| *** takashin has joined #openstack-placement | 00:43 | |
| *** tetsuro has joined #openstack-placement | 02:45 | |
| *** Nel1x has quit IRC | 04:07 | |
| *** jaypipes has quit IRC | 04:27 | |
| *** tetsuro has quit IRC | 05:35 | |
| *** e0ne has joined #openstack-placement | 05:52 | |
| *** e0ne has quit IRC | 06:24 | |
| *** e0ne has joined #openstack-placement | 06:52 | |
| *** takashin has left #openstack-placement | 07:05 | |
| *** helenafm has joined #openstack-placement | 07:25 | |
| *** gibi is now known as giblet | 07:38 | |
| *** bauzas is now known as PapaOurs | 08:03 | |
| *** stephenfin is now known as finucannot | 08:27 | |
| *** ttsiouts has joined #openstack-placement | 08:32 | |
| openstackgerrit | Merged openstack/placement master: Move qs parsing to placement.lib.RequestGroup https://review.openstack.org/602495 | 08:36 |
|---|---|---|
| openstackgerrit | Merged openstack/placement master: Fix member_of doc in RequestGroup.dict_from_request https://review.openstack.org/605427 | 08:36 |
| *** ttsiouts has quit IRC | 09:15 | |
| *** ttsiouts has joined #openstack-placement | 09:18 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: DNM: Experimenting with integration gabbits https://review.openstack.org/601614 | 09:18 |
| openstackgerrit | Chris Dent proposed openstack/placement master: WIP: Add placeload to integration test https://review.openstack.org/602484 | 09:18 |
| *** tssurya has joined #openstack-placement | 09:46 | |
| *** cdent has joined #openstack-placement | 09:58 | |
| *** e0ne has quit IRC | 10:16 | |
| *** e0ne has joined #openstack-placement | 10:25 | |
| *** ttsiouts has quit IRC | 11:09 | |
| *** cdent has quit IRC | 11:37 | |
| *** ttsiouts has joined #openstack-placement | 12:21 | |
| *** mriedem has joined #openstack-placement | 12:29 | |
| *** nicolasbock_ has joined #openstack-placement | 12:32 | |
| *** s10 has joined #openstack-placement | 12:45 | |
| *** cdent has joined #openstack-placement | 13:06 | |
| *** yikun has quit IRC | 13:21 | |
| * efried ducks & covers | 13:25 | |
| *** efried is now known as fried_rice | 13:25 | |
| cdent | i look forward to reading that fried_rice but I need to move locations first | 13:41 |
| cdent | brb | 13:41 |
| *** cdent has quit IRC | 13:41 | |
| *** Nel1x has joined #openstack-placement | 13:56 | |
| *** s10 has quit IRC | 13:57 | |
| *** s10 has joined #openstack-placement | 13:59 | |
| *** cdent has joined #openstack-placement | 14:05 | |
| cdent | fried_rice: you prefer I jump on that or wait for others? | 14:13 |
| fried_rice | cdent: Well, I surely didn't ask you (or anyone) before I tossed that grenade into the bunker. | 14:14 |
| fried_rice | I just read the latest on the ironic profile thread and hit my tipping point of being sick and tired of the "originally intended as capabilities" argument. | 14:14 |
| cdent | a) I agree with you, b) I don't want to overly bias the discussion by less embedded people | 14:14 |
| fried_rice | If I had waited an hour to settle in, I probably wouldn't have said anything, again. | 14:14 |
| fried_rice | a) is good to know, and b) mm, interesting. | 14:16 |
| fried_rice | I guess yeah, there will always be time to enter the discussion later. But of course it's up to you. | 14:16 |
| *** jaypipes has joined #openstack-placement | 14:21 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: DNM: Experimenting with integration gabbits https://review.openstack.org/601614 | 14:28 |
| openstackgerrit | Chris Dent proposed openstack/placement master: WIP: Add placeload to integration test https://review.openstack.org/602484 | 14:28 |
| *** edleafe has quit IRC | 14:29 | |
| *** edleafe has joined #openstack-placement | 14:30 | |
| cdent | fried_rice: on (b) I'm super conscious that when those of us who are primarily focused on this stuff front load a discussion other people feel that things are already decided and they will have to shout loudly to be heard. I sometimes want to avoid that by letting a topic percolate before diving in | 14:33 |
