| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-12-20 | |||
| 19:46:33 | edleafe | cdent: I'm not disagreeing | |
| 19:47:56 | cdent | I know | |
| 19:50:56 | mriedem | edleafe: i think we're pretty close on https://review.openstack.org/#/c/511358/ but i don't know why you had to modify the FilterScheduler in there | |
| 19:52:50 | mriedem | i think you might be working around something in the claims code in the conductor build_instances method that i pointed out | |
| 19:53:07 | mriedem | also, i think claim_resources might be the wrong method to call from conductor, seems put_allocations is what we want instead | |
| 19:55:15 | cdent | I got things done today, but none of the things I mean to to do, so I'm giving up | |
| 19:55:38 | mriedem | that's most of my days | |
| 19:55:55 | cdent | true | |
| 19:56:14 | cdent | but today seemed especially bad. or maybe I just feel guilty for not doing the stuff I said I'd do for efried_cya_jan | |
| 19:56:16 | cdent | anyway | |
| 19:56:19 | cdent | goodnight | |
| 19:56:26 | SamYaple | /win/win 20 | |
| 20:03:05 | edleafe | mriedem: the filter scheduler change was because the functional tests turned up an issue: that the Selection objects in the alternates had no allocation_request in them. Since the Selection objects weren't used until this patch, it never showed up | |
| 20:03:47 | mriedem | ummm | |
| 20:04:15 | mriedem | edleafe: then what is this? https://github.com/openstack/nova/blob/master/nova/scheduler/filter_scheduler.py#L380 | |
| 20:04:36 | mriedem | i realize the first selected host doesn't have an allocation_request in it https://github.com/openstack/nova/blob/master/nova/scheduler/filter_scheduler.py#L348 | |
| 20:04:51 | mriedem | which you did by design because you said conductor wouldn't need it, because we claim on that host during scheduling | |
| 20:05:01 | mriedem | but the alternates should have the allocation requests | |
| 20:08:59 | openstackgerrit | Chris Dent proposed openstack/nova master: [placement] Enable limiting GET /allocation_candidates https://review.openstack.org/513526 | |
| 20:09:04 | edleafe | mriedem: it looks like when I rebased on your patch, it pulled an older version. I'm not sure how | |
| 20:09:12 | mriedem | also realized just now that the claim logic in conductor won't work for the caching scheduler, which does have alternates but doesn't do claims | |
| 20:09:24 | mriedem | edleafe: ok, it seems like this is easily fixed | |
| 20:09:26 | edleafe | well, maybe no | |
| 20:09:41 | edleafe | yeah, I'll fix it, but I'm curious how that happened | |
| 20:09:53 | mriedem | my guess would just be like you said, rebase wonk | |
| 20:10:00 | mriedem | when there were lots of patches | |
| 20:10:13 | mriedem | note my latest comment in the conductor manager about the caching scheduler | |
| 20:10:34 | mriedem | we should have a test for that - where we have alternates, but they don't have allocation_request entries | |
| 20:10:40 | mriedem | we still reschedule, but we don't claim | |
| 20:11:05 | edleafe | ok, let me switch gears and start working on that | |
| 20:20:05 | mriedem | i think we should probably run one of our CI jobs with the caching scheduler, but not sure which one | |
| 20:20:17 | mriedem | could be the nova-next job | |
| 20:20:52 | mriedem | if we threw it in an experimental queue job we'd likely break it and not realize | |
| 20:33:15 | mriedem | https://review.openstack.org/529410 | |
| 20:48:57 | edleafe | mriedem: so for the caching scheduler case, I should just assume that if there is no allocaiton_request, that it's fine to build on the host? | |
| 20:50:20 | mriedem | edleafe: i think so yeah | |
| 20:50:35 | mriedem | b/c we still get an alternates list for the caching scheduler | |
| 20:50:42 | mriedem | it just won't have anything to do with placement | |
| 20:50:55 | edleafe | do we get alternates for caching? | |
| 20:51:12 | edleafe | or is it just a single-element list per instance? | |
| 20:51:48 | edleafe | oh wait | |
| 20:52:01 | edleafe | yeah, it still uses the filter scheduler underneath | |
| 21:02:58 | edleafe | so calling put_allocations() instead of claim_resources() would mean having to hack into the allocation_request to extract the resource provider | |
| 21:03:15 | edleafe | One of the goals of allocation_request was it should be opaque | |
| 21:03:38 | edleafe | Just because we've violated that once doesn't mean we should continue to do so in other cases. | |
| 21:04:01 | edleafe | So what if I added a "don't check for moves" param to claim_resources? | |
| 21:08:50 | mriedem | let me look at what you're talking about | |
| 21:10:15 | mriedem | edleafe: oh yeah you're right, | |
| 21:10:19 | mriedem | didn't think about the rp_uuid parameter | |
| 21:10:21 | mriedem | nevermind then | |
| 21:10:32 | mriedem | we have the move thing fixed by deleting the allocatoins in compute before casting to the cell conductor to do the reschedule | |
