Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-02-23
16:47:13 mriedem NostawRm: most of us are in a room all week at the PTG so you're likely not going to get a discussion started here today,
16:47:30 mriedem NostawRm: i'd suggest asking the question in the dev mailing list, tag the subject with [nova] so it's filtered properly,
16:47:45 mriedem and explain the use case enough so that others can see if there is a generic way to solve it in nova
16:47:48 mriedem without using hooks
16:48:06 NostawRm ah, I see, didn't know that was going on, ty mriedem
16:49:23 mriedem there are notifications that nova emits when deleting an instance, so you could listen for when nova starts deleting the instance and do your backup, but it gets racy
16:50:57 nic Odds are good that you'll need to subclass/extend the virt driver with your own that does what you want, but that is the definition of "technical debt"
16:51:43 nic If you budget enough for the maintenance coding that will be required, it might turn out OK ;)
16:54:30 openstackgerrit Sujitha proposed openstack/nova master: Change tags to default field in Instance object. https://review.openstack.org/415298
16:54:31 openstackgerrit Sujitha proposed openstack/nova master: Adding tags field to InstancePayload https://review.openstack.org/407228
16:56:09 openstackgerrit Sujitha proposed openstack/nova master: [WIP] Reduce calls to load_tags() to 0 https://review.openstack.org/435146
16:57:52 NostawRm heh, nic, that's likely what I'm looking at now, until I get happy enough with it to submit it for public scrutiny
17:01:50 nic We maintained our own Nova virt driver for years with various custom features for our clients, but as of Liberty we were able to use the stock driver. It's better to not chase code.
17:06:05 NostawRm I would hope that this method would work well enough that others would want it and it would make it into the stock driver, but I'll be a bit honest, I'm not sure of that just from the fact that I've seen a lot of people ask for it and I'm working on custom because its not there
17:07:16 NostawRm Because I've had to read through the code to get an idea of what I want, I can kind of picture how it would go in, but who knows if I could work that out, kind of a nublet
17:25:11 openstackgerrit Luigi Toscano proposed openstack/nova master: More usage of ostestr and cleanup an unused dependency https://review.openstack.org/437536
18:02:52 johnthetubaguy ildikov: you mean this one? https://review.openstack.org/#/c/373203/
18:08:23 openstackgerrit Alex Xu proposed openstack/nova master: [placement] Add Traits related table to the api database https://review.openstack.org/376198
18:30:55 med_ so, at lunch someone mentioned a "--failure" flag that meant nothing to me. I was too gutless to ask at lunch what that meant. Oddly hard to google for. If someone could cluebat me with the obvious I would be forever grateful.
18:31:14 dansmith med_: are you thinking of me talking about --failing?
18:31:18 openstackgerrit Maxim Nestratov proposed openstack/nova master: libvirt: Fix misleading error in Ploop imagebackend https://review.openstack.org/437512
18:31:27 med_ ah yes. so I failed many ways
18:31:31 dansmith med_: when messing with unit tests, you can re-run just the ones that failed last time by doing "tox -epy27 -- --failing"
18:31:39 med_ danke.
18:32:12 med_ dansmith, so now twice in one conference, I've more or less misquote misremembered what you said. I think you should add me to your /ignore list
18:35:42 ildikov johnthetubaguy: yes :)
18:39:43 openstackgerrit Sujitha proposed openstack/nova master: Remove mox from nova/tests/unit/api/openstack/compute/test_create_backup.py https://review.openstack.org/435610
18:52:20 openstackgerrit Lee Yarwood proposed openstack/nova master: Make BlockDeviceMapping object support uuid https://review.openstack.org/242603
18:52:23 openstackgerrit Lee Yarwood proposed openstack/nova master: Add uuid column to BlockDeviceMapping https://review.openstack.org/242602
18:57:28 openstackgerrit Balazs Gibizer proposed openstack/nova master: Improve assertJsonEqual error reporting https://review.openstack.org/428199
19:05:19 openstackgerrit Stephen Finucane proposed openstack/nova master: libvirt: drop MIN_QEMU_NUMA_HUGEPAGE_VERSION https://review.openstack.org/437560
19:05:19 openstackgerrit Stephen Finucane proposed openstack/nova master: libvirt: drop MIN_LIBVIRT_NUMA_VERSION https://review.openstack.org/437561
19:05:20 openstackgerrit Stephen Finucane proposed openstack/nova master: libvirt: drop MIN_LIBVIRT_HUGEPAGE_VERSION https://review.openstack.org/437562
19:05:30 sfinucan mriedem2: ^
19:05:54 oomichi alex_xu: I found some non-good api on tempest https://review.openstack.org/#/c/435630 . https://developer.openstack.org/api-ref/compute/?expanded=check-tag-existence-detail#check-tag-existence should be HEAD, not GET
19:10:20 bauzas yay, Launchpad timeout issue \o/
19:17:10 mriedem sfinucan: yeah so you found out why i didn't tackle the numa stuff right? :)
19:17:13 mriedem b/c of the test gorp
19:18:22 sfinucan mriedem: Aye. Looked tough until I realised having hugepages meant having NUMA support :)
19:18:49 sfinucan *needed NUMA support
19:18:58 sfinucan ...and having NUMA support implicitly meant you had huge page support
19:19:25 sfinucan Hence, bye bye swathes of test code
19:38:47 openstackgerrit Anusha Unnam proposed openstack/nova master: Add unit tests for cinderclient https://review.openstack.org/437575
19:45:32 openstackgerrit Diana Clarke proposed openstack/python-novaclient master: Fix devstack python-novaclient warning https://review.openstack.org/436975
19:57:46 diana_clarke andreykurilin: You at PTG? I'm not seeing existing tests in python-novaclient for stuff like: https://review.openstack.org/436975 But perhaps I'm just missing the obvious.
