollback cases like failing unshelve | 14:34 |
| efried | yeah, what gibi said | 14:35 |
| efried | filter scheduler too | 14:35 |
| jaypipes | well, then my view is that this patch doesn't make our existing situation any *worse* and solves a micro-race that might happen (very unlikely). I don't think it's a reason to not merge the patch as-is | 14:35 |
| gibi | deleting the allocation held by the migration uud after move | 14:36 |
| jaypipes | but I do acknowledge the problem that efried has outlined. | 14:36 |
| jaypipes | we could solve this long-term by caching consumer generations. | 14:36 |
| jaypipes | which would be made easier if we had a GET /consumers endpoint. | 14:37 |
| jaypipes | but I digress. | 14:37 |
| efried | Okay. I'll buy the change if can we put a prominent NOTE acking the issue. gibi cool? | 14:37 |
| *** dpawlik has quit IRC | 14:37 | |
| gibi | efried: super cool | 14:37 |
| jaypipes | I'm fine with that. | 14:37 |
| efried | Thanks for the discuss y'all. | 14:37 |
| gibi | thank you | 14:37 |
| efried | #agreed to keep https://review.openstack.org/#/c/591597/ in play with a NOTE acking the race issue | 14:37 |
| efried | moving on | 14:38 |
| efried | #link latest pupdate: http://lists.openstack.org/pipermail/openstack-dev/2018-September/134977.html | 14:38 |
| efried | If you wouldn't mind clicking through ^ and scrolling down to the specs section | 14:38 |
| efried | Give the titles a quick skim and see if there's anything you'd like to talk about. | 14:38 |
| mriedem | i haven't been reviewing specs | 14:39 |
| gibi | efried: about my specs open on placement, I'm totally OK deferring those while placement is in freezed state | 14:40 |
| jaypipes | efried: not that I'd like to talk about. just need to do the reviews. :( | 14:40 |
| *** dpawlik has joined #openstack-meeting-alt | 14:40 | |
| jaypipes | gibi: does min bandwidth sched depend on any-traits? | 14:40 |
| *** dpawlik has quit IRC | 14:40 | |
| gibi | jaypipes: the multisegment use case depends on any-traits but I think that can wait | 14:41 |
| jaypipes | ack | 14:41 |
| gibi | jaypipes: if a network only maps to a single physnet then we are good | 14:41 |
| efried | I think alex_xu merged gibi's specs, not sure what the bp tracking process does at this point. mriedem? | 14:41 |
| jaypipes | right. | 14:41 |
| gibi | I think mriedem approved the bp after the spec was merged | 14:41 |
| gibi | and mriedem also noted that placement is freezed in the bp | 14:42 |
| efried | but now we're saying we want to defer to Train? Or just let it ride and *probably* defer to Train at the end of the cycle? | 14:42 |
| gibi | efried: I more like the later | 14:43 |
| alex_xu | we freeze to Train? | 14:43 |
| efried | We've got a handful of placement bps in play for this cycle. I was under the impression we were assuming placement would un-freeze at some point, early enough for us to get work done. | 14:43 |
| mriedem | it won't unfreeze until grenade is done | 14:43 |
| mriedem | so that's my focus | 14:43 |
| efried | We're not truly frozen at the moment, fwiw. The openstack/placement repo is open for patches; we've been merging stuff that's not purely related to extraction. | 14:44 |
| mriedem | what kinds of stuff? | 14:44 |
| efried | I agree we shouldn't merge features until the upgrade stuff is sorted. | 14:44 |
| gibi | efried: still I'd not start steeling review time from the extraction with any-traits and similar features | 14:44 |
| bauzas | can we call Anna for unfreezing ? | 14:44 |
| * bauzas makes sad jokes | 14:44 | |
| efried | mriedem: Nothing big, mainly refactoring. E.g. for reducing complexity. | 14:45 |
| efried | actually, looking at it, nothing significant has merged, but there's open patches along those lines. | 14:46 |
