openstackgerrit | tianhui proposed openstack/nova master: Fix bug to filter_scheduler https://review.openstack.org/571998 | 02:20 |
---|---|---|
openstackgerrit | jichenjc proposed openstack/nova master: z/VM Driver: Initial change set of z/VM driver https://review.openstack.org/523387 | 02:46 |
openstackgerrit | jichenjc proposed openstack/nova master: z/VM Driver: Spawn and destroy function of z/VM driver https://review.openstack.org/527658 | 02:46 |
openstackgerrit | jichenjc proposed openstack/nova master: z/VM Driver: add snapshot function https://review.openstack.org/534240 | 02:46 |
openstackgerrit | jichenjc proposed openstack/nova master: z/VM Driver: add power actions https://review.openstack.org/543340 | 02:46 |
openstackgerrit | jichenjc proposed openstack/nova master: z/VM Driver: add get console output https://review.openstack.org/543344 | 02:47 |
openstackgerrit | tianhui proposed openstack/nova master: Fix bug to filter_scheduler https://review.openstack.org/571998 | 03:49 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Add rules column to instance_group_policy table. https://review.openstack.org/560832 | 03:50 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Add policy to InstanceGroup object and api models. https://review.openstack.org/563375 | 03:50 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Add policy field to ServerGroup notification object https://review.openstack.org/563401 | 03:50 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Change the anti-affinity Filter to adapt to new policy https://review.openstack.org/571166 | 03:50 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Adapt _validate_instance_group_policy to new policy model https://review.openstack.org/571465 | 03:50 |
openstackgerrit | Takashi NATSUME proposed openstack/nova master: Remove mox in unit/network/test_neutronv2.py (1) https://review.openstack.org/572018 | 05:06 |
openstackgerrit | Merged openstack/nova stable/pike: libvirt: handle DiskNotFound during update_available_resource https://review.openstack.org/571426 | 05:31 |
openstackgerrit | Merged openstack/nova stable/pike: libvirt: Skip fetching the virtual size of block devices https://review.openstack.org/571427 | 05:31 |
*** e0ne has joined #openstack-placement | 06:34 | |
openstackgerrit | jichenjc proposed openstack/nova master: Remove unused _disk_qcow2_to_raw https://review.openstack.org/572025 | 06:48 |
openstackgerrit | jichenjc proposed openstack/nova master: Enhance api-guide general info some updates https://review.openstack.org/561773 | 06:53 |
*** sususuryashines has joined #openstack-placement | 07:04 | |
openstackgerrit | jichenjc proposed openstack/nova master: Downgrade overquota warning https://review.openstack.org/572028 | 07:05 |
*** sususuryashines is now known as tssurya | 07:08 | |
openstackgerrit | jichenjc proposed openstack/nova master: Enhance api-guide general info some updates https://review.openstack.org/561773 | 07:23 |
*** giblet is now known as gibi | 07:27 | |
openstackgerrit | Zhenyu Zheng proposed openstack/nova master: Use ThreadPoolExecutor for max_concurrent_live_migrations https://review.openstack.org/563505 | 07:27 |
*** e0ne has quit IRC | 07:30 | |
*** e0ne_ has joined #openstack-placement | 07:30 | |
openstackgerrit | Zhenyu Zheng proposed openstack/nova master: Only run placement request filters when Placement will be called https://review.openstack.org/569969 | 07:33 |
*** ttsiouts has joined #openstack-placement | 07:50 | |
*** ttsiouts has quit IRC | 07:53 | |
*** ttsiouts has joined #openstack-placement | 07:57 | |
openstackgerrit | Takashi NATSUME proposed openstack/nova-specs master: Change a validation in creating a server group https://review.openstack.org/546484 | 08:16 |
openstackgerrit | Takashi NATSUME proposed openstack/nova-specs master: Change a validation in creating a server group https://review.openstack.org/546484 | 08:16 |
*** ttsiouts has quit IRC | 08:16 | |
*** ttsiouts has joined #openstack-placement | 08:46 | |
openstackgerrit | Chen proposed openstack/nova master: Fix a typo https://review.openstack.org/572061 | 09:33 |