19:58:33 diana_clarke andreykurilin: Anyhoo, when you have a change, can you please sanity check that and point me in the right direction from tests. I noticed test_auth.py was recently deleted. Thx!
19:58:43 diana_clarke s/from/for
19:59:18 andreykurilin diana_clarke: hi! Yes, I'm at PTG :)
19:59:38 diana_clarke andreykurilin: I thought I saw your hair in one of the pics :)
19:59:53 andreykurilin diana_clarke: heh
20:00:34 miss_hypothermia mriedem: ok, I think I got your point regarding that terminate_connection call
20:01:22 diana_clarke andreykurilin: Anyhoo, the TL;DR is that on stable/ocata devstack, you see this warning for each nova command: http://paste.openstack.org/show/598407/
20:01:53 miss_hypothermia mriedem: as for unreserving the volume in case of Quota error, I got myself confused with where the volume creation happens, when Nova creates it compared to _validate_bdm
20:02:04 miss_hypothermia mriedem: can you by any chance point me to that?
20:02:55 andreykurilin diana_clarke: I'm not at Sheraton right now, will ping you when I will be there or at least the PC will be near me:)
20:03:46 diana_clarke andreykurilin: Okie dokie, no rush. I'm almost done for the day anyway. But I think you're the best person to take a look at this when you have time.
20:03:57 sfinucan lyarwood: http://ricostacruz.com/cheatsheets/rst.html
20:04:29 sfinucan lyarwood: https://github.com/ralsina/rst-cheatsheet/blob/master/rst-cheatsheet.rst
20:08:18 openstackgerrit Lee Yarwood proposed openstack/nova master: Make BlockDeviceMapping object support uuid https://review.openstack.org/242603
20:08:19 openstackgerrit Lee Yarwood proposed openstack/nova master: Add uuid column to BlockDeviceMapping https://review.openstack.org/242602
20:08:19 mriedem miss_hypothermia: nova creates the volume in nova/virt/block_device.py for the image/snapshot/blank drivers
20:08:19 openstackgerrit Lee Yarwood proposed openstack/nova master: Add attachment_id to BlockDeviceMapping https://review.openstack.org/437597
20:09:04 mriedem miss_hypothermia: in the cinder channel smcginnis said that the volume drivers in cinder should not fail on terminate_connection if the init never happened
20:09:48 mriedem so, the only difference in behavior is that now in boot from volume when you provide a volume to nova, if something fails the volume is going to be 'locked' by nova until the instance is deleted
20:10:05 mriedem 'locked' meaning it's going to be in 'attaching' status until the admin force detaches/unreserves it, something like that
20:10:15 mriedem we'd need that in a release note
20:12:08 gibi jo
20:12:11 gibi sorry
20:14:53 openstackgerrit Jay Pipes proposed openstack/nova master: placement: ironic inventory reporting https://review.openstack.org/437602
20:15:10 jaypipes dansmith, mriedem, cdent: ^
20:15:32 cdent ✔
20:23:48 dansmith jaypipes: that looks a lot like what I was thinking
20:23:55 dansmith jaypipes: I shall return to it when I have full brains
20:24:09 dansmith melwitt: so..........
20:24:31 dansmith melwitt: target_cell doesn't target mq, only database
20:24:47 dansmith melwitt: which we discussed previously as handled in the rpc layer and I think re-discussed recently as also putting in target_cell
20:25:15 dansmith melwitt: so now I'm hitting a problem because I need to call conductor's rpcapi for an instance, and have it go to the right cell's conductor
20:25:30 dansmith melwitt: and since we don't have that in there, I go by default to superconductor and fail
20:25:39 dansmith which I think is further justification for my opinion that target_cell should do mq too
20:25:41 melwitt dansmith: I know. it's on my todo list to add it back to target_cell and move the caching to there, maybe
20:25:42 dansmith thoughts?
20:25:50 dansmith okay
20:26:07 melwitt dansmith: yeah, I agree. maybe just a small patch to start to let it be possible, then the move of the cache and everything will be a bigger haul
20:26:18 dansmith melwitt: yeah, I shall do that thing then
20:27:28 melwitt dansmith: I can put up a change based on what I had way back when, before all of it moved to compute/rpcapi
20:27:40 melwitt it's an earlier PS of that patch
20:28:09 dansmith melwitt: okay would you rather do that than me just do it?
20:30:16 melwitt dansmith: you can do it, I didn't remember what there was to it. turns out it was only this https://review.openstack.org/#/c/274955/12/nova/context.py
20:30:36 dansmith melwitt: ack, yeah, thanks
20:31:08 melwitt I'm not sure if we need the special-casing of 'none:///'. that was to avoid doing something for cell0
20:31:44 spiantino hey all - i found some logs from november discussing adding the ability to include "<kvm><hidden state='on'/><kvm>" to the libvirt domain via nova
20:32:16 spiantino is there an implementation of that? or any advice on how to hackily get that field included in a VM? without it NVidia GPUs don't work in a VM
20:35:24 spiantino mgariepy: were you ever able to get that option added to driver.py?
20:35:40 spiantino happy to submit the pr
20:39:49 oomichi alex_xu: sdague: johnthetubaguy: can you take a look at https://review.openstack.org/#/c/406377 ?
20:40:07 oomichi that could fix potential server internal error
20:40:55 miss_hypothermia mriedem: that seems like a fair deal, I mean have the volume locked
20:41:42 miss_hypothermia mriedem: so in this sense I will remove the compute service version bump and the additional unreserve call and add a release note to describe the change
20:42:06 mriedem miss_hypothermia: ok
20:42:10 mriedem i think i'm ok with that :)

Earlier   Later