Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-02-12
14:46:02 gibi mriedem, bauzas: I don't think we can enforce to have image_id in the request spec
14:46:08 mriedem gibi: that vm is worthless though right?
14:46:31 gibi mriedem: yeah, I oversimplified it, but I could attach my-vol to a vm, populate it, detach it, then boot from it
14:47:48 mriedem sure, i wasn't saying we should 'require' an id in the requestspec.image field,
14:48:04 mriedem just that in certain cases (presumably the more normal cases), it's available via the volume_image_metadata
14:48:19 mriedem and furthermore, we can use that information later if a cell is down
14:48:30 mriedem and we use the request spec to represent the instance in the GET /servers/{id}/detail response
14:48:39 mriedem *if we use
14:49:28 gibi mriedem, bauzas: OK, it is clearer now. I had the feeling that we want to make image_id mandatory in the request spec
14:50:10 mriedem jroll: added the healthcheck middleware to both https://etherpad.openstack.org/p/self-healing-queens-ptg and https://etherpad.openstack.org/p/api-sig-ptg-rocky
14:50:21 mriedem we probably need a few more SIGs for this
14:51:18 bauzas mriedem: heh for "more SIGs"
14:51:30 bauzas that can become a very discussed and argued thing
14:51:36 openstackgerrit Chris Dent proposed openstack/nova master: WIP: Move resource provider objects into placement hierarchy https://review.openstack.org/540049
14:51:37 openstackgerrit Chris Dent proposed openstack/nova master: Optional separate database for placement API https://review.openstack.org/362766
14:51:37 openstackgerrit Chris Dent proposed openstack/nova master: WIP: Isolate placement database config https://review.openstack.org/541435
14:51:41 jroll mriedem: what have I done
14:51:58 jroll this really wasn't something I intended to tilt that hard at
14:51:58 bauzas gibi: mriedem: for the image.id being mandatory, I think I understand the point
14:52:19 bauzas gibi: mriedem: the problem is that AFAICS we assert in a lot of places about having an instance with an image
14:52:33 mriedem jroll: i just wanted to see it socialized more before it's adopted across a few projects,
14:52:53 mriedem because i'm projecting it on the community wide goals process
14:52:54 bauzas so we can 'fix' that, but that won't solve the design issue that we accepted in the past a different paradigm
14:52:59 jroll mriedem: also, I wasn't aware of the "new" thing until today, so adding the new thing you linked wasn't my intention at all
14:53:19 mriedem jroll: so your ironic patch and graham's spec aren't the same thing?
14:53:21 jroll mriedem: the new thing sounds like it'll take a year just to get through committee and have code written
14:53:23 jroll correct
14:53:27 bauzas jroll: from a very practical standpoint, do you feel blocked ?
14:53:29 jroll mine is the "old" one
14:53:38 mriedem bauzas: jim just wants to upstream it, but meh
14:53:40 mriedem it sounds like
14:53:45 jroll bauzas: no, what matt said
14:53:45 mriedem it's middleware
14:53:46 bauzas jroll: because I don't really see *why*it needs to be something
14:54:13 jroll bauzas: because it's what we use for every paste-enabled service, and so trying to standardize the rest there
14:54:49 bauzas jroll: well, you wanna standardize your own monitoring model :)
14:55:04 jroll bauzas: s/your own/oslo's/
14:55:10 gibi bauzas: on the image_id, theoretically we can forbid booting from volume in nova where the volume doesn't have the image_id in the metadata, but I think that would be restrict some use cases that is possible today
14:55:12 cdent jroll: I'd say go ahead with the old thing. the new thing needs to be discussed but is a whole different kettle
14:55:14 jroll I'm not forcing it on anyone ¯\_(ツ)_/¯
14:55:20 bauzas I'm not sure operators are sold on using middleware healthchecks for knowing whether my app is up
14:55:48 cdent jroll: I reckon the way it is done in ironic is just fine for placement
14:56:06 bauzas gibi: If I'm sure about one thing, it's how operators can find solutions with things we don't officially support
14:56:07 mriedem gibi: i wasn't suggesting making that break now,
14:56:13 cdent it doesn't preclude some different future
14:56:19 mriedem gibi: was just wondering if we have the information from the volume meta, if we should store it in the request spec
14:56:43 bauzas mriedem: at the request_spec creation time, then ?