openstackgerrit | tianhui proposed openstack/nova master: Fix bug for hypervisors https://review.openstack.org/572063 | 09:37 |
openstackgerrit | Zhenyu Zheng proposed openstack/nova master: Use ThreadPoolExecutor for max_concurrent_live_migrations https://review.openstack.org/563505 | 09:47 |
*** ttsiouts has quit IRC | 09:50 | |
*** ttsiouts has joined #openstack-placement | 09:51 | |
*** ttsiouts has quit IRC | 09:55 | |
*** alex_xu_ has joined #openstack-placement | 10:17 | |
*** purplerbot has quit IRC | 10:20 | |
*** alex_xu has quit IRC | 10:20 | |
openstackgerrit | Kashyap Chamarthy proposed openstack/nova master: libvirt: Remove usage of migrateToURI{2} APIs https://review.openstack.org/567258 | 10:30 |
*** ttsiouts has joined #openstack-placement | 10:32 | |
*** jrollinhatin is now known as jroll | 11:20 | |
*** rubasov has joined #openstack-placement | 11:24 | |
*** gibi has quit IRC | 11:31 | |
openstackgerrit | Jan Gutter proposed openstack/nova-specs master: Spec to implement vRouter HW offloads https://review.openstack.org/567148 | 11:32 |
openstackgerrit | Jan Gutter proposed openstack/nova master: Convert vrouter legacy plugging to os-vif https://review.openstack.org/571325 | 11:32 |
openstackgerrit | Jan Gutter proposed openstack/nova master: [WIP] Add support for vrouter HW offloads https://review.openstack.org/572082 | 11:32 |
*** gibi has joined #openstack-placement | 11:43 | |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Add policy to InstanceGroup object and api models. https://review.openstack.org/563375 | 11:49 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Add policy field to ServerGroup notification object https://review.openstack.org/563401 | 11:49 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Change the anti-affinity Filter to adapt to new policy https://review.openstack.org/571166 | 11:49 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Adapt _validate_instance_group_policy to new policy model https://review.openstack.org/571465 | 11:49 |
openstackgerrit | Yikun Jiang (Kero) proposed openstack/nova master: Microversion 2.63 - Use new format policy in server group https://review.openstack.org/567534 | 11:49 |
openstackgerrit | Damon Li proposed openstack/nova master: Remove minimum version check when save service https://review.openstack.org/572089 | 12:07 |
*** gibi has quit IRC | 12:10 | |
*** gibi has joined #openstack-placement | 12:10 | |
*** edmondsw has joined #openstack-placement | 12:12 | |
openstackgerrit | Damon Li proposed openstack/nova master: Remove minimum version check when save service https://review.openstack.org/572089 | 12:13 |
*** jroll has quit IRC | 12:29 | |
*** jroll has joined #openstack-placement | 12:30 | |
*** gibi has quit IRC | 12:30 | |
openstackgerrit | Jan Gutter proposed openstack/nova-specs master: Spec to implement vRouter HW offloads https://review.openstack.org/567148 | 12:32 |
*** openstackgerrit has quit IRC | 12:34 | |
*** gibi has joined #openstack-placement | 13:05 | |
*** gibi has quit IRC | 13:14 | |
*** jaypipes has joined #openstack-placement | 13:20 | |
*** gibi has joined #openstack-placement | 13:20 | |
*** tetsuro has joined #openstack-placement | 13:38 | |
*** openstackgerrit has joined #openstack-placement | 13:44 | |
openstackgerrit | Chen proposed openstack/nova master: Fix some inconsistencies in doc https://review.openstack.org/570407 | 13:44 |
*** cdent has joined #openstack-placement | 13:54 | |
*** mriedem has joined #openstack-placement | 14:00 | |
openstackgerrit | MultipleCrashes proposed openstack/nova master: Retry decorator fix for autoscale delete https://review.openstack.org/563418 | 14:10 |
*** e0ne_ has quit IRC | 14:16 | |
*** e0ne has joined #openstack-placement | 14:17 | |
*** alex_xu_ has quit IRC | 14:20 | |
*** alex_xu has joined #openstack-placement | 14:20 | |
*** takashin has joined #openstack-placement | 14:29 | |
*** tetsuro has quit IRC | 14:35 | |
*** ttsiouts has quit IRC | 14:40 | |
*** cdent has quit IRC | 14:44 | |
*** ttsiouts has joined #openstack-placement | 14:46 | |
*** alex_xu has quit IRC | 14:47 | |
*** alex_xu has joined #openstack-placement | 14:49 | |
gibi | I've tried to add another alternative to the https://etherpad.openstack.org/p/placement-migrate-operations from line 108 | 14:53 |
