Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-05-02
15:28:15 efried stephenfin: You gonna go around and change all the other deprecations in the docs? There were about a dozen I counted yesterday.
15:31:09 stephenfin efried: I can but where?
15:31:56 efried stephenfin: I went and looked for them yesterday. I can go find 'em again.
15:32:06 efried stephenfin: But I guess if I'm gonna do that, I may as well fix 'em myself :)
15:33:19 shewless Hi there. I have an instance which is allocated 2GB of RAM but when I look at the qemu process it looks like it's comsuming closer to 8GB of RAM. This is a huge discrepency. Can anyone help me figure out what's going on?
15:33:34 shewless It kind of looks like we are seeing an extra 6GB of memory usage (over what's been allocated) PER instance?
15:35:55 stephenfin efried: That...would make sense, yes :)
15:36:10 shewless I thought that in general a qemu process would use less or equal to what's been provisioned for that instance. Is that wrong?
15:38:30 efried # rgrep -i deprecated doc/source | awk -F: '$1 ~ /rst$/'
15:38:30 efried stephenfin: This produced 29 results, which will need to be inspected manually to see if a .. deprecated:: marker is actually appropriate:
15:42:31 openstackgerrit Matt Riedemann proposed openstack/nova master: Remove IronicHostManager and baremetal scheduling options https://review.openstack.org/565805
15:42:32 openstackgerrit Matt Riedemann proposed openstack/nova master: Remove [scheduler]/host_manager config option https://review.openstack.org/565806
15:42:35 mriedem jroll: ^
15:42:44 jroll woot
15:43:04 sahid shewless: we know that there are some kind of overheads depending of the number of vCPUs, devices... but 6GB looks crazy, perhaps you should ask #virt in OFTC servers with the exact arguments you are passing to QEMU
15:43:53 shewless sahid thanks I'll check there
15:45:02 openstackgerrit sahid proposed openstack/nova master: network: update pci request spec to handle trusted tags https://review.openstack.org/458820
15:45:03 openstackgerrit sahid proposed openstack/nova master: libvirt: configure trust mode for vfs https://review.openstack.org/458514
15:45:04 openstackgerrit sahid proposed openstack/nova master: pci: don't consider case when match tags specs https://review.openstack.org/565808
15:50:54 efried stephenfin: That deprecated role ain't especially pretty IMO.
15:51:12 stephenfin In what regard?
15:51:59 efried let's see if your delta has rendered yet...
15:53:11 efried stephenfin: http://logs.openstack.org/67/565367/9/check/build-openstack-sphinx-docs/c46fab1/html/cli/nova-consoleauth.html
15:53:28 efried stephenfin: I guess I was expecting a box, like .. note::
15:53:33 openstackgerrit Jay Pipes proposed openstack/nova-specs master: update add-consumer-generation to focus on API https://review.openstack.org/565565
15:53:34 efried stephenfin: It's aight, I suppose.
15:53:34 sahid cfriesen__ it's implemented
15:56:45 stephenfin efried: Yeah, that's openstackdocstheme. Wonder what the reasoning for that was
15:57:27 efried stephenfin: I'm sure it's related to it never being used, but I wonder if it's cause or effect :)
15:59:02 stephenfin efried: Could be. I thought oslo.config used it but it actually uses '.. warning' too. I should fix that https://docs.openstack.org/nova/queens/configuration/config.html
15:59:15 openstackgerrit Aditya Vaja proposed openstack/nova master: remove IVS plug/unplug as they're moved to separate plugin https://review.openstack.org/534371
16:05:04 openstackgerrit Eric Fried proposed openstack/nova master: Use .. deprecated:: theme for deprecations https://review.openstack.org/565822
16:05:06 efried stephenfin: Here it is ^ but I'll probably abandon it, unless we want to fix the theme to make it prettier.
16:10:08 jroll jaypipes: do you know if we need to report these as zero, or can we just drop them all? https://github.com/openstack/nova/blob/master/nova/virt/ironic/driver.py#L764
16:13:04 evrardjp mriedem: thanks for the patch in OSA :)
16:14:35 jroll jaypipes: ah, I think unit tests answered that for me, never mind. :)
16:16:50 mriedem evrardjp: np
16:20:03 arvindn05 mriedem: replied back to the thread on rebuild with image traits. Will look for further update from melwitt or dansmith
16:20:55 mriedem evrardjp: s/OSA/literally everything/ +(status:open+OR+status:merged)
16:20:59 mriedem https://review.openstack.org/#/q/topic:exact-filters+(status:open+OR+status:merged)
16:22:26 evrardjp mriedem: I haved guessed that ;)
16:22:32 evrardjp mriedem: but still thank you.
16:22:40 evrardjp :D
16:22:46 arvindn05 mriedem: i think we are on the same page in case we choose to reject the request at the API layer. the spec had the same intent
16:23:03 EmilienM mriedem: thanks for https://review.openstack.org/#/c/565824/ - I took over, just see my comment for the FYI
16:24:07 arvindn05 > If the image's required traits have changed from the original image, we can reject the rebuild request. Effectively, since the qualitative requirements have changed the user should create a new instance.
