| *** tetsuro has joined #openstack-placement | 00:02 | |
| *** ttsiouts has quit IRC | 00:11 | |
| openstackgerrit | Eric Fried proposed openstack/placement master: Remove pep8 whitespace ignores https://review.openstack.org/640554 | 00:18 |
|---|---|---|
| openstackgerrit | Eric Fried proposed openstack/placement master: Nit collector https://review.openstack.org/640555 | 00:18 |
| openstackgerrit | Brin Zhang proposed openstack/nova-specs master: Support for changing deleted_on_termination after boot https://review.openstack.org/580336 | 00:39 |
| openstackgerrit | Brin Zhang proposed openstack/nova-specs master: Support delete_on_termination in volume attach api https://review.openstack.org/612949 | 00:40 |
| efried | cdent tetsturo jaypipes re+A https://review.openstack.org/#/c/640554/ (rebased) if you please | 00:47 |
| *** mriedem has quit IRC | 01:00 | |
| *** tetsuro has quit IRC | 01:32 | |
| *** tetsuro has joined #openstack-placement | 05:04 | |
| *** bhagyashris_ has joined #openstack-placement | 05:12 | |
| *** bhagyashris_ has quit IRC | 05:43 | |
| *** rubasov has quit IRC | 05:53 | |
| *** e0ne has joined #openstack-placement | 06:31 | |
| openstackgerrit | Merged openstack/placement master: Remove pep8 whitespace ignores https://review.openstack.org/640554 | 06:42 |
| openstackgerrit | Abhishek Kekane proposed openstack/nova-specs master: Support multiple backend of Glance https://review.openstack.org/641210 | 06:52 |
| openstackgerrit | Tetsuro Nakamura proposed openstack/placement master: Use oslo_utils.excutils for reraise exception https://review.openstack.org/641256 | 07:12 |
| openstackgerrit | Abhishek Kekane proposed openstack/nova-specs master: Support multiple backend of Glance https://review.openstack.org/641210 | 07:16 |
| *** bhagyashris has joined #openstack-placement | 07:39 | |
| *** e0ne has quit IRC | 07:47 | |
| *** e0ne has joined #openstack-placement | 07:54 | |
| *** rubasov has joined #openstack-placement | 07:56 | |
| *** e0ne has quit IRC | 08:16 | |
| *** helenafm has joined #openstack-placement | 08:24 | |
| *** takashin has left #openstack-placement | 08:30 | |
| *** tssurya has joined #openstack-placement | 08:33 | |
| *** ttsiouts has joined #openstack-placement | 09:08 | |
| gryf | c | 09:09 |
| *** ttsiouts has quit IRC | 09:13 | |
| *** tetsuro has quit IRC | 09:35 | |
| *** bhagyashris has quit IRC | 09:54 | |
| *** e0ne has joined #openstack-placement | 10:35 | |
| *** cdent has joined #openstack-placement | 10:49 | |
| *** rubasov has quit IRC | 10:51 | |
| *** helenafm_ has joined #openstack-placement | 11:04 | |
| *** helenafm has quit IRC | 11:06 | |
| *** ttsiouts has joined #openstack-placement | 12:16 | |
| sean-k-mooney | cdent: i do not have a vote in the placement ptl elections but you would get my +1 if i did. | 12:31 |
| *** tetsuro has joined #openstack-placement | 12:31 | |
| cdent | sean-k-mooney: there was some talk of using nova commits to gather eligible voters, because history is short. thank you. | 12:31 |
| cdent | (I'll check on that) | 12:31 |
| sean-k-mooney | im going to try and attend some of the placement form/ptg sessions but if i can help out with "get the house in order" it would be a good way for me to learn more about how placement is implemented | 12:32 |
| *** rubasov has joined #openstack-placement | 12:32 | |
| cdent | thank you | 12:33 |
| sean-k-mooney | cdent: well i dont have enough/any real commits to placement code intree or extracted | 12:33 |
| cdent | run_once? | 12:33 |
| sean-k-mooney | ya that was the only one | 12:33 |
| sean-k-mooney | most of my contibutions to placement in the past have been "how could we use it in nova" and "what is missing to do x" | 12:34 |
| cdent | I think most of the work for placement over the next several months is going to be work in nova; getting shared and nested really going | 12:36 |
| sean-k-mooney | ya im kind of interested to work on modeling sriov in placement but i need to run that by our team next week | 12:37 |