| efried | anyway, my point is, the code could get proposed and reviewed while upgrade stuff is being worked. | 14:46 |
| efried | We don't have to keep hands off the repo entirely. | 14:47 |
| efried | okay, any other specs or reviews to bring up specifically? | 14:48 |
| *** Leo_m has joined #openstack-meeting-alt | 14:48 | |
| efried | #topic Extraction | 14:48 |
| efried | cdent is on PTO. Anyone have any updates beyond what mriedem talked about earlier? | 14:48 |
| *** openstack changes topic to "Extraction (Meeting topic: nova_scheduler)" | 14:48 | |
| bauzas | nothing but the fact I just uploaded a new revision for the libvirt reshape | 14:49 |
| bauzas | and then I'll try to look whether it works | 14:49 |
| bauzas | with my machine | 14:49 |
| jaypipes | bauzas: link? | 14:49 |
| efried | #link vgpu reshape https://review.openstack.org/#/c/599208/ | 14:50 |
| *** dave-mccowan has quit IRC | 14:50 | |
| jaypipes | danke | 14:50 |
| bauzas | jaypipes: https://review.openstack.org/#/c/599208/5 | 14:50 |
| bauzas | shit too late | 14:50 |
| * efried holsters six-shooter | 14:50 | |
| efried | #topic bugs | 14:50 |
| efried | #link Placement bugs https://bugs.launchpad.net/nova/+bugs?field.tag=placement | 14:50 |
| *** openstack changes topic to "bugs (Meeting topic: nova_scheduler)" | 14:50 | |
| bauzas | we should also discuss about reshapes | 14:50 |
| efried | okay | 14:51 |
| bauzas | eg. should we have some specific module for reshapes ? | 14:51 |
| efried | #topic reshapes | 14:51 |
| *** openstack changes topic to "reshapes (Meeting topic: nova_scheduler)" | 14:51 | |
| efried | bauzas: module to do what? | 14:51 |
| bauzas | efried: for example, say I need a reshape for vGPU | 14:51 |
| bauzas | efried: then, once we agree on NUMA, we could have yet another reshape | 14:51 |
| bauzas | then, say PCPU will need a new reshape | 14:51 |
| bauzas | etc. | 14:51 |
| efried | You think there's some portion of the reshaping algorithm in update_provider_tree that could be common for all reshapes in all virt drivers? | 14:51 |
| bauzas | so, I was thinking of having a specific module that upgraders would use (like FFU folks) | 14:52 |
| efried | or are you saying we should ask virt drivers to keep those things somewhere separate for ease of review/maintenance? | 14:52 |
| bauzas | efried: maybe having a pattern, just that | 14:52 |
| bauzas | so we could see all the reshapes | 14:53 |
| bauzas | then, knowing for example when they were created, and for which cycle | 14:53 |
| bauzas | I mean, I dunno | 14:53 |
| efried | I think the idea has merit, if only for the sake of not having to keep adding and removing virt driver code every cycle. Keep 'em all together, kind of like we do for db upgrades? | 14:53 |
| bauzas | yup | 14:54 |
| bauzas | if so, I'll provide a new revision for https://review.openstack.org/#/c/599208/ | 14:54 |
| bauzas | and people will discuss in it ^ | 14:54 |
| efried | bauzas: Maybe you can write up something that demonstrates what you're talking about? | 14:54 |
| bauzas | yeah, the above ^ | 14:54 |
| efried | okay. | 14:55 |
| efried | #topic opens | 14:55 |
| *** openstack changes topic to "opens (Meeting topic: nova_scheduler)" | 14:55 | |
| efried | We're going to run out of time, but I put this one on there | 14:55 |
| efried | How to handle min_unit (and others) in a forward-looking (i.e. generic NRP) way? | 14:55 |
| efried | #link IRC discussion with belmoreira http://eavesdrop.openstack.org/irclogs/%23openstack-placement/%23openstack-placement.2018-09-20.log.html#t2018-09-20T14:11:59 | 14:55 |
| efried | My take was that, if we're going to have any hope of allowing operators to configure things like allocation ratios, min/max units, etc. in the future (with nrp/sharing) then we're going to need a generic solution that doesn't get us into the config nightmare we're currently experiencing with alloc ratios. | 14:56 |
