Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-11-28
20:50:15 mriedem efried: that's what i said, i think it's a limitation in the api until 1.12 and we have to handle it in the cli
20:50:42 mriedem melwitt: well, we could just hold off on approving the first unmerged patch with the new option until people are happy with the entire series
20:50:42 melwitt mriedem: so I'm not seeing a better way to do the reno except at the end https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:master+topic:bp/websocket-proxy-to-host-security
20:50:43 efried mriedem rgr. Though "limitation" might be a bit strong.
20:50:55 melwitt mriedem: okay, that works
21:00:30 openstackgerrit Merged openstack/nova master: Enable cold migration with target host(1/2) https://review.openstack.org/408955
21:25:43 melwitt weird, yet another time when it says merge conflict and I rebase locally but there was no conflict ...
21:27:29 melwitt mriedem: should I just try the rebase button on https://review.openstack.org/#/c/498983 ? it says merge conflict but there's no conflict to resolve when I pull it down and rebase locally
21:28:35 mriedem sure
21:29:07 melwitt k, trying that now
21:29:34 openstackgerrit melanie witt proposed openstack/nova master: Save updated libvirt domain XML after swapping volume https://review.openstack.org/498983
21:44:50 penick Hey folks, we're gearing up to replace our quota by flavor (and az) patches with something we could actually upstream. At the moment (based on hazy recollection of conversations at the summit) the direction i'm heading in is to do quota by custom resource class, wherein a resource class could be a combination of AZ and Flavor. Does that sound sane?
21:45:05 mriedem mlavalle: is this the neutron port binding extension that i'd want for the live migration thingy? http://paste.openstack.org/show/627643/
21:45:21 penick Wanted to see if anyone had some input before we started working on the spec.
21:45:58 melwitt penick: I think what we discussed was leveraging the quota class facility we already have and adding the ability to tag flavors with a quota class
21:47:37 penick melwitt: ah, I also have to do quota by flavor and AZ as well. Can I make that work using quota classes too?
21:47:45 mriedem mlavalle: i'm guessing it's not, since it looks like that extension is just an indication of the binding:profile stuff on the port, i.e. https://developer.openstack.org/api-ref/network/v2/index.html#port-binding-extended-attributes
21:48:02 mlavalle mriedem: it's not
21:48:26 mlavalle that is the existing extension
21:49:00 melwitt penick: this gives a high level view of what quota classes are https://docs.openstack.org/nova/pike/user/quotas.html so it's already possible to set quota on a per class basis i.e. "openstack quota set --class myclass --instances 5"
21:49:52 melwitt it's just that currently there is no way to specify a quota class with any one request. rackspace used to do it via their own custom paste middleware, IIRC
21:51:52 penick oh ok, I didn't realize that it could be custom resource classes too, not just the typical resources like instances, ram, etc
21:52:08 mriedem not the same thing
21:52:16 melwitt as for AZ, I guess to do it that way there would need to be a way to tag an AZ with a quota class
21:52:16 mriedem resource classes are a placement concept
21:52:21 gryf quota class is different thing
21:52:21 mriedem quota classes are a nova concept
21:52:59 mriedem placement doesn't have any concept of quota
21:53:20 penick Ah, that makes sense.
21:53:21 mriedem we've talked about replacing nova's quota calculation code using the consumer allocation / resource class stuff in placement at some point, but there are issues
21:53:42 mriedem because the quota usage in nova is also tracked via allocatoins in placemetn for things like vcpu/memory_mb/disk_gb
21:54:35 melwitt yeah. a quota class is a logical grouping of quotas. like you could create a 'dev class' that has certain quota limits and you could create a different 'prod class' that has different quota limits
21:54:42 mriedem penick: you could start with a backlog spec to document the problem and use case
21:54:51 mriedem w/o getting into design or implementation details
21:55:15 mriedem note that today you can create any number of custom quota classes in nova, but the only one that the code actually gives 2 shits about is the 'default' quota class
21:55:38 mriedem ala https://review.openstack.org/#/c/411035/
21:55:42 melwitt they've proposed a spec about it in the past, it got gridlocked on the implementation details
21:55:55 mriedem they = yahoo, past = juno?
21:56:22 mriedem and was it the same guy that was doing the zookeeper spec that gridlocked for 3 years?
21:56:32 melwitt I was thinking a reasonable way to actually get it done would be to use quota classes and provide a way to tag flavor with a quota class
21:56:36 mriedem or tooz or kazoo or whatever it was
21:56:37 melwitt mriedem: yes
21:57:13 penick heh
21:57:25 mriedem welp, i'd dig up the old spec i guess because i'm not familiar with the use case / problem description
21:57:32 mriedem and i'm sure that predated anything with placement
21:57:48 mriedem *plus* no chance for queens, so you'd have to tee this up for discussion in dublin
21:58:12 melwitt mriedem: this is the old spec https://review.openstack.org/#/c/206160/
21:59:11 melwitt it was agreed to be a reasonable use case, just that all of the implementation ideas were pretty hairy
21:59:29 mriedem anything involving AZs is going to be hairy as hell
21:59:35 mriedem or quotas
21:59:44 mriedem is there some way we can work boot from volume into this?