| sean-k-mooney | i need to figure out if i will be looking at cyborg stuff next cycle or not and that will determin how much time i have to work on other features. | 12:38 |
| cdent | I'm also really curious to get ironic using it directly, but I don't know that anyone is going to have time for that | 12:38 |
| sean-k-mooney | as in for ironic standalone | 12:39 |
| cdent | yeah | 12:39 |
| sean-k-mooney | ya that would make sense. alther do they have a standalone schduler. i taught you had to specify the host when it was used standalone | 12:40 |
| cdent | I'm not clear on what the current situation is. Just seems an area worth investigating. | 12:40 |
| sean-k-mooney | i guess if they modeled everything in placmente you coudl still use placement to find the set of candiates and impemetn your own schduler then call ironic direcly to deploy | 12:41 |
| sean-k-mooney | so if you were driving ironic with OSM or ONAP it could be interesting | 12:42 |
| cdent | yeah, my etcd-compute experiments made it pretty clear that "scheduler" can be a very lightweight api for "make it go" | 12:42 |
| sean-k-mooney | ya | 12:42 |
| sean-k-mooney | you know i was talking to stephen a while ago about how if i had infinit time and energy i would love to replace the current libvirt driver with either an agentless model or distributed state model(like you etcd-compute) to reduce the need for rabbitmq | 12:45 |
| sean-k-mooney | i still think we would likely need to keep rabbit for db operations like looking up flavors but we could also cache those in etcd so i think 90% or more of the usage of rabbitmq could be eliminated. | 12:47 |
| sean-k-mooney | did you fully remove it in your etcd-compute driver | 12:48 |
| sean-k-mooney | oh https://github.com/cdent/etcd-compute does not actully plug into nova | 12:51 |
| cdent | when I started the experiment I had no intention of it actually being able to launch real vms, but then that part proved to be "easy" | 12:58 |
| cdent | it's was done more as an example of "small things with placement" than "let's change nova" | 12:59 |
| *** ttsiouts has quit IRC | 13:17 | |
| *** ttsiouts has joined #openstack-placement | 13:18 | |
| *** ttsiouts has quit IRC | 13:23 | |
| *** ttsiouts has joined #openstack-placement | 13:28 | |
| efried | sean-k-mooney: If you're interested in cyborg, there's some discussion going on right now around the design. | 13:30 |
| efried | sean-k-mooney: Sundar implemented a PoC of what is essentially the design we agreed upon in Denver last time. | 13:30 |
| sean-k-mooney | efried: yes i am interested and plan deploy the poc before denver | 13:31 |
| efried | sean-k-mooney: Series starting at https://review.openstack.org/#/c/631242/ | 13:31 |
| sean-k-mooney | thats the nova side right and he has a cyborg branch too | 13:32 |
| sean-k-mooney | with the same topic? | 13:32 |
| efried | cyborg side is here: https://review.openstack.org/#/c/626420/ | 13:32 |
| sean-k-mooney | ok can we at least put them on the same gerrit topic so that they are easy to find? | 13:32 |
| efried | We're still trying to convince Jay this is the right approach. He wants to see custom resource classes and no dynamic programming (unless orchestrated externally) | 13:33 |
| efried | sean-k-mooney: They're not the same topic. One is a pilot branch for the cyborg API itself; the other is consuming that API in nova. They'll have different blueprints, specs, etc. | 13:33 |
| efried | The cyborg series is linked from the commit message of the patch that starts using it. | 13:34 |
| sean-k-mooney | well i need to look at what was actully implemented in the piolt before i can comment | 13:34 |
| *** mriedem has joined #openstack-placement | 13:34 | |
| efried | sure. | 13:35 |
| sean-k-mooney | efried: i personally thing that programing a device should not change its representation in placment | 13:35 |
| efried | sean-k-mooney: There's an etherpad with an overview of the design (and a bit of internal discussion): https://etherpad.openstack.org/p/nova-cyborg-design-notes | 13:35 |