gibi | It is a lot less API heavy | 14:53 |
efried | gibi: responding | 14:59 |
gibi | efried: looking | 15:00 |
*** takashin has left #openstack-placement | 15:01 | |
*** cdent has joined #openstack-placement | 15:10 | |
openstackgerrit | melanie witt proposed openstack/nova stable/pike: zuul: Move legacy jobs to project https://review.openstack.org/572130 | 15:17 |
*** purplerbot has joined #openstack-placement | 15:17 | |
*** e0ne has quit IRC | 15:19 | |
*** ttsiouts has quit IRC | 15:19 | |
openstackgerrit | melanie witt proposed openstack/nova stable/ocata: zuul: Move legacy jobs to project https://review.openstack.org/572132 | 15:21 |
jaypipes | friggin etherpad... | 15:26 |
jaypipes | I don't know wtf is going on with it. | 15:26 |
jaypipes | efried: grabbing some nosh. back in about an hour. | 15:27 |
efried | ack | 15:28 |
openstackgerrit | Matt Riedemann proposed openstack/nova-specs master: Spec for volume multiattach enhancements https://review.openstack.org/552078 | 15:40 |
openstackgerrit | Dan Smith proposed openstack/nova master: Use oslo.messaging per-call monitoring https://review.openstack.org/566696 | 15:44 |
efried | jaypipes, cdent, gibi: We might be making this more complicated than it needs to be. All we really need is an atomic transaction that can | 15:45 |
efried | - set inventories on >1 RP | 15:45 |
efried | - set multiple allocations. | 15:45 |
efried | We already have a JSON payload format for setting an inventory. | 15:45 |
efried | We already have a JSON payload format for setting multiple allocations. | 15:45 |
efried | So let's just combine those. | 15:45 |
efried | Working this up on the etherpad... | 15:45 |
*** tssurya has quit IRC | 15:48 | |
cdent | efried: are you sure? I'm not sure that I entirely understand the problem. There seemed to be some edge cases that made your description incomplete. However if it is possible to do what you're saying, that would be great | 15:48 |
efried | cdent: Which edge cases? | 15:48 |
efried | cdent: I don't think it's important for the payload to describe which units are *moving* from where to where. I think we only need to know which units *are* where by the time we're done. | 15:49 |
cdent | That's the point: I don't know. But the etherpad, at least in its earliest forms seemed to suggest that splitting of some kind was required. | 15:49 |
cdent | I agree that representing desired state, rather than the transformation, is better. | 15:49 |
cdent | but it wasn't clear to me if that was going to be suitable for the allocations | 15:50 |
cdent | where "suitable" might mean "concise" (not that I'm really that concerned about concise, just trying to be fair to people who do care about such things) | 15:51 |
efried | splitting still possible with this. | 15:52 |
openstackgerrit | MultipleCrashes proposed openstack/nova master: Retry decorator fix for autoscale delete https://review.openstack.org/570370 | 15:53 |
cdent | the concise form is "move these allocations for these consumers", but if we just say "make the allocations look like this" that achieves the same thing, so I'd be in favor of that | 15:53 |
efried | and in fact, the only additional bloat is having to include the full inv for the "source" RPs, I think. | 15:53 |
efried | also having to include the proj/user id info in the allocations, I suppose. | 15:54 |
cdent | from a RESTy standpoint expressing the full state of everything is nice | 15:54 |
efried | cdent: Have we not put RP gen into POST /allocations? | 15:58 |
efried | cdent: Was that part of what's happening in the consumer gen bp? | 15:58 |
cdent | just consumer, not rp, because there are^wcan be multiple rps involved. | 15:59 |
efried | cdent: Right, the rp gen would have to go in the allocation dict for each rp. | 16:01 |
efried | { | 16:02 |
efried | "30328d13-e299-4a93-a102-61e4ccabe474": { | 16:02 |
efried | "project_id": "131d4efb-abc0-4872-9b92-8c8b9dc4320f", | 16:02 |