21:59:51 penick You bet your ass we can
21:59:55 melwitt being able to tag a flavor with a quota class would get us quota by flavor at least, and the same use case has come up in the preemptable instances discussions
22:01:08 melwitt that is, for preemptable instances, you could have a quota class 'preemtable' which is unlimited quota and tag preemptable instance flavors with that class. so that users booting preemptable instances don't use up their quota when they boot them
22:02:35 mriedem mlavalle: ok so i'm looking for the "binding-extended" extension per https://review.openstack.org/#/c/484389/8/neutron_lib/api/definitions/portbindings_extended.py
22:03:12 mriedem mlavalle: i'm looking at starting to build up the nova patches for this thing, and part of that is just asking neutron if the api extension is even available to do the deed
22:05:13 mlavalle mriedem: that is the extension definition. as far as the implementation, I am working on it
22:06:02 openstackgerrit Ken'ichi Ohmichi proposed openstack/nova master: Remove unnecessary spaces on JSON samples https://review.openstack.org/523543
22:06:24 penick So If I want to limit tenant $foo to 10 instances of flavor $bar, and I want to limit tenant $baz to 20 instances of flavor bar, i'd create 2 quota classes, one for tenant $foo, one for tenant $baz, then tag flavor $bar with both quota classes?
22:09:39 penick is that how the flow would go? Assuming I could tag a flavor with the quota class
22:09:42 melwitt penick: hm, no. what I was saying would only work to have 1 set of quotas for a flavor if every tenant gets the same quotas for that flavor
22:09:52 penick oh, crud
22:10:42 melwitt basically, you'd need a quota class per unique set of quota limits
22:11:14 melwitt I would think to a large extent many tenants would be able to use the same class?
22:12:23 penick i'm checking the DB on one of our prod clusters right now..
22:12:41 mriedem if we ever get limits information from keystone, doesn't that kind of throw a wrench into all of this?
22:12:50 melwitt what I mean is, you would only need one class per tenant/flavor if they were all literally different
22:12:51 mriedem because that's the eventual replacement for quota classes in nova
22:13:53 melwitt it would throw a wrench if we won't have quota classes in keystone
22:16:10 penick Ok, so i'll need to assume one class per tenant and flavor
22:16:25 melwitt hang on, let me look at this again real quick.
22:16:34 penick mriedem are limits definitely moving to keystone? Because i've heard talk about that for a year or two..and nothing's happened.
22:16:57 dansmith well, people keep getting laid off
22:17:00 dansmith kinda gets in the way of progress
22:17:37 mriedem someone from huawei was supposedly working on picking up that spec in keystone, but last i heard in sydney it was dropped in lbragstad's lap, and he's trying to move the rbac stuff forward
22:18:34 lbragstad we do have a new owner for that work
22:18:59 lbragstad https://review.openstack.org/#/c/455709/
22:19:14 lbragstad which is still under sdague's name but wxy is picking it up
22:20:02 lbragstad so if there is anything nova needs out of that interface from keystone, let us know
22:21:07 openstackgerrit Matt Riedemann proposed openstack/nova master: Add check if neutron "binding-extended" extension is available https://review.openstack.org/523548
22:22:14 openstackgerrit Matt Riedemann proposed openstack/nova master: Add check if neutron "binding-extended" extension is available https://review.openstack.org/523548
22:22:23 penick if Keystone is going to be able to land the limits api in queens, will Nova move to use it for Rocky?
22:22:37 lbragstad yeah - that's the idea
22:23:05 lbragstad but there are other things that playing into it too, which sean was really driving in Boston (like what enforcement models to use)
22:23:21 lbragstad we're not quite sure how that's going to get applied as the service yet
22:30:05 melwitt penick: I looked through the code again and indeed quota classes are not per tenant. so it won't be as simple as what I thought earlier
22:30:33 penick melwitt: dang
22:40:33 openstackgerrit Matt Riedemann proposed openstack/nova master: Add check if neutron "binding-extended" extension is available https://review.openstack.org/523548
22:43:11 openstackgerrit Thomas Goirand proposed openstack/nova master: qemu-img do not use cache=none if no O_DIRECT support https://review.openstack.org/523554
22:45:15 zigo melwitt: ^
22:46:05 melwitt cool, looking
22:51:41 openstackgerrit Mathieu Gagné proposed openstack/nova master: Fix rebuild of baremetal instance when vm_state is ERROR https://review.openstack.org/523559
22:53:08 mriedem efried: can i make simple rest calls like post() from a ksa adapter object?
22:53:37 mriedem mgagne: broken already :)
22:54:13 mgagne mriedem: what's broken?
22:54:18 mriedem mgagne: oh nvm, was thinking the rebuild + config drive thing
22:54:45 mgagne mriedem: hehe, we have a theme this month: rebuilding stuff =)
22:55:40 dansmith and re-rebuilding
22:55:48 mgagne that one had an interesting side effect: rebuild is actually happening on Ironic side but Nova stays in ERROR while rebuild continues
22:57:45 mriedem efried: nvm, answer is yes, that's exactly how it's used for the placement client in SchedulerReportClient

Earlier   Later