| jaypipes | efried: morning. | 13:36 |
| jaypipes | sean-k-mooney: you too :) | 13:36 |
| efried | sean-k-mooney: In the PoC, programming the device changes its traits, not its inventory. | 13:36 |
| efried | Hi jaypipes | 13:36 |
| sean-k-mooney | jaypipes: o/ | 13:36 |
| sean-k-mooney | efried: it feels at first glance that cyborge by changing its traits is just using placement as metadata storage | 13:37 |
| jaypipes | sean-k-mooney: bingo. | 13:38 |
| sean-k-mooney | but it has its uses in that i may avoid having to reprogram a deivce if you trust that a guest cant do somthing malious to it | 13:38 |
| jaypipes | sean-k-mooney: course, we've had this conversation about ten times already at each of the past PTGs over the last 1.5 years at least. | 13:38 |
| sean-k-mooney | e.g. you could reuse the device for another requset without needing to reprogram | 13:38 |
| sean-k-mooney | haha ya | 13:38 |
| efried | jaypipes: Yes, we have, and we have always come back to the model proposed in the PoC. | 13:39 |
| jaypipes | OK, since the risk of derailment is high, I am going back to getting reviews done this morning, sorry guys... | 13:39 |
| sean-k-mooney | didnt we lay out a 3 phase plan in both denver and dublin | 13:39 |
| efried | changing traits isn't perfect, but custom resource classes are the greater evil | 13:40 |
| efried | sean-k-mooney: Yes, we did. | 13:40 |
| sean-k-mooney | jaypipes: no worries i too need to jump to a neuton patch that i need to rethink | 13:40 |
| sean-k-mooney | efried: well the custom resouce class would be fixed | 13:40 |
| efried | no | 13:40 |
| sean-k-mooney | e.g. it also would not change based in the programmed image | 13:41 |
| efried | cyborg would be changing the inventory every time it reprogrammed. | 13:41 |
| sean-k-mooney | it would be based on the device | 13:41 |
| efried | no | 13:41 |
| efried | cyborg would be changing the inventory every time it reprogrammed. | 13:41 |
| sean-k-mooney | it does not have to however | 13:41 |
| jaypipes | efried: what does the kubernetes scheduler team think about FPGAs and GPU resources? | 13:41 |
| jaypipes | efried: are you familiar with the PCI management proposals in k8s land for tracking custom hardware? | 13:42 |
| efried | jaypipes: no idea. I would only be guessing if I said a container wants its accelerators to be programmed by the time the container is stood up. | 13:42 |
| sean-k-mooney | jaypipes: they have device pluggins | 13:42 |
| jaypipes | right. everything is a plugin. no standards, no idea how to do standardized resource tracking. it's complete wild west. | 13:43 |
| sean-k-mooney | you can freeley add new ones for your custom device without needeing to change the core | 13:43 |
| sean-k-mooney | ya | 13:43 |
| efried | right, was gonna ask if k8s people use placement? | 13:43 |
| sean-k-mooney | pros and cons | 13:43 |
| jaypipes | also no ability to share resources like this between tenants with any isolation. | 13:43 |
| cdent | (not yet!) | 13:43 |
| sean-k-mooney | they dont need to have this conversation but each deployment is totally different | 13:43 |
| jaypipes | I'm actually going to get back to reviews, but my point is that k8s land has done the same wild-west approach that the original nova pci manager made and it's going to bite them as hard long-term. | 13:45 |
| jaypipes | see the proposals from nvidia and intel over there. | 13:45 |
| jaypipes | same stuff, different hype. | 13:45 |
| efried | The proposed use of placement makes this anything but a wild west. It is tightly controlled and structured. | 13:50 |
| *** tetsuro has quit IRC | 13:50 | |
| jaypipes | we'll have to agree to disagree. | 13:51 |
| *** tetsuro has joined #openstack-placement | 13:51 | |
| efried | jaypipes: I would be happy to do so if we could move the cyborg-in-nova effort forward with the consensus-but-one that we've reached every time we've talked about this. | 13:52 |
