| *** mriedem_afk is now known as mriedem | 00:12 | |
| *** irclogbot_0 has joined #openstack-placement | 00:17 | |
| *** mriedem has quit IRC | 03:29 | |
| openstackgerrit | Boxiang Zhu proposed openstack/nova-specs master: Scheduler filters evaluated even forced host https://review.openstack.org/645458 | 03:30 |
|---|---|---|
| openstackgerrit | Boxiang Zhu proposed openstack/nova-specs master: Scheduler filters evaluated even forced host https://review.openstack.org/645458 | 03:46 |
| *** takashin has joined #openstack-placement | 06:10 | |
| *** david-lyle has joined #openstack-placement | 06:27 | |
| *** dklyle has quit IRC | 06:27 | |
| *** david-lyle has quit IRC | 06:29 | |
| *** dklyle has joined #openstack-placement | 06:29 | |
| *** dims has quit IRC | 06:39 | |
| *** dims has joined #openstack-placement | 06:41 | |
| *** takashin has left #openstack-placement | 07:30 | |
| *** ttsiouts has joined #openstack-placement | 08:13 | |
| *** helenaAM has joined #openstack-placement | 08:23 | |
| *** ttsiouts has quit IRC | 08:26 | |
| *** ttsiouts has joined #openstack-placement | 08:27 | |
| *** ttsiouts has quit IRC | 08:31 | |
| *** ttsiouts has joined #openstack-placement | 08:33 | |
| *** ttsiouts has quit IRC | 08:45 | |
| *** ttsiouts has joined #openstack-placement | 08:46 | |
| *** ttsiouts has quit IRC | 08:50 | |
| *** ttsiouts has joined #openstack-placement | 09:04 | |
| *** tssurya has joined #openstack-placement | 09:04 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: Initial structure for in-tree specs https://review.openstack.org/645195 | 09:07 |
| *** ttsiouts has quit IRC | 09:21 | |
| *** ttsiouts has joined #openstack-placement | 09:22 | |
| *** ttsiouts has quit IRC | 09:26 | |
| *** ttsiouts has joined #openstack-placement | 09:44 | |
| *** cdent has joined #openstack-placement | 09:51 | |
| cdent | pupdate: https://anticdent.org/placement-update-19-11.html | 10:40 |
| *** ttsiouts has quit IRC | 10:42 | |
| *** ttsiouts has joined #openstack-placement | 10:43 | |
| *** ttsiouts_ has joined #openstack-placement | 10:48 | |
| *** ttsiouts has quit IRC | 10:48 | |
| *** ttsiouts_ has quit IRC | 11:54 | |
| *** ttsiouts has joined #openstack-placement | 11:55 | |
| *** ttsiouts has quit IRC | 11:59 | |
| *** ttsiouts has joined #openstack-placement | 12:09 | |
| *** altlogbot_2 has quit IRC | 13:01 | |
| *** irclogbot_0 has quit IRC | 13:01 | |
| *** altlogbot_1 has joined #openstack-placement | 13:02 | |
| *** irclogbot_1 has joined #openstack-placement | 13:02 | |
| *** dklyle has quit IRC | 13:07 | |
| openstackgerrit | Surya Seetharaman proposed openstack/nova-specs master: Support adding the reason behind a server lock https://review.openstack.org/638629 | 13:11 |
| *** dklyle has joined #openstack-placement | 13:12 | |
| *** ttsiouts has quit IRC | 13:20 | |
| *** ttsiouts has joined #openstack-placement | 13:21 | |
| *** ttsiouts has quit IRC | 13:26 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: Initial structure for in-tree specs https://review.openstack.org/645195 | 13:33 |
| *** mriedem has joined #openstack-placement | 13:51 | |
| *** efried is now known as fried_rice | 13:58 | |
| fried_rice | cdent: This has come up in my head several times while thinking about what the next enhancements to placement might "need to" be: | 14:19 |
| cdent | oh hi | 14:20 |
| fried_rice | perhaps it's time we reopen the idea of a GET /a_c that accepts a JSON request payload. | 14:20 |
| fried_rice | basically, whatever we add to GET /a_c *next* is going to be super awkward to fit into querystring-ness. | 14:20 |
| cdent | it does seem like something that will happen someday and we can do it when someday comes | 14:21 |
| cdent | but until then, why worry? | 14:22 |
| cdent | the flip side of that is: the query string handling is a useful constraint that prevents us from writing war and peace when making a GET /a_c query, which helps keep that API better | 14:23 |
| cdent | you're right that we will probably find a reason, eventually | 14:23 |
| cdent | but your statement was a classic YAGNI | 14:23 |
| cdent | (especially given how much catch up with need to do from services that use placement) | 14:23 |
| cdent | which makes it sounds like I'm being all "no, don't think about that". I do think it is worth thinking about, just not doing (yet) | 14:24 |