14:56:46 mriedem yes
14:56:50 bauzas well
14:56:56 mriedem the confusion will be,
14:57:01 bauzas if you rebuild ?
14:57:04 mriedem instance.image_ref == None == volume-backed,
14:57:09 gibi copying image_id over from the volume meta to the request_spec is fine by me
14:57:09 bauzas also
14:57:11 mriedem but requestspec.image.id == None....?
14:57:55 mriedem bauzas: if you rebuild a volume-backed instance with a new image, we fail today
14:58:09 jroll cdent: thanks
14:58:23 mriedem bauzas: https://github.com/openstack/nova/blob/master/nova/compute/api.py#L2979
14:59:24 bauzas mriedem: okay, gtk
15:01:18 takashin mriedem: Would you check my reply to you comment in https://review.openstack.org/#/c/513721/ ?
15:04:11 mriedem takashin: done
15:05:18 takashin mriedem: Thanks.
15:08:59 mriedem bauzas: https://bugs.launchpad.net/nova/+bug/1748858
15:09:00 openstack Launchpad bug 1748858 in OpenStack Compute (nova) "allocation_ratio set by aggregate metadata don't work in NUMATopologyFilter" [Undecided,New] - Assigned to yangjie (yang.jie)
15:09:07 mriedem sounds familiar
15:09:30 bauzas mriedem: sec, thinking hard of https://review.openstack.org/#/c/543263/2/nova/scheduler/filters/isolated_hosts_filter.py
15:09:44 bauzas the change isn't trivial
15:10:04 mriedem bauzas: really?
15:10:40 mriedem bauzas: that's basically what it did before it was regressed here https://review.openstack.org/#/c/202678/35/nova/scheduler/filters/isolated_hosts_filter.py
15:10:41 bauzas mriedem: yeah, my guts feel some problem that I still don't see
15:10:52 mriedem maybe you need to go to the bathroom
15:11:07 bauzas probably
15:11:13 bauzas but I'm looking at https://review.openstack.org/#/c/543263/2/nova/scheduler/filters/isolated_hosts_filter.py@43
15:11:16 gibi bauzas: shall we pull the patch from the gate as I +Wd it recently
15:11:42 mriedem if you look at how it worked before https://review.openstack.org/#/c/202678/35/nova/scheduler/filters/isolated_hosts_filter.py
15:11:43 bauzas gibi: I'm not saying we *have* a problem
15:11:50 mriedem the image id was always something it handled as optional
15:11:53 mriedem using dict.get()
15:11:59 bauzas the thing is
15:12:01 mriedem changing to request spec in that patch made it required
15:12:06 bauzas if we say we don't have an image
15:12:17 bauzas then the image isn't isolated, right
15:12:28 mriedem bauzas: that's the point of my email,
15:12:39 bauzas in that case, we should accept host1
15:12:40 mriedem which is, this never did anything for volume-backed instances,
15:12:44 mriedem but that's not the point of my fix right now
15:12:56 mriedem i'm restoring this filter to how it previously worked, before mitaka
15:13:01 bauzas then the doc is invalid
15:13:09 bauzas at least the docstring I mean
15:13:16 bauzas and I'd like to fix that
15:13:19 mriedem so clean that up in a follow up?
15:13:45 bauzas well now that gibi pulled the trigger, /me shrufs
15:13:46 cdent Is there a general opinion about the merit of conditional imports in nova code? I'd like to only import rpc and sqlalchemy_api here https://github.com/openstack/nova/blob/master/nova/config.py#L54-L58 if those conditionals are true. The other option (which might make better long term sense) is to make a different parse_args
15:13:54 bauzas shrugs*
15:14:13 bauzas mriedem: the thing is, I thought you wanna backport ?
15:14:17 bauzas moar fun then
15:14:20 mriedem bauzas: i do
15:14:35 bauzas so, we should backport both your change *and* the doc change
15:14:37 mriedem i don't think the docstring is something that bars us from backporting this fix
15:14:38 bauzas looks sad

Earlier   Later