efried | "user_id": "131d4efb-abc0-4872-9b92-8c8b9dc4320f", | 16:02 |
efried | "allocations": { | 16:02 |
efried | "e10927c4-8bc9-465d-ac60-d2f79f7e4a00": { | 16:02 |
efried | "resources": { | 16:02 |
efried | "VCPU": 2, | 16:02 |
efried | "MEMORY_MB": 3 | 16:02 |
efried | } | 16:02 |
efried | "resource_provider_generation": <gen>, <==== HERE - this is what we don't have today (at least in the docs) | 16:02 |
efried | } | 16:02 |
efried | } | 16:02 |
efried | }, | 16:02 |
cdent | why do you want to do that? | 16:03 |
efried | want? | 16:03 |
efried | :) | 16:03 |
cdent | why do you want gen in the allocations? | 16:03 |
efried | I'm just spitballing here. I noticed it's missing and I'm trying to figure out if we need it. | 16:03 |
efried | we don't need it? | 16:03 |
efried | Don't we update rp gen when we allocate against the rp? | 16:03 |
* efried looks | 16:03 | |
* cdent is on a call | 16:05 | |
efried | answer is yes. | 16:05 |
efried | https://github.com/openstack/nova/blob/master/nova/api/openstack/placement/objects/resource_provider.py#L2288 | 16:06 |
cdent | yes, but why do you want to send it? | 16:06 |
cdent | s/you want/considering/ | 16:07 |
efried | I think we discussed this in Dublin when we were conceiving the consumer gen work. | 16:07 |
efried | and came to the conclusion that we don't. | 16:07 |
efried | but I don't remember why | 16:07 |
efried | If something has changed about the provider, I want to re-evaluate whether I really want to allocate against it? | 16:08 |
efried | But I guess that way lies thrashing in the scheduler. | 16:08 |
cdent | what you care about with allocations is whether you can make the allocation or not, that's all | 16:09 |
cdent | if we make it more complicated than that, that way lies madness | 16:09 |
efried | I can buy it. | 16:11 |
efried | jaypipes, cdent, edleafe, bauzas, gibi: Okay, finished proposal for "specify the end state". See L67-98. https://etherpad.openstack.org/p/placement-migrate-operations | 16:25 |
efried | please do read the preamble | 16:26 |
cdent | efried: superficial read (while on the phone) it seems sane and complete but it is hard for me to think about clearly without test cases. Will make some notes | 16:35 |
efried | cdent: ack, thanks for looking. | 16:35 |
jaypipes | efried: ack. will do shortly. just wrapping up yet another revision on the cpu-resources spec. | 16:38 |
efried | thx | 16:38 |
efried | you really should just get that thing merged already | 16:39 |
jaypipes | lol | 16:40 |
*** edmondsw has quit IRC | 16:41 | |
*** ttsiouts has joined #openstack-placement | 16:41 | |
*** tssurya has joined #openstack-placement | 16:46 | |
*** edmondsw has joined #openstack-placement | 16:46 | |
*** e0ne has joined #openstack-placement | 16:47 | |
*** ttsiouts has quit IRC | 16:49 | |
*** ttsiouts has joined #openstack-placement | 17:24 | |
*** ttsiouts has quit IRC | 17:41 | |
*** ttsiouts has joined #openstack-placement | 17:46 | |
*** tssurya has quit IRC | 18:20 | |
*** tssurya has joined #openstack-placement | 18:37 | |
*** mriedem has quit IRC | 18:40 | |
*** mriedem has joined #openstack-placement | 18:43 | |
openstackgerrit | Matt Riedemann proposed openstack/nova master: Fix typo in enable_certificate_validation config option help https://review.openstack.org/572185 | 18:51 |
openstackgerrit | Jay Pipes proposed openstack/nova-specs master: Standardize CPU resource tracking https://review.openstack.org/555081 | 18:55 |
cdent | mriedem, efried, jaypipes: wanted to get a sanity check on a desire: I'd like to set the default for the new [placement_database]/connection conf setting to an environment variable . That doesn't seem to be a thing that is commonly done in nova and derivatives, but makes containerisation happy. Is there a reason to not do it? | 18:56 |
jaypipes | cdent: you mean doing something like default=os.environ.get('PLACEMENT_DB_CONNECTION', DEFAULT_PLACEMENT_DB_CONNECTION) or something like that? | 18:59 |
cdent | yeah | 18:59 |
mriedem | we don't have defaults on db connections do we? | 19:00 |
jaypipes | cdent: ++ I would definitely support that. | 19:00 |