| 21:10:35 | mriedem | good call | |
| 21:11:10 | mriedem | mtreinish: thanks for hitting those backports | |
| 21:16:55 | mtreinish | mriedem: sure, np | |
| 21:17:17 | mtreinish | I just want to eol newton :) | |
| 21:23:28 | mriedem | i just stepped in a pile of zaqar in the ML | |
| 21:23:50 | mriedem | did everyone else know something about user-space notifications using zaqar and how nova core is "dead set" against integrating with anything like this? | |
| 21:26:04 | edleafe | mriedem: I know that there is a general understanding among non-Nova teams that Nova won't cooperate unless it helps Nova | |
| 21:26:31 | mriedem | huh | |
| 21:27:11 | mriedem | could it be that we, god forbid, actually have to prioritize some things based on how many people we have to work on and review stuff? | |
| 21:27:40 | mriedem | pretty sure we cooperated on the py35 goal, and the wsgi goal | |
| 21:28:46 | jroll | mriedem: there was a discussion years and years ago about that thing | |
| 21:29:06 | jroll | I assume the answer was "not right now" and that means never | |
| 21:29:09 | mriedem | oh ok | |
| 21:29:22 | mriedem | and by years and years ago, it means the current nova cores are the fuckers that said 'not now' | |
| 21:29:42 | jroll | heh, of course | |
| 21:29:54 | jroll | also I think that depended on the creds thing aka the kevin fox thing | |
| 21:30:23 | mriedem | oh | |
| 21:30:42 | mriedem | ok anything to do with the creds fiasco is over my head | |
| 21:30:50 | jroll | :P | |
| 22:04:38 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: [placement] Add sending global request ID in delete https://review.openstack.org/525231 | |
| 22:04:58 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: [placement] Add sending global request ID in delete (2) https://review.openstack.org/526263 | |
| 22:05:15 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: [placement] Add sending global request ID in delete (3) https://review.openstack.org/526557 | |
| 22:05:41 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: [placement] Add sending global request ID in post https://review.openstack.org/526823 | |
| 22:05:58 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: [placement] Add functional tests for traits API https://review.openstack.org/524094 | |
| 22:06:25 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: Fix 500 error while passing 4-byte unicode data https://review.openstack.org/407514 | |
| 22:06:45 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: [placement] Fix getting placement request ID https://review.openstack.org/523606 | |
| 22:06:58 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: [cellv2] Improve getting BDMs in multiple cells https://review.openstack.org/521400 | |
| 22:14:02 | openstackgerrit | Chris Dent proposed openstack/nova master: Do not set allocation.id in AllocationList.create_all() https://review.openstack.org/529397 | |
| 22:40:01 | oomichi | alex_xu: yeah, I reviewed that. Thanks for your work. The other https://review.openstack.org/#/c/529207 needs to be updated in your patch sequence | |
| 23:07:29 | openstackgerrit | Ed Leafe proposed openstack/nova master: Don't try to delete build request during a reschedule https://review.openstack.org/528835 | |
| 23:07:30 | openstackgerrit | Ed Leafe proposed openstack/nova master: Make conductor pass and use host_lists https://review.openstack.org/511358 | |
| 23:07:30 | openstackgerrit | Ed Leafe proposed openstack/nova master: Change compute RPC to use alternates for resize https://review.openstack.org/526436 | |
| 23:07:37 | edleafe | mriedem: ^^ addressed your comments | |
| 23:07:44 | edleafe | and rebased your bottom patch | |
| 23:07:52 | edleafe | sorry | |
| 23:09:11 | mriedem | as long as i don't lose the +2 it's fine :) | |
| 23:13:55 | edleafe | you're still good :) | |
| 23:32:05 | mriedem | edleafe: so can we split the unnecessary-for-this-patch scheduler thing out of this change and out of the series and just have it on it's own? | |
| 23:41:50 | edleafe | mriedem: if I did that, I'd have to also undo the test cleanups that assumed that host_list[0] had an empty allocation_request | |
| 23:42:08 | mriedem | in test_filter_scheduler? | |
| 23:42:15 | mriedem | or test_conductor? | |
| 23:43:02 | edleafe | both | |
| 23:44:01 | edleafe | the test_filter_scheduler stuff could be broken out easily enough, but the conductor tests are more tangled | |
| 23:44:26 | mriedem | https://www.amazon.com/Orangutangled-Sudipta-Bardhan-Quallen/dp/1477847170 | |
| 23:44:29 | mriedem | you said the word of the day | |
| 23:44:30 | mriedem | ahhhhhhhhhhhhhh | |
| 23:44:44 | mriedem | sorry, it's a book i've had to read several times | |
| 23:45:30 | mriedem | alright nevermind then | |
| 23:45:42 | mriedem | just makes it hard to review big changes with unrelated stuff mixed into them | |