| edleafe | "We'll burn that bridge when we come to it" | 14:26 |
| cdent | burn all the bridges | 14:27 |
| fried_rice | I agree we shouldn't just do it with the syntax we have today | 14:27 |
| fried_rice | what I'm saying is, everything that's "coming next" that I've pondered (e.g. subtree affinity) I can't think of a way to slot into querystring without it being really horrible. | 14:27 |
| cdent | in that vein I wish we could get some input from edge folk who talk about wanting partitioned placement. it's a thing on the radar too, but needs real info | 14:28 |
| fried_rice | whereas when I start contemplating what it would look like in a JSON payload, I'm like, "oh, yeah, that makes sense and is pretty" | 14:28 |
| cdent | are you familiar with the phrase: "make it work then make it right" and its variants | 14:28 |
| fried_rice | more or less. | 14:29 |
| edleafe | I would need to see the request that would be too cumbersome for a query string first | 14:29 |
| cdent | while you may not have had ideas on how to do it pretty in a query string, if you do it ugly, other people (and you too) can look at that concrete thing and say either "you could do that like X" or, conclusisvely and concretely "we need json for this" | 14:29 |
| cdent | yeah edleafe said it shorter | 14:29 |
| openstackgerrit | Surya Seetharaman proposed openstack/nova-specs master: Support adding the reason behind a server lock https://review.openstack.org/638629 | 14:30 |
| fried_rice | edleafe: Okay, for this particular example, I'm trying to express: | 14:35 |
| fried_rice | I want providers from group 3 (e.g. NUMA VCPU+MEMORY_MB) and group 4 (e.g. VGPU affined to same NUMA node) to come from the same subtree, and that sub-root is indicated $somehow, but | 14:35 |
| fried_rice | a) we don't want to hardcode it to something like "always one level down from the root"; and | 14:35 |
| fried_rice | b) we want it to be generic enough so that, if there are N subtrees that could satisfy, we don't have to build an N-component OR, especially since N could be different from one host to the next. | 14:35 |
| fried_rice | Initial thought is something like: | 14:37 |
| fried_rice | Tag the subtree roots with a trait, e.g. NUMA_SUBTREE_STARTS_HERE. Importantly, the trait is the *same* for each numa subtree provider. (This gets us (b).) | 14:37 |
| fried_rice | Provide a syntax that says, "the resources from group numbers [...] should be satisfied by providers under a common NUMA_SUBTREE_STARTS_HERE" | 14:37 |
| edleafe | fried_rice: I'm not able to visualize that request, so I can't suggest how it should be done. I would need to sit down and study the NUMA arrangements and corresponding trees to see what would work best | 14:37 |
| fried_rice | in querystring format, it would be something like: | 14:38 |
| fried_rice | common_subtree_required=NUMA_SUBTREE_STARTS_HERE&common_subtree_request_groups=3,4 | 14:38 |
| fried_rice | ...assuming we only want to do it once per query. If more than once, we need to somehow number the new keys. | 14:38 |
| fried_rice | it also means promoting the request group numbers to first-class citizens in some sense. | 14:38 |
| fried_rice | which we've kind of started talking about doing to return the "which providers correspond to which request groups" metadata. | 14:39 |
| fried_rice | ...whereas if we had a JSON request payload, we could do the grouping simply by wrapping the $trait(s) and $request_groups in a dict. Which would allow us to do multiple without awkward numbering, etc. | 14:41 |
| fried_rice | thinking about that ^ makes, like, my chest relax and I start breathing easier. | 14:41 |
| cdent | have you considered that the difficulty you are having with your chest is an important signal, a true fear, which should not be ignored? | 14:41 |
| cdent | if the thing you are trying to do is making you sweat | 14:42 |
| cdent | maybe it is wrong | 14:42 |
| cdent | (is only a bit serious) | 14:42 |
| edleafe | thinking about any of this makes my head spin. It's difficult for me to conceive when a) I'm distracted by other work and b) I don't have an image in my head to work with | 14:42 |
| fried_rice | Sure I've considered that cdent. | 14:42 |