mriedem | so wouldn't the env var default to None? | 19:00 |
cdent | mriedem: yes | 19:00 |
jaypipes | cdent: FTR, I've always written my python code like that. It's a pattern I think should be used in OpenStack. (see https://github.com/jaypipes/os-lively/blob/master/os_lively/conf.py for example...) | 19:01 |
cdent | right, I'll make it happens. Also looking into seeing if it is doable for some of the keystone middleware settings | 19:01 |
cdent | jaypipes: yeah, totally agree it is the right way to go | 19:01 |
*** ttsiouts has quit IRC | 19:04 | |
* cdent lunches | 19:04 | |
*** cdent has quit IRC | 19:05 | |
jaypipes | cdent: would you be able to do 11am EST / 8am PST tomorrow morning for a decision-making hangout on the migration API endpoint thing? | 19:05 |
efried | jaypipes: Seen proposal at L67 yet? | 19:20 |
efried | egads, there's three of you on the etherpad now. | 19:20 |
efried | one of you is apparently old | 19:20 |
jaypipes | efried: yeah, no idea why there's two of me on the damn thing :( | 19:25 |
efried | jaypipes: It's effectively the same as what you're proposing at the bottom - except it uses existing payload formats. | 19:27 |
openstackgerrit | Dan Smith proposed openstack/nova master: Change consecutive build failure limit to a weigher https://review.openstack.org/572195 | 19:47 |
*** ttsiouts has joined #openstack-placement | 19:53 | |
*** cdent has joined #openstack-placement | 19:57 | |
openstackgerrit | Matt Riedemann proposed openstack/nova-specs master: Remove device from volume attach requests (spec) https://review.openstack.org/452546 | 20:06 |
jaypipes | efried: gotta love etherpad "conversations" :) | 20:08 |
efried | jaypipes: ikr. Especially with the colors, my eyes come back here and everythnig looks all washed out. | 20:08 |
jaypipes | efried: still haven't figured out where "old jaypipes" is coming from :( | 20:09 |
efried | "Ghost of Jay" | 20:09 |
jaypipes | heh | 20:09 |
*** e0ne has quit IRC | 20:13 | |
efried | cdent: 8am Pacific work for you tomorrow? | 20:13 |
efried | edleafe: 10am Central work for you tomorrow? | 20:13 |
cdent | efried: should be fine. free until 10, and can skip the thing at 10 if needed | 20:14 |
efried | cdent: What's the reason we've avoided PATCH thus far? | 20:15 |
cdent | PATCH and PUT are supposed to have very specific meanings, PATCH moreso than PUT, but both are often usefully violated. In the most strict sense PATCH is supposed to use a diff-like or tranformattion-oriented syntax on a _single_ resource (a thing that can be identified by a URI) | 20:17 |
cdent | so while we might consider PATCHing an inventory, resource provider, or allocation, doing that operation on a suite of things would be weird | 20:17 |
openstackgerrit | Jay Pipes proposed openstack/nova-specs master: Standardize CPU resource tracking https://review.openstack.org/555081 | 20:18 |
efried | cdent: Mm, gotcha. | 20:18 |
efried | cdent: That said, what's the usual verb for an operation that *does* work on a suite of things? | 20:20 |
cdent | POST | 20:21 |
efried | ight | 20:21 |
cdent | POST is the "transform state in arbitrary ways" method | 20:21 |
edleafe | efried: I should be available at 10am CDT tomorrow | 20:23 |
* cdent is trying to decode the etherpad | 20:23 | |
*** ttsiouts has quit IRC | 20:23 | |
efried | cdent: I'm about to clean it up into a new pad. Glad to get feedback in its current state if you can stomach it, but feel free to wait. | 20:24 |
cdent | I think I've digested it. The one thing I'm missing is the positive justification for the PATCH format. I understand that it avoids some of the dangers and verbosity of the POST version, but interested what else. | 20:25 |
efried | jaypipes, cdent, edleafe, gibi, bauzas: Tuesday 1500 UTC for discussing "placement API to support upgrades". | 20:25 |
cdent | ✔ | 20:25 |
efried | cdent: which POST? The "additions+removals" one, or the "whole payload" one? | 20:26 |
cdent | whole payload | 20:26 |
jaypipes | yup, efried ++ | 20:27 |
efried | cdent: Okay, jaypipes's concerns there start on L102 | 20:28 |