| *** cloudrancher has quit IRC | 14:57 | |
| efried | jaypipes suggested | 14:57 |
| efried | #link Spec (rocky, abandoned): Standardized provider descriptor file https://review.openstack.org/#/c/550244/ | 14:57 |
| efried | which almost gets us there, but falls a bit short in that it doesn't solve the chicken/egg of being able to identify a provider before you can tweak it. | 14:57 |
| gibi | efried: I agree but right now I have no good sollution | 14:58 |
| efried | yeah | 14:58 |
| efried | the efforts around device passthrough | 14:58 |
| efried | #link Spec: Modelling passthrough devices for report to placement ("generic device management") https://review.openstack.org/#/c/591037/ | 14:58 |
| efried | #link Spec: Generic device discovery policy https://review.openstack.org/#/c/603805/ | 14:58 |
| efried | are leading towards defining file formats in a similar spirit | 14:58 |
| efried | but I think that fundamental problem still exists - how do I identify a provider that's going to be automatically generated by nova (or neutron or cyborg or cinder or...) | 14:59 |
| efried | ...in order to provide customized inventory settings for it? | 14:59 |
| bauzas | mmmm | 14:59 |
| efried | Since we're out of time, consider ^ food for thought. | 14:59 |
| bauzas | maybe we should autogenerate this file if needed ? | 14:59 |
| bauzas | like we do for options ? | 14:59 |
| bauzas | a getter/setter way | 15:00 |
| jaypipes | efried: I specifically left out the "identification of the provider before you need it" because the clients of such a descriptor file would undoubtedly have different ideas of how to map local identifiers to RP identifiers. | 15:00 |
| efried | it would have to be updated frequently. basically kept in sync with placement. | 15:00 |
| efried | That's time, folks. Let's continue in -nova if desired. | 15:00 |
| efried | #endmeeting | 15:00 |
| *** openstack changes topic to "OpenStack Meetings || https://wiki.openstack.org/wiki/Meetings/" | 15:00 | |
| openstack | Meeting ended Mon Sep 24 15:00:35 2018 UTC. Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4) | 15:00 |
| openstack | Minutes: http://eavesdrop.openstack.org/meetings/nova_scheduler/2018/nova_scheduler.2018-09-24-14.00.html | 15:00 |
| openstack | Minutes (text): http://eavesdrop.openstack.org/meetings/nova_scheduler/2018/nova_scheduler.2018-09-24-14.00.txt | 15:00 |
| openstack | Log: http://eavesdrop.openstack.org/meetings/nova_scheduler/2018/nova_scheduler.2018-09-24-14.00.log.html | 15:00 |
| *** cloudrancher has joined #openstack-meeting-alt | 15:02 | |
| *** helenafm has left #openstack-meeting-alt | 15:02 | |
| *** cloudrancher has quit IRC | 15:06 | |
| *** mriedem has left #openstack-meeting-alt | 15:11 | |
| *** dave-mccowan has joined #openstack-meeting-alt | 15:12 | |
| *** yamamoto has joined #openstack-meeting-alt | 15:13 | |
| *** dave-mccowan has quit IRC | 15:17 | |
| *** dave-mccowan has joined #openstack-meeting-alt | 15:22 | |
| *** jesusaur has quit IRC | 15:26 | |
| *** dustins has quit IRC | 15:26 | |
| *** dustins has joined #openstack-meeting-alt | 15:30 | |
| *** jesusaur has joined #openstack-meeting-alt | 15:31 | |
| *** efried has quit IRC | 15:37 | |
| *** dpawlik has joined #openstack-meeting-alt | 15:38 | |
| *** efried has joined #openstack-meeting-alt | 15:39 | |
| *** dpawlik has quit IRC | 15:42 | |
| *** alexchadin has quit IRC | 15:47 | |
| *** e0ne has quit IRC | 16:01 | |
| *** cloudrancher has joined #openstack-meeting-alt | 16:02 | |
| *** lbragstad has quit IRC | 16:10 | |
| *** lbragstad has joined #openstack-meeting-alt | 16:18 | |
| *** janki has quit IRC | 16:20 | |
| *** spotz is now known as spotz_ | 16:25 | |
| *** spotz_ is now known as spotz | 16:25 | |
| *** rdopiera has quit IRC | 16:27 | |