16:24:08 arvindn05 https://review.openstack.org/#/c/560718/6/specs/rocky/approved/glance-image-traits.rst - line 100. I will make it more clear like i did in the ML
16:25:51 mriedem EmilienM: ack thanks
16:26:40 mriedem EmilienM: i suspect the depends-on is backward anyway for any of these deployment projects that gate on ironic, which tripleo does?
16:26:58 mriedem but i didn't want the nova change depending on 20 random other deployment projects
16:27:04 EmilienM mriedem: yeah it's fine
16:38:26 openstackgerrit Arvind Nadendla proposed openstack/nova master: update scheduler to use image-traits https://review.openstack.org/560596
16:46:34 openstackgerrit Jim Rollenhagen proposed openstack/nova master: Ironic: report 0 for vcpus/memory_mb/disk_gb resources https://review.openstack.org/565841
16:46:48 jroll mriedem: one more in your ironic chain for you ^
16:51:18 mriedem question inline
16:53:24 jroll ah thanks
16:56:42 mriedem dansmith: so for this CLI to fix allocations in placement for instances scheduled using the CachingScheduler, i'm trying to think where that should live, since none of the nova-manage subcommands really fit that; we could do something like,
16:56:47 mriedem a) nova-manage placement ...
16:56:54 mriedem b) placement-manage ...
16:56:57 mriedem c) throw it in osc-placement
16:57:09 mriedem although (c) isn't really meant for something like this
16:57:20 dansmith yeah, it's a pretty oddball thing
16:57:30 mriedem we're fixing allocations for nova instances,
16:57:33 dansmith it's also a placement client sort of thing, so it's weird to have it in nova-manage,
16:57:36 mriedem so thinking: nova-manage placement fix_allocations
16:57:40 dansmith but placement-manage would be the opposite
16:58:06 mriedem right, we need the instance uuid to find it's host/node to get the node uuid to find the resource provider it should have allocations against
16:58:13 dansmith mriedem: maybe we need a generic nova-manage group, like "nova-manage fix-our-mess" or "nova-manage maintenance ..."
16:58:17 mriedem so doing instance lookups from a placement-manage is weird
16:58:56 dansmith could use that generic group for syncing aggregate membership, fixing cinder attachments (historically)
16:59:25 mriedem jaypipes: where were you going to throw the nova-manage sync CLI for aggregates?
16:59:37 dansmith mriedem: wherever you tell him to
16:59:38 dansmith :P
17:00:29 mriedem was just wondering if his gears were grinding on that yet
17:03:42 openstackgerrit Matt Riedemann proposed openstack/nova master: Remove [scheduler]/host_manager config option https://review.openstack.org/565806
17:03:50 mriedem i'm partial to 'nova-manage placement ...'
17:03:59 mriedem but don't care too much
17:04:02 mriedem gonna get lunch
17:04:20 dansmith melwitt: mriedem: tssurya: I've been completely covered up in other stuff for the last week.. I have nothing cellsy to talk about, unless the cern peeps have things to report?
17:04:35 dansmith mriedem: well, I just don't know that we'll have enough in there to justify the group, but whatever you want
17:05:13 mriedem dansmith: we'll have at least 2 things, this and jay's sync aggs thing
17:05:20 mriedem i don't have any cells stuff to talk about
17:07:07 tssurya dansmith: not really except that we moved to cellsv2, and basic operations take more time than for cellsv1 , we are trying to optimize whatever possible, but since belmiro won't be around today as its a Euro unfriendly meeting timing, we can give the overall experience in the next week's meeting
17:07:56 dansmith tssurya: okay, yeah would be good to hear an analysis of what takes longer so we can think about what to do
17:08:27 tssurya dansmith: yep
17:09:00 dansmith cellsv1 cheated with cache coherency, so not surprising that some things will take longer to be correct ;)
17:09:43 tssurya dansmith: haha, yea I get it but querying to all 70 DBs versus top cell having all info is also the toll maybe
17:09:59 dansmith tssurya: you guys are on pike yeah?
17:10:10 tssurya dansmith: skipped pike, moved straight to queens :)
17:10:17 tssurya from ocata
17:10:38 dansmith oh
17:11:02 dansmith yeah, so you should have the parallel instance listing,
17:11:13 dansmith although we do iterate the cells in other places where we could optimize
17:11:37 dansmith tssurya: https://review.openstack.org/#/c/509003/
17:11:41 dansmith that would help you more than most
17:11:56 jaypipes mriedem: nova-manage placement ... works for me.
17:11:57 tssurya dansmith: exactly, we are trying to optimize such things, however we also had some DB side issues, some being slower than the others,
17:12:36 tssurya dansmith: oh I have never come across that patch
17:12:40 tssurya thanks
17:13:04 dansmith tssurya: that would make us only query the cells a particular tenant has instances in, which I think in your case would be a big win.. like going from 70 to 2 or something

Earlier   Later