jaypipes | cdent: the PATCH format was added as an option. Not sure there's a positive justification for it, besides just being another possibility. | 20:28 |
efried | (or for those of you with weird line numbers, right after the JSON for POST /inventories_and_allocations) | 20:28 |
jaypipes | cdent: I added it in response to gibi's statement earlier today that he was looking for "what we need is a series of existing placement API calls executed in an atomic way." | 20:29 |
jaypipes | cdent: and so I popped the PATCH stuff up as a possible thing that did that. | 20:30 |
jaypipes | cdent: or kind of did that ... :) | 20:30 |
*** ttsiouts has joined #openstack-placement | 20:31 | |
cdent | blargh, office talkers distracting me | 20:31 |
efried | jaypipes: https://tools.ietf.org/html/rfc7396 is another option for PATCH, less verbose than the http://tools.ietf.org/html/rfc6902 format. Still has the ookiness of the fact that we're operating on multiple resources at once, which IIUC is going to be the objection to *any* PATCH. | 20:35 |
efried | (I'm thinking maybe we're not quiiite ready yet to distill down to a smaller set of proposals :) | 20:35 |
efried | (cause we're still in 'expand' mode, pretty much) | 20:35 |
cdent | I'll be lightly -1 to any merge/diff/transform format simply because it is less scrutable (and thus, to me, less testable) than a full-state thing | 20:37 |
efried | something about cdent's scrutum... | 20:37 |
*** e0ne has joined #openstack-placement | 20:38 | |
jaypipes | efried: but merge patch has no ability to represent the removal of elements, right? | 20:39 |
efried | jaypipes: yeah: null | 20:39 |
jaypipes | ah. | 20:39 |
jaypipes | cdent: my personal opinion is that the list of operations in the JSON Patch is actually pretty easy to scrutinize. It's explicit about what is being added, removed, and tested for, as opposed to just "replace the representation of this thing with this other thing" (and needing to have the "test" actions represented in other ways (the generation checks, etc) | 20:41 |
cdent | yeah, I see that, thus "lightly" | 20:41 |
jaypipes | cdent: but at the end of the day, we just need to settle on one of these methods and go with it. not having this functionality means not being able to do in-place compute upgrades which means a bunch of nested providers patch series are dead in the water... | 20:42 |
* cdent nods | 20:42 | |
cdent | I'm pretty sure that the placement side of this fix is the easy part, yeah? | 20:42 |
jaypipes | as much as it would please me to say "f it, just have the operators perform the slide puzzle" :) | 20:42 |
efried | cdent: The nova side is actually not bad. | 20:42 |
cdent | I'll believe you when I see it, efried | 20:43 |
jaypipes | cdent: I don't think either side is all that complicated. just needs to be agreed upon which side is responsible for what. | 20:43 |
efried | jaypipes: Remember, if we want to buy time, we can hack the upgrade temporarily without changing the placement API at all ("locking" providers by setting reserved=total while we shuffle stuff around). | 20:43 |
efried | then we can take our time figuring out the PERFECT form for this single-transaction migration. | 20:44 |
jaypipes | cdent: I personally think that placement should be responsible for all locking/transactional awareness and nova side should just be responsible for telling placement *either* what it wants the desired state to be *or* a list of instructions for placement to take in an atomic fashion. | 20:44 |
cdent | I thinnk efried has just landed on the solution: custom http verb: PERFECT | 20:44 |
efried | (Yeah, that's a lie, still have to have one leetle API change to allow us to "overallocate" when reserved==total) | 20:44 |
jaypipes | heh | 20:45 |
efried | heh | 20:45 |
jaypipes | efried: we won't be able to shuffle anything around while reserved=total... | 20:45 |
jaypipes | efried: that just prevents scheduling, really. | 20:45 |
cdent | jaypipes: I think we're in agreeement. My assertion is that the state awareness that nova is going to need to have to be able to express its needs to placement (whatever the format) is ... messy | 20:45 |