| jaypipes | efried: you can just say "dude, you're being an asshole blocker for this" :) | 13:55 |
| jaypipes | but at the end of the day, I have as much right as anyone to point out what I think is a bad design and a bad misuse of the placement API. | 13:55 |
| efried | jaypipes: Absolutely. And your viewpoint is greatly respected. But at the end of the day, we have to be able to make progress even if there's a conflicting respected viewpoint on the table. | 13:57 |
| jaypipes | understood completely. | 13:57 |
| jaypipes | of course, I might disagree this is progress, but OK :) | 13:57 |
| cdent | If it's true that we agreed to this plan at the last PTG, unless there are new reasons for disagreement, can we move forward, even if it is somewhat icky? | 13:58 |
| efried | ...that's what I would like to see. IMO Sundar has done an excellent job of crisping up the details and proving them in code. AFAIK nothing new has come up - no surprises or revelations as this was being implemented. | 14:06 |
| jaypipes | efried: so, the only thing I had asked of the Cyborg team at the last 2 PTGs was the following: working (shown in a test) code that plumbs a VM to an FPGA programmed region. Has that been done? | 14:07 |
| efried | jaypipes: In a test that you can see in gerrit? I don't think so. But the PoC series works in a devstack. | 14:08 |
| efried | the device is currently faked. | 14:09 |
| efried | but that part - putting the pci address in the xml - isn't new | 14:09 |
| jaypipes | efried: which is the ONLY thing I'd asked for in the last 2 years. show some code that functionally connects a VM to the FPGA. | 14:09 |
| efried | yup, that's what the PoC patches do. | 14:10 |
| jaypipes | it's a faked device. | 14:10 |
| efried | are we expecting a real device in a CI? | 14:11 |
| jaypipes | we better be, otherwise we'll end up in NUMA live migration land. | 14:11 |
| jaypipes | and we all know how much fun that's been. | 14:11 |
| efried | standing up a CI with real hardware for a PoC seems like kind of an unreasonable ask. | 14:11 |
| jaypipes | same for PCI devicemanager land. | 14:11 |
| jaypipes | efried: not CI, sorry. | 14:11 |
| efried | for the real thing, fair | 14:11 |
| jaypipes | efried: I'd be happy with just a demo of real hardware. | 14:12 |
| jaypipes | but I've never seen anything like that. instead, the focus has been on yet another REST API service. | 14:12 |
| jaypipes | sorry if I'm annoyed about that | 14:12 |
| jaypipes | but at least I've been consistently annoyed for 2 years on the same exact point. | 14:12 |
| efried | Okay. I'll go ask Sundar to deploy his PoC against real hardware and put together some kind of demo... video or something. | 14:12 |
| jaypipes | in any case, it doesn't sound like anything is going to change, so I'll relent. | 14:13 |
| efried | Thanks. The compromise is appreciated. | 14:13 |
| efried | (no sarcasm, in case that came across wrong) | 14:14 |
| mriedem | we should uh, maybe, not yell at people https://docs.openstack.org/placement/latest/placement-api-microversion-history.html#granular-resource-requests-to-get-allocation-candidates | 14:14 |
| jaypipes | heh. compromise... | 14:14 |
| *** jaypipes has quit IRC | 14:14 | |
| cdent | mriedem: but it is really important! | 14:15 |
| *** jaypipes has joined #openstack-placement | 14:15 | |
| cdent | mriedem: you fixing it, or shall i? | 14:16 |
| mriedem | i'm doing it | 14:16 |
| efried | mriedem: That's a ``this`` role | 14:16 |
| mriedem | i know | 14:16 |
| mriedem | in a title | 14:16 |
| efried | I guess in other titles we're just not using a role? | 14:16 |
| efried | I don't know that that's better. But okay. | 14:16 |
| mriedem | you don't really need to in a title | 14:16 |
| mriedem | b/c it's a title | 14:16 |
| efried | stephenfin: do you have a way we can make a code-ish chunk look code-ish in a title? | 14:17 |
| efried | (without shouting) | 14:18 |
| stephenfin | efried: What do you mean? | 14:19 |
| stephenfin | ``code`` | 14:19 |
| stephenfin | -------- | 14:19 |
