Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-12-04
17:00:58 edleafe mriedem: ^^ there ya go
17:01:01 mriedem thanks
17:01:36 efried cfriesen The other thing I do when I'm working on a series is: source .tox/pep8/bin/activate to get into the pep8 virtualenv, and leave that window open. Then from within that window I'll do: git diff --name-only HEAD~1 | grep '\.py$' | xargs flake8
17:01:51 efried cfriesen HEAD~{whatever-number-here} to go back that many patches.
17:02:39 emccormick Hey, if anyone has a sec for a possibly dumb question: Is there a reason that it's required to have the instance type in the domain xml match the instance type in nova.conf?
17:03:25 emccormick I ask because I'm working on migrating backends from local to rbd, and while running instances continue to function after the change to nova.conf, and can also be migrated, they cannot be rebooted.
17:03:44 emccormick just wondering if that's something that could be changed or if it would break things horribly
17:04:24 efried cfriesen The only caveat to the permanent-venv deal is if you need to rebuild the venv for whatever reason; sometimes running via tox will figure that bit out for you without needing to use -r; but sitting in the venv you get nuffink.
17:05:03 edleafe efried: I use aliases to do roughly the same thing
17:05:06 cfriesen efried: that's a useful trick and a good caveat.
17:05:34 efried edleafe Aliases for what, running flake8 in the venv?
17:06:02 openstackgerrit Chris Friesen proposed openstack/nova master: unreserve volumes after scheduling has failed https://review.openstack.org/525253
17:06:17 edleafe efried: like this:
17:06:18 edleafe alias ptt='source .tox/py27/bin/activate; python -m testtools.run'
17:06:28 cfriesen ^ we ran into this while testing Pike, looks like it's still an issue in master
17:06:39 edleafe I have different aliases for functional, py3, etc.
17:06:41 efried edleafe Ah, gotcha.
17:08:34 openstackgerrit Stephen Finucane proposed openstack/nova master: Resolve review comments from I41b03a347 https://review.openstack.org/525255
17:08:34 openstackgerrit Stephen Finucane proposed openstack/nova master: zuul: Move legacy jobs to project https://review.openstack.org/514309
17:08:48 cfriesen The issue that the patch above ^^^ fixes is that if you do boot-from-vol and then fail scheduling, currently the volume remains reserved and is stuck in the "attaching" state.
17:09:50 efried mriedem The _ensure_resource_provider error handling cherry-pick for pike is looking green now (https://review.openstack.org/#/c/524618/) -- will proceed to backport to ocata unless some reason not to?
17:11:50 mriedem go ahead
17:12:26 openstackgerrit Surya Seetharaman proposed openstack/nova master: update_cell allows more than once cell to have the same db/transport url https://review.openstack.org/518556
17:16:23 efried urgh, merge conflict (which I guess isn't surprising). Will hit it after "lunch".
17:26:46 dmsimard kashyap: o/ mriedem said you could help with a bug we're trying to nail down. Have you gotten familiar with https://bugs.launchpad.net/nova/+bug/1735823 ?
17:26:48 openstack Launchpad bug 1735823 in OpenStack Compute (nova) "Nova can hang when creating a VM with disk injection" [Medium,In progress] - Assigned to Matt Riedemann (mriedem)
17:27:35 dmsimard tl;dr, we thought we ran libguestfs with force_tcg for disk injection but we didn't, so in some environments (such as OVH nodepool cloud regions) with broken nested virtualization, this failed horribly
17:28:19 dmsimard We believe the fix is here: https://review.openstack.org/#/c/524748/ -- however, when testing it here: https://review.openstack.org/#/c/524750/, we're seeing increased failures.. and outside of OVH too.
17:29:14 dmsimard Before the fix (at OVH) http://logs.openstack.org/46/523646/1/check/legacy-tempest-dsvm-neutron-full-centos-7/5bf092c/job-output.txt#_2017-11-29_03_02_38_031436 -- After the fix (at RAX): http://logs.openstack.org/50/524750/1/check/legacy-tempest-dsvm-neutron-full-centos-7/a7f051e/job-output.txt.gz#_2017-12-04_15_25_13_211317
17:29:45 dmsimard I did a recheck in https://review.openstack.org/#/c/524750/ to make sure it wasn't a flap and it does look like it breaks things.
17:32:07 stephenfin efried: Interesting little way to speed up tox - 'touch nova.egg-ingo'
17:32:41 stephenfin I read the tox source a while back to figure out why, but it skips the "check if any dependencies are out-of-date" step, which half of what takes so long
17:32:50 cdent mriedem: I’m thinking that https://bugs.launchpad.net/nova/+bug/1734491 may have back port potential?
17:32:51 openstack Launchpad bug 1734491 in OpenStack Compute (nova) "placement keystonemiddleware_authtoken ignores OS_PLACEMENT_CONFIG_DIR" [Undecided,In progress] - Assigned to Chris Dent (cdent)
17:32:57 stephenfin Probably a bug in tox but I haven't figured out why yet
17:36:18 jaypipes efried: let's go with get_provider_uuids() then.
17:38:16 ericyoung Any idea when the patch for ScaleIO ephemeral/image volume support will bubble to the top of the review list? https://review.openstack.org/#/c/407440/
17:40:58 cdent jaypipes: the point on the jsonschema comment was: webob is passing a string to jsonschema, one that looks like an int, but is not an int
17:41:08 cdent just int will _not_ work
17:41:58 jaypipes cdent: if that's the case, then we shouldn't have "integer" in there at all, since we'll never be supplying actual numbers.