efried | jaypipes: Right, which is the only thing we actually need to prevent here. | 20:45 |
cdent | but then I tend to think stuff is messy a lot sooner than either of you two | 20:46 |
efried | cdent: Most of that messy awareness is already in update_from_provider_tree | 20:46 |
efried | cdent: with the exception of the allocations | 20:46 |
cdent | but yes: placement needs to be the authority on locking/transactions | 20:47 |
jaypipes | cdent: well, there's definitely gonna be mess... but the virt driver is really the only thing that knows how to say "this single compute node is now a parent and there are two children that represent these multiple NUMA nodes. put the allocations for instances X, Y, and Z on child A and put the allocations for instances Xa, Ya, and Za on child B" | 20:47 |
cdent | sure, I'm not suggesting we can avoid the mess, simply that we should make sure we position our undergarments before we leap in | 20:48 |
cdent | manage expectations, etc | 20:48 |
jaypipes | ack | 20:49 |
efried | jaypipes, cdent: It's a good point: right now the virt driver, via update_provider_tree, is essentially providing us with the *full picture* (minus the allocations atm) so that's a mark in favor of the "full spec" version of the API. | 20:49 |
jaypipes | efried: ack. again, I'm not opposed to the whole "tell me the desired state" thing. I just have some reservations about giving users a "super API" that may end up being overused. | 20:54 |
jaypipes | efried: if you thought CERN having issues with our overly loquacious API calls for refreshing provider associations, having *any* caller make overuse of such a proposed "replace the world" API would be a similar performance issue.. | 20:56 |
*** ttsiouts has quit IRC | 20:57 | |
*** ttsiouts has joined #openstack-placement | 20:57 | |
cdent | jaypipes, efried: Isn't that just rope that the operators can choose to hang themselves with or not? Especially since the state gathering for huge operations will be itself painful, so people probably won't want to? | 20:57 |
jaypipes | cdent: not sure... | 20:58 |
efried | jaypipes: To be fair, refreshing provider associations was making many calls vs. one. Not sure how a single "refresh" (à la https://review.openstack.org/#/c/521875/) would have affected their env... | 20:58 |
efried | cdent: I agree with that. Though we're not talking about "operators" here so much as "API consumers" like potentially e.g. cyborg. | 20:59 |
jaypipes | efried: a single "replace the world" operation would essentially block all activities (in the scheduler and elsewhere) for all providers in that provider tree for the duration of the operation. | 20:59 |
cdent | these are all system-admin tasks, though, yes? | 21:00 |
jaypipes | cdent: I'm thinking of Cinder and Cyborg as examples here. | 21:00 |
cdent | it's not like an end user can bring down multiple nova-computes with a simple script posted josh | 21:00 |
cdent | presumably code review in cinder and cyborg would help avoid that? | 21:00 |
jaypipes | cdent: imagine if cinder, because it was easier, decided to do a "replace the world" API call every time it updated something for a shared volume storage provider... | 21:00 |
cdent | wouldn't they learn quickly that was a bad idea? | 21:01 |
jaypipes | cdent: :) I have no idea. | 21:01 |
jaypipes | in any case, I'm just logging concerns I have, nothing more. | 21:02 |
*** e0ne has quit IRC | 21:02 | |
cdent | presumably the replace the world problems are present with the patch version too, assuming an adventurous client | 21:03 |
cdent | and I think it is fair to assume that if we have any mode for replacing the world, people will explore ways to use it | 21:03 |
cdent | and that's just how it goes? | 21:03 |
jaypipes | perhaps | 21:04 |
*** tssurya has quit IRC | 21:10 | |
*** tssurya has joined #openstack-placement | 21:11 | |
efried | jaypipes: Okay, so as I put together the curated/distilled etherpad, are you okay if I use merge-patch (RFC 7396) as the PATCH proposal? | 21:17 |
jaypipes | efried: as an alternative to RFC6902? or as a replacement (pun intended) for that? | 21:18 |