| *** dave-mccowan has quit IRC | 16:38 | |
| *** ttsiouts has joined #openstack-meeting-alt | 16:38 | |
| *** ttsiouts has quit IRC | 16:45 | |
| *** ganso has quit IRC | 16:49 | |
| *** diablo_rojo has joined #openstack-meeting-alt | 16:51 | |
| *** derekh has quit IRC | 16:59 | |
| *** dpawlik has joined #openstack-meeting-alt | 17:00 | |
| *** belmoreira has quit IRC | 17:04 | |
| *** alex_xu has quit IRC | 17:05 | |
| *** dave-mccowan has joined #openstack-meeting-alt | 17:35 | |
| *** sambetts is now known as sambetts|afk | 17:39 | |
| *** munimeha1 has quit IRC | 17:47 | |
| *** cloudrancher has quit IRC | 17:48 | |
| *** cloudrancher has joined #openstack-meeting-alt | 17:49 | |
| *** dpawlik has quit IRC | 17:51 | |
| *** e0ne has joined #openstack-meeting-alt | 18:06 | |
| *** dpawlik has joined #openstack-meeting-alt | 18:17 | |
| *** dpawlik_ has joined #openstack-meeting-alt | 18:17 | |
| *** dpawlik has quit IRC | 18:18 | |
| *** ianychoi_ is now known as ianychoi | 18:54 | |
| *** panda has quit IRC | 18:56 | |
| *** yamahata has quit IRC | 19:20 | |
| *** iyamahat has quit IRC | 19:20 | |
| *** iyamahat has joined #openstack-meeting-alt | 19:33 | |
| *** yamahata has joined #openstack-meeting-alt | 19:35 | |
| *** panda has joined #openstack-meeting-alt | 19:37 | |
| *** dpawlik_ has quit IRC | 19:51 | |
| *** dpawlik has joined #openstack-meeting-alt | 19:55 | |
| *** dpawlik has quit IRC | 19:56 | |
| *** erlon has quit IRC | 20:05 | |
| *** raildo_ has joined #openstack-meeting-alt | 20:11 | |
| *** raildo_ has quit IRC | 20:12 | |
| *** dpawlik has joined #openstack-meeting-alt | 20:13 | |
| *** raildo has quit IRC | 20:13 | |
| *** dpawlik has quit IRC | 20:18 | |
| *** e0ne has quit IRC | 20:37 | |
| *** dpawlik has joined #openstack-meeting-alt | 20:40 | |
| *** slaweq has quit IRC | 21:01 | |
| *** amito has quit IRC | 21:10 | |
| *** amito has joined #openstack-meeting-alt | 21:11 | |
| *** dpawlik has quit IRC | 21:12 | |
| *** jtomasek has quit IRC | 21:37 | |
| *** cloudrancher has quit IRC | 21:42 | |
| *** cloudrancher has joined #openstack-meeting-alt | 21:43 | |
| *** dustins has quit IRC | 22:01 | |
| *** dpawlik has joined #openstack-meeting-alt | 22:16 | |
| *** dpawlik has quit IRC | 22:20 | |
| *** panda is now known as panda|off | 22:38 | |
| *** rcernin has joined #openstack-meeting-alt | 22:45 | |
| *** hongbin has quit IRC | 22:45 | |
| *** dims_ has joined #openstack-meeting-alt | 22:59 | |
| *** mmedvede_ has joined #openstack-meeting-alt | 23:00 | |
| *** stephenfin_ has joined #openstack-meeting-alt | 23:00 | |
| *** dims has quit IRC | 23:00 | |
| *** jlvillal has quit IRC | 23:00 | |
| *** strigazi has quit IRC | 23:00 | |
| *** mmedvede has quit IRC | 23:00 | |
| *** stephenfin has quit IRC | 23:00 | |
| *** mmedvede_ is now known as mmedvede | 23:00 | |
| *** cloudrancher has quit IRC | 23:01 | |
| *** andreykurilin has quit IRC | 23:03 | |
| *** purplerbot has quit IRC | 23:04 | |
| *** cloudrancher has joined #openstack-meeting-alt | 23:05 | |
| *** cloudrancher has quit IRC | 23:05 | |
| *** andreykurilin has joined #openstack-meeting-alt | 23:05 | |
| *** cloudrancher has joined #openstack-meeting-alt | 23:07 | |
| *** cloudrancher has quit IRC | 23:07 | |
| *** cloudrancher has joined #openstack-meeting-alt | 23:10 | |
| *** cloudrancher has quit IRC | 23:10 | |
| *** cloudrancher has joined #openstack-meeting-alt | 23:12 | |
| *** tetsuro has joined #openstack-meeting-alt | 23:24 | |
| *** cloudrancher has quit IRC | 23:41 | |
| *** tpsilva has quit IRC | 23:43 | |
| *** erlon has joined #openstack-meeting-alt | 23:46 | |
| *** cloudrancher has joined #openstack-meeting-alt | 23:51 | |
| *** cloudrancher has quit IRC | 23:57 | |
| *** Leo_m has quit IRC | 23:58 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!