| openstackgerrit | Matt Riedemann proposed openstack/placement master: Stop yelling the 1.25 microversion history at people https://review.openstack.org/641381 | 14:19 |
| mriedem | stephenfin: ^ | 14:19 |
| mriedem | https://docs.openstack.org/placement/latest/placement-api-microversion-history.html#granular-resource-requests-to-get-allocation-candidates | 14:19 |
| stephenfin | mriedem: Well that's annoying. That's a bug in openstack-docs-theme. Happy to fix if you fancy creating a quick bug at some point | 14:20 |
| efried | ah, that would be a better solution imo. | 14:21 |
| mriedem | fix how? just ignore `` `` in titles? | 14:21 |
| efried | not ignore. reduce font size and match color? | 14:21 |
| stephenfin | You could make it monospace but not bold and black | 14:21 |
| efried | yeah, or that | 14:21 |
| mriedem | https://bugs.launchpad.net/openstack-doc-tools/+bug/1818840 | 14:23 |
| openstack | Launchpad bug 1818840 in openstack-doc-tools "``this`` in titles is extra loud" [Undecided,New] | 14:24 |
| openstackgerrit | Matt Riedemann proposed openstack/placement master: Stop yelling the 1.11 and 1.25 microversion history at people https://review.openstack.org/641381 | 14:25 |
| openstackgerrit | Tetsuro Nakamura proposed openstack/placement master: Prepare for negative member_of queryparam https://review.openstack.org/641391 | 14:31 |
| openstackgerrit | Tetsuro Nakamura proposed openstack/placement master: Prepare for negative member queryparam 2 https://review.openstack.org/641392 | 14:31 |
| openstackgerrit | Tetsuro Nakamura proposed openstack/placement master: Negative member_of query with microversion 1.32 https://review.openstack.org/641393 | 14:31 |
| cdent | tetsuro++ | 14:31 |
| tetsuro | I'm going to bed since it's too late now. | 14:32 |
| tetsuro | bye | 14:34 |
| *** tetsuro has quit IRC | 14:34 | |
| cdent | how do people feel about visual indent? | 14:38 |
| efried | define, please | 14:39 |
| edleafe | Hate it | 14:39 |
| efried | you considering adding a pep ignore and/or hacking rule? | 14:39 |
| efried | is this the one where the start of the continuation line is lined up with the opening paren of the first line? | 14:39 |
| cdent | https://github.com/openstack/placement/blob/master/placement/tests/unit/objects/test_resource_provider.py#L162-L164 | 14:40 |
| cdent | edleafe: I also hate it. very bad for diffs | 14:40 |
| cdent | openstacksdk forbids it, if I remember right | 14:41 |
| cdent | at this point, I'm thinking, not proposing | 14:41 |
| efried | My opinion: sometimes it's appropriate, sometimes it's not. I don't agree with a rule that enforces "*never* do it". But wouldn't block such a rule if majority agreed on it. | 14:43 |
| efried | for instance, the example you linked doesn't offend me at all. | 14:43 |
| efried | or the one below it. | 14:44 |
| edleafe | One fundamental concept with Python is that indentation is semantic. Things at the same nesting level should have the same indentation. | 14:45 |
| openstackgerrit | Eric Fried proposed openstack/placement master: DNM: Use ``code`` role in api-ref titles https://review.openstack.org/641404 | 14:53 |
| efried | mriedem, stephenfin: there's that ^ for when the docs theme fix is in. | 14:54 |
| stephenfin | efried: Sweet, fix is here https://review.openstack.org/641408 | 15:01 |
| efried | stephenfin: nice. Does depends-on work for docstheme? (don't see why it wouldn't) | 15:01 |
| stephenfin | I would imagine so, yup | 15:02 |
| efried | trying... | 15:02 |
| openstackgerrit | Eric Fried proposed openstack/placement master: DNM: Use ``code`` role in api-ref titles https://review.openstack.org/641404 | 15:03 |
| openstackgerrit | Stephen Finucane proposed openstack/placement master: Group API versions by release https://review.openstack.org/641422 | 15:46 |
| stephenfin | efried: You'd also like that ^^^ coupled with this https://review.openstack.org/641423, I suspect | 15:46 |
| efried | ... | 15:47 |
| efried | did you mean to make the former dependent on the latter? | 15:48 |