efried | replacement. | 21:19 |
*** ttsiouts has quit IRC | 21:19 | |
*** ttsiouts has joined #openstack-placement | 21:19 | |
efried | jaypipes: Sayin, we're likely to argue pro or con PATCH on the merits of PATCH as opposed to on the exact syntax of the payload | 21:20 |
efried | though maybe that's a wrong assumption. | 21:20 |
efried | just trying to pre-cull as much as possible. | 21:20 |
jaypipes | efried: sure, that's cool with me. maybe just leave a note about RFC6902 being an alternative format for PATCH. | 21:20 |
efried | roger that. | 21:20 |
jaypipes | cool, thx. | 21:20 |
jaypipes | efried: personally, I'm looking forward to not seeing "old jaypipes" on this new etherpad. | 21:21 |
efried | heh. | 21:21 |
efried | jaypipes: You'll have to join as "puggsy_malone" | 21:22 |
jaypipes | heh | 21:22 |
jaypipes | mmm, Jules is making succotash for dinner. yum. | 21:22 |
efried | you're not suffering, then, Yosemite Sam? | 21:24 |
efried | ah, boo, that's Sylvester, not Sam. | 21:24 |
*** tssurya has quit IRC | 21:27 | |
jaypipes | efried: heh, no :) | 21:29 |
*** tssurya has joined #openstack-placement | 21:29 | |
*** tssurya has quit IRC | 21:30 | |
*** edmondsw has quit IRC | 21:35 | |
openstackgerrit | Lee Yarwood proposed openstack/nova stable/queens: Allow cinderv2 endpoints within the request context catalog https://review.openstack.org/572213 | 21:46 |
*** ttsiouts has quit IRC | 21:52 | |
openstackgerrit | David Bingham proposed openstack/nova master: Adding apply_cells to nova-manage to enable automated cells_v2 configuration. https://review.openstack.org/572216 | 22:08 |
efried | jaypipes: Having written up the merge-patch version, it's so close to the removals-and-additions proposal that I'm just going to strike the latter. Basically it would be the same proposal but with a different `VERB /endpoint`. | 22:11 |
efried | jaypipes: So I'm going to consolidate those with a note, if you're cool with that. | 22:11 |
efried | cdent: ^ | 22:13 |
efried | jaypipes, cdent: https://etherpad.openstack.org/p/placement-making-the-(up)grade <== wanna take a first pass before I send out a note? | 22:16 |
mriedem | who is going to be rodney dangerfield in this skit? | 22:18 |
mriedem | oh wait, different 80s movie https://www.imdb.com/title/tt0087666/ | 22:18 |
openstackgerrit | David Bingham proposed openstack/nova master: Adding apply_cells to nova-manage to enable automated cells_v2 configuration. https://review.openstack.org/568987 | 22:28 |
cdent | efried: looking | 22:29 |
jaypipes | efried: gimme a bit to digest, ok? | 22:29 |
efried | jaypipes: cdent: I'm adding preamble about the big picture. Jump to "New Placement API Call" | 22:29 |
*** mriedem has quit IRC | 22:30 | |
efried | jaypipes, cdent: Okay, I think I finished the preamble. | 22:42 |
efried | jaypipes, cdent: Have a note drafted announcing the etherpad and the hangout. Let me know when I can punch the Send button. | 22:46 |
cdent | efried: at the very end there is an unfinished sentence? | 22:48 |
efried | cdent: No, it was going to be a third proposal, but I realized it was going to be basically the same as the second. I've deleted it. Thanks for the catch. | 22:48 |
*** mriedem has joined #openstack-placement | 22:48 | |
jaypipes | efried: lgtm. | 22:48 |
efried | thx | 22:48 |
cdent | the "needs to suport" section is a good idea | 22:49 |
efried | cdent: What am I forgetting therein, that we already discussed? | 22:53 |
* cdent re-reads | 22:54 | |
cdent | efried: I can't think of anything else | 22:58 |
efried | jaypipes: cdent: Okay, thanks. Emailing, then bailing out for the evening. See y'all in the a.m. | 22:59 |
cdent | thanks efried | 22:59 |
*** cdent_ has joined #openstack-placement | 23:03 | |
*** cdent has quit IRC | 23:04 | |
*** cdent_ is now known as cdent | 23:04 | |
*** cdent_ has joined #openstack-placement | 23:24 | |
*** cdent has quit IRC | 23:26 | |
*** cdent_ is now known as cdent | 23:26 | |
*** takashin has joined #openstack-placement | 23:34 | |
*** mriedem has quit IRC | 23:48 |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!