| cdent | I think an email with some pretty drawings might help move the conversation along, IRC is unlikely to get anywhere | 14:43 |
| cdent | as I said in today's pupdate, I'm hoping we can have many of the conversations of the PTG nearly finished before we get to the PTG, otherwise none of them will finish | 14:43 |
| cdent | so we'll have to do email in advance | 14:43 |
| cdent | you should put this topic on the ptg etherpad if you haven't already) | 14:44 |
| fried_rice | cdent: yeah, this is the inter-provider affinity thing. I haven't fleshed it out yet. | 14:51 |
| cdent | I figured that, but was saying if you wanted to put the general topic of "json body on request" on the agenda, the option is there | 14:52 |
| *** Sundar has joined #openstack-placement | 14:53 | |
| fried_rice | ack | 14:55 |
| fried_rice | did that | 14:57 |
| cdent | I have to edit inter-provider affinity to intra-provider affinity in my head | 15:05 |
| cdent | because it's all in one parent | 15:05 |
| *** Sundar is now known as DarthNau | 15:14 | |
| *** altlogbot_1 has quit IRC | 15:21 | |
| *** altlogbot_1 has joined #openstack-placement | 15:26 | |
| *** irclogbot_1 has quit IRC | 15:30 | |
| *** irclogbot_2 has joined #openstack-placement | 15:32 | |
| *** irclogbot_2 has quit IRC | 15:36 | |
| *** irclogbot_1 has joined #openstack-placement | 15:38 | |
| fried_rice | intra-tree, not intra-provider | 15:41 |
| cdent | right, I'm not saying it's right, just that I do it as a mental tweak to distinguish from inter-tree | 15:48 |
| cdent | intra-tree is maybe the best term for the thing you're actually wanting to do (assuming you're pursuing the limited scope plan) | 15:49 |
| cdent | because otherwise every time you say inter-provider people will have the same questions and concerns that I keep raising | 15:50 |
| *** mriedem is now known as mriedem_afk | 16:07 | |
| *** helenaAM has quit IRC | 16:15 | |
| * cdent walkies | 16:19 | |
| fried_rice | Yeah, "subtree" is the best working title imo. Intra-tree is a superset, and the extras are weirdness I can't even come up with a scenario for. | 16:30 |
| *** fried_rice is now known as fried_rolls | 16:30 | |
| *** e0ne has joined #openstack-placement | 16:48 | |
| openstackgerrit | melanie witt proposed openstack/nova-specs master: Move Stein implemented specs https://review.openstack.org/645749 | 16:52 |
| *** e0ne has quit IRC | 16:59 | |
| cdent | I think I'm tapped out, unless somebody has something lsat minute for me to care about? | 17:15 |
| *** cdent has left #openstack-placement | 17:18 | |
| *** cdent has joined #openstack-placement | 17:24 | |
| *** cdent has left #openstack-placement | 17:24 | |
| *** tssurya has quit IRC | 17:30 | |
| openstackgerrit | Kashyap Chamarthy proposed openstack/nova-specs master: Add "CPU selection with hypervisor consideration" spec https://review.openstack.org/645814 | 17:33 |
| openstackgerrit | melanie witt proposed openstack/nova-specs master: Move Stein implemented specs https://review.openstack.org/645749 | 17:37 |
| openstackgerrit | melanie witt proposed openstack/nova-specs master: Move Stein implemented specs https://review.openstack.org/645749 | 17:40 |
| *** mriedem_afk is now known as mriedem | 17:55 | |
| openstackgerrit | Merged openstack/nova-specs master: Update Network Bandwidth resource provider spec https://review.openstack.org/644810 | 18:22 |
| openstackgerrit | melanie witt proposed openstack/nova-specs master: Move Stein implemented specs https://review.openstack.org/645749 | 18:45 |
| openstackgerrit | Merged openstack/nova-specs master: Update alloc-candidates-in-tree https://review.openstack.org/639033 | 19:01 |
| *** fried_rolls is now known as fried_rice | 19:07 | |
| openstackgerrit | melanie witt proposed openstack/nova-specs master: Move Stein implemented specs https://review.openstack.org/645749 | 19:12 |
| *** mriedem has quit IRC | 19:46 | |
| *** DarthNau has quit IRC | 19:46 | |
| *** mriedem has joined #openstack-placement | 19:47 | |
| *** e0ne has joined #openstack-placement | 21:54 | |
| *** e0ne has quit IRC | 22:36 | |
| openstackgerrit | melanie witt proposed openstack/nova-specs master: Move Stein implemented specs https://review.openstack.org/645749 | 22:37 |
| *** mriedem has quit IRC | 23:04 | |
| *** N3l1x has quit IRC | 23:34 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!