| cdent | In this particular case I'm tempted to dive in early, because of how strongly I feel about (a) | 14:33 |
| *** s10 has quit IRC | 14:37 | |
| *** s10 has joined #openstack-placement | 14:39 | |
| melwitt | what would be an example of a non-boolean trait though, out of curiosity? things like the PapaOurs example of CUSTOM_I_HAVE_MORE_THAN_2_GPUS? because I could see that getting untenable to maintain for trying to represent different values | 14:43 |
| fried_rice | I'm _less_ concerned about non-boolean traits (though that's surely something that would be useful). | 14:45 |
| fried_rice | Things like identifying NUMA nodes, tagging the service that "owns" a provider | 14:46 |
| fried_rice | Even tagging physnets, though we seem to be okay with that one. Perhaps because it'll always be CUSTOM_ | 14:46 |
| melwitt | ok, I see. well, the owner thing we agreed would be a standard trait. the NUMA node identification, that one was where the naming convention talk came in, I think | 14:49 |
| melwitt | at a high level, I guess I can say I find it a lot easier to reason about traits when they're boolean capabilities, and then it becomes confusing for the other uses | 14:51 |
| melwitt | as evidenced by the ML thread, I don't even understand if the ComputeCapabilitiesFilter can be replaced by trait usage | 14:52 |
| fried_rice | I'm not sure I understand either, but I would like to be able to see that discussion develop rather than being shut down prematurely by the "intended purpose" argument. | 14:54 |
| fried_rice | If there's some other reason, by all means let's bring it into the conversation. | 14:54 |
| fried_rice | As for the owner thing, I agree we agreed it would be a standard trait. And yet: https://review.openstack.org/#/c/602160/ | 14:55 |
| melwitt | yeah, I intend to arrive at a more simplified answer from someone like, is it "possible" and if so, how, and what does that look like, and if it's a bad UX, how can we get to something that is a good UX. I just have to keep asking | 14:56 |
| melwitt | ah, I see | 14:57 |
| *** e0ne has quit IRC | 14:58 | |
| melwitt | yeah, I'm conflicted about it because at least for me, reasoning about traits becomes more difficult when they mean things beyond "I have this ability" and can include "my name is this" "my service owner is this" etc | 14:59 |
| *** e0ne has joined #openstack-placement | 15:02 | |
| edleafe | I've stated my position, but what I'd really like is for jaypipes to weigh in on this, as the "capabilities, not state" design was his. | 15:04 |
| *** jaypipes is now known as leakypipes | 15:06 | |
| leakypipes | melwitt: I've been in openstack-ironic channel for last hour disucssing this... | 15:06 |
| melwitt | oh, heh. I'll take a look at that | 15:09 |
| melwitt | I have to go help pick up the tacoma, bbl | 15:11 |
| *** cdent has quit IRC | 15:18 | |
| leakypipes | melwitt: ComputeCapabilitiesFilter cannot be replaced by traits, no. Because key/value pairs != capabilities. That said, ComputeCapabilitiesFilter could probably be replaced with placement request filters, since the majority of the ComputeCapabilitiesFilter use case is this: find me a compute node that has cpu_model=XXX (or some similar key/value constraint). That is perfect for a host aggregate metadata lookup, which would return a set | 15:21 |
| leakypipes | of aggregate UUIDs that then would get passed on to the placement GET /allocation_candidates request as a member_of=in:<AGG_UUID_LIST> query. The problem with the ComputeCapabilitiesFilter is the weird edge cases where it's not just key/value lookups that they're doing but rather crazy-ass things like "find me a compute host that can do RAID5, but where RAID5 is actually activated on the baremetal host and where the firmware of the node | 15:21 |
| leakypipes | is greater than 1.24.51912" | 15:21 |