| stephenfin | I probably should have, yeah, but it's not compulsory | 15:48 |
| stephenfin | It still looks better (IMO) even without the openstackdocstheme chang | 15:49 |
| stephenfin | *change | 15:49 |
| *** rubasov has quit IRC | 16:04 | |
| *** ttsiouts has quit IRC | 16:08 | |
| *** ttsiouts has joined #openstack-placement | 16:09 | |
| openstackgerrit | Chris Dent proposed openstack/placement master: Move ResourceClass and ResourceClassList to own module https://review.openstack.org/641432 | 16:10 |
| openstackgerrit | Chris Dent proposed openstack/placement master: Remove the ResourceClassList class https://review.openstack.org/641433 | 16:10 |
| * cdent switches back to osc-placement | 16:10 | |
| *** ttsiouts has quit IRC | 16:11 | |
| *** ttsiouts has joined #openstack-placement | 16:11 | |
| *** Sundar has joined #openstack-placement | 16:15 | |
| *** tssurya has quit IRC | 16:18 | |
| efried | cdent: https://review.openstack.org/#/c/640555/ is now first-in-queue, and contains valid changes. We could keep it around for a while as a collector, or we could just merge it and move on, or we could split it into two changes since it's really two changes (though a whole patch for a one-character typo in a comment is the kind of thing we usually don't accept from anyone), or... something else? | 16:33 |
| cdent | efried: I think you should tune up the commit message and we should simply merge it | 16:34 |
| cdent | it's no longer really a "collector" when it is first | 16:34 |
| efried | okay, stand by. | 16:36 |
| openstackgerrit | Eric Fried proposed openstack/placement master: Add explicit short-circuit in get_all_by_consumer_id https://review.openstack.org/640555 | 16:38 |
| efried | cdent: ^ | 16:38 |
| cdent | aye | 16:39 |
| openstackgerrit | Merged openstack/placement master: Stop yelling the 1.11 and 1.25 microversion history at people https://review.openstack.org/641381 | 16:59 |
| *** rubasov has joined #openstack-placement | 17:17 | |
| *** ttsiouts has quit IRC | 17:24 | |
| *** ttsiouts has joined #openstack-placement | 17:25 | |
| *** ttsiouts has quit IRC | 17:29 | |
| *** e0ne has quit IRC | 17:34 | |
| *** helenafm_ has quit IRC | 17:34 | |
| openstackgerrit | Chris Dent proposed openstack/osc-placement master: Add support for 1.19 microversion https://review.openstack.org/641094 | 18:30 |
| openstackgerrit | Chris Dent proposed openstack/osc-placement master: Microversion 1.21 support https://review.openstack.org/641123 | 18:30 |
| *** Sundar has quit IRC | 18:31 | |
| *** jaypipes has quit IRC | 18:44 | |
| *** Sundar has joined #openstack-placement | 19:29 | |
| *** cdent_ has joined #openstack-placement | 19:59 | |
| *** cdent has quit IRC | 20:02 | |
| *** cdent_ is now known as cdent | 20:02 | |
| *** e0ne has joined #openstack-placement | 20:03 | |
| *** Sundar has quit IRC | 20:19 | |
| *** e0ne has quit IRC | 20:21 | |
| *** cdent has quit IRC | 20:42 | |
| *** e0ne has joined #openstack-placement | 21:13 | |
| *** takashin has joined #openstack-placement | 21:50 | |
| *** e0ne has quit IRC | 21:58 | |
| *** e0ne has joined #openstack-placement | 22:01 | |
| *** e0ne has quit IRC | 22:01 | |
| openstackgerrit | Eric Fried proposed openstack/placement master: Get rid of backslash continuations in code https://review.openstack.org/641497 | 22:19 |
| openstackgerrit | Chris Dent proposed openstack/placement master: Move ResourceClass and ResourceClassList to own module https://review.openstack.org/641432 | 22:20 |
| openstackgerrit | Chris Dent proposed openstack/placement master: Remove the ResourceClassList class https://review.openstack.org/641433 | 22:20 |
| openstackgerrit | Chris Dent proposed openstack/placement master: Move Trait and TraitList to own module https://review.openstack.org/641500 | 22:20 |
| openstackgerrit | Chris Dent proposed openstack/placement master: Make Trait._get_by_name_from_db a reader https://review.openstack.org/641501 | 22:20 |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!