17:42:21 cdent unless we change webob out from under it (this has been discussed further up the review stack)
17:42:47 cdent I can make it just string if you like, but the way it is now is to be as flexible yet accurate as possible
17:43:03 cdent it is also what the nova api does
17:43:06 cdent so I copied it
17:44:15 jaypipes cdent: meh...
17:44:43 cdent jaypipes: much meh made merry many months, mostly
18:11:01 mriedem edleafe: ok i've made it through the code parts of https://review.openstack.org/#/c/510159/
18:11:11 mriedem edleafe: some things that need to be cleaned up in there, but we could arguably do those in a follow up,
18:11:17 mriedem i will go through test changes after lunch
18:15:34 edleafe mriedem: a follow-up is fine.
18:26:39 openstackgerrit Merged openstack/nova master: XenAPI: get vGPU stats from hypervisor https://review.openstack.org/512965
19:43:12 efried stephenfin That's a cool tip, thanks. I doubt it would be considered a "bug" since you're effectively hacking a metadata marker tox uses internally for exactly that purpose.
19:59:44 mriedem edleafe: ok, found a few more missing tests in https://review.openstack.org/#/c/510159/
19:59:52 mriedem edleafe: should probably just add those in there rather than a follow up
20:00:04 mriedem the comment updates could all be done in a follow up
20:07:17 mriedem moving onto the next change in the series
20:12:50 efried edleafe jaypipes What about uuid_iter() ? (And make it an iterable)
20:13:30 efried Ah, I think I remember why I didn't do that originally - don't want the caller holding (and possibly abandoning) the lock.
20:20:09 openstackgerrit Eric Fried proposed openstack/nova master: ProviderTree.get_provider_uuids() https://review.openstack.org/520243
20:20:16 efried edleafe jaypipes ^
20:28:28 edleafe mriedem: if I have to change that patch, might as well do 'em all there.
20:30:14 jaypipes efried: +2 from me. nice.
20:30:25 efried jaypipes Thanks
20:30:25 mriedem edleafe: well, if you're going to update all of those comments,
20:30:41 jaypipes efried: though you didn't update the make _provider really private patch.
20:30:45 mriedem you should then do that in the change earlier in the series that actually makes the scheduler drivers return Selection objects back up to the SchedulerManager
20:30:54 efried jaypipes Uh, was I supposed to?
20:31:14 jaypipes efried: nah, just figured you would... rebase the entire thing to get em up to the same rev
20:31:38 jaypipes efried: since gerrit shows the get_provider_uuids() based on an old rev of the dep patch
20:31:38 efried jaypipes Hum, that's weird; I wouldn't have thought I needed to do that.
20:31:43 efried yeah - why?
20:31:49 efried I rebased both of 'em earlier.
20:31:50 mriedem edleafe: specifically https://review.openstack.org/#/c/495854/ made the comments invalid
20:32:00 mriedem https://review.openstack.org/#/c/510159/ just needs more tests
20:32:16 edleafe mriedem: ok
20:32:22 efried jaypipes ohh, I know what happened.
20:32:23 mriedem edleafe: comments in https://review.openstack.org/#/c/516707/ too - i think we need to not touch anything in the compute manager/claims/RT code there in that patch
20:32:35 efried I manually edited two characters in the commit message of the bottom patch.
20:33:11 openstackgerrit Eric Fried proposed openstack/nova master: ProviderTree.get_provider_uuids() https://review.openstack.org/520243
20:33:27 efried jaypipes ^ simple rebase from gerrit, kept your +2. Thanks for pointing it out.
20:36:12 jaypipes efried: coolio.
20:40:39 openstackgerrit Dan Smith proposed openstack/nova master: Fix cellsv1 messaging test https://review.openstack.org/525304
20:40:39 openstackgerrit Dan Smith proposed openstack/nova master: Fix cellsv1 build request deletion if multiple api databases are used https://review.openstack.org/523187
20:41:07 efried dansmith Got some review bandwidth for these two? https://review.openstack.org/#/c/523932/ https://review.openstack.org/#/c/520243/ They're pretty simple.
20:41:49 dansmith efried: no, and also, you're doing it wrong
20:42:01 efried Well, that figures. What am I doing wrong?
20:42:11 dansmith efried: acting like I have review bandwidth will never work.. playing to my guilt usually will
20:42:15 dansmith watch and learn:
20:42:48 dansmith dansmith: you haven't reviewed any of my patches in a while and I'm starting to feel like you're a bad person. prove me wrong and review these easy ones I have selected for you...
20:43:00 efried okay, let me give that a try:
20:43:15 mriedem edleafe: some comments that can be updated in https://review.openstack.org/#/c/511357/ when you update the series
20:43:36 efried dansmith: reviewing these simple patches will not only further the cause of Very Important blueprints, it will make you more handsome and popular.
20:43:50 dansmith efried: weak, but passable
20:43:57 efried Dangit, yeah, there's almost no guilt in there at all.
20:44:12 dansmith but it plays to other things that work for me, so..
20:44:37 mriedem passive aggressive works too
20:44:57 efried Like this: dansmith I think it's great how much time you have for cells stuff.
20:45:05 mriedem no
20:45:09 efried damn
20:45:13 efried I really suck at this.

Earlier   Later