| leakypipes | melwitt: so, in short, no, the ComputeCapabilitiesFilter probably will never go away, but we can do more efficient placement request filters for a large portion of the use cases it currently meets. | 15:23 |
| mriedem | leakypipes: doesn't that mean we shouldn't deprecate that filter then? | 15:33 |
| *** helenafm has quit IRC | 15:37 | |
| *** s10 has quit IRC | 15:41 | |
| leakypipes | mriedem: yes. pretty sure I never said to deprecate it. | 15:44 |
| leakypipes | mriedem: not sure why deprecation was even brought up. | 15:44 |
| leakypipes | mriedem: I just am resigned to the fact that we're going to continue to have crazypants filters for crazy edge cases in the filter scheduler until the end of time. | 15:45 |
| leakypipes | mriedem: the only thing I've ever said is to move the things we *can* to placement request filters when the use case can be more efficiently handled that way. | 15:45 |
| mriedem | sure, i didn't mean to say it was your idea, it know it wasn't. | 15:46 |
| mriedem | i also think, given enough time, people are going to complain about having to decorate providers with 100 individual traits | 15:47 |
| mriedem | it'll be a different type of mgmt nightmare | 15:47 |
| *** ttsiouts has quit IRC | 15:55 | |
| *** cdent has joined #openstack-placement | 15:58 | |
| fried_rice | cdent: Hey, gotta run out, but real quick before I do: We're going to have to figure out what to do with the service-types-authority record for placement wrt extraction. What and when. | 16:14 |
| fried_rice | cdent: Talked to mordred a bit about it yesterday: http://eavesdrop.openstack.org/irclogs/%23openstack-dev/%23openstack-dev.2018-09-27.log.html#t2018-09-27T21:58:52 | 16:14 |
| cdent | noted | 16:14 |
| fried_rice | okay, bbl. | 16:14 |
| *** fried_rice is now known as fried_rolls | 16:15 | |
| *** mriedem is now known as mriedem_lunch | 16:21 | |
| *** Nel1x has quit IRC | 17:04 | |
| *** mnaser has joined #openstack-placement | 17:09 | |
| melwitt | leakypipes: thanks for the clarity on that. I understand a lot more now | 17:13 |
| *** tssurya has quit IRC | 17:31 | |
| *** e0ne has quit IRC | 17:44 | |
| *** mriedem_lunch has quit IRC | 17:52 | |
| *** mriedem has joined #openstack-placement | 17:54 | |
| *** alex_xu has quit IRC | 17:56 | |
| *** alex_xu has joined #openstack-placement | 18:01 | |
| *** e0ne has joined #openstack-placement | 18:02 | |
| *** e0ne has quit IRC | 18:05 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: Remove placement.db.base https://review.openstack.org/606154 | 18:10 |
| *** e0ne has joined #openstack-placement | 18:25 | |
| *** mriedem has quit IRC | 18:44 | |
| *** e0ne has quit IRC | 18:50 | |
| *** tssurya has joined #openstack-placement | 19:22 | |
| *** e0ne has joined #openstack-placement | 19:29 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: DNM: Experimenting with integration gabbits https://review.openstack.org/601614 | 19:31 |
| openstackgerrit | Chris Dent proposed openstack/placement master: WIP: Add placeload to integration test https://review.openstack.org/602484 | 19:31 |
| * cdent is going against the current | 19:31 | |
| *** e0ne has quit IRC | 19:34 | |
| *** mriedem has joined #openstack-placement | 19:50 | |
| *** openstackgerrit has quit IRC | 20:07 | |
| *** PapaOurs is now known as bauzas | 21:05 | |
| *** tssurya has quit IRC | 21:26 | |
| *** mriedem has quit IRC | 21:37 | |
| *** openstackgerrit has joined #openstack-placement | 22:13 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: DNM: Experimenting with integration gabbits https://review.openstack.org/601614 | 22:13 |
| openstackgerrit | Chris Dent proposed openstack/placement master: WIP: Add placeload to integration test https://review.openstack.org/602484 | 22:13 |
| *** fried_rolls is now known as efried | 22:20 | |
| *** cdent has quit IRC | 23:22 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!