| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-06-11 | |||
| 14:09:58 | artom | Err, dansmith ^^ | |
| 14:10:21 | dansmith | artom: ack, I feel like an ass complaining about the hardest problem in computer science, but.. | |
| 14:10:36 | dansmith | might be that I'm just feeling my base state | |
| 14:11:04 | artom | dansmith, in terms of where to put them... get_virt_ in to libvirt/driver.py and get_manager_ in the compute manager? They're 2 sides of the same logic, so I kinda think they should go in the same place, one that's callable by both libvirt and the compute manager, so...? | |
| 14:11:11 | artom | dansmith, no, naming is important, it's fine :) | |
| 14:12:07 | dansmith | artom: virtapi is the wrong place for that, IMHO. I'm not opposed to them being adjacent | |
| 14:12:24 | kashyap | efried: Thanks for not minding my nit-pick! New PS looks good to me | |
| 14:12:47 | kashyap | [With the test failures fixed.] | |
| 14:13:00 | efried | thanks. Any idea wtf is up with the doubled words hacking test?? | |
| 14:13:02 | artom | dansmith, fair enough about virtapi being wrong, but... what else do we have? We talked about putting them in utils somewhere | |
| 14:13:29 | kashyap | efried: Which are those? | |
| 14:14:45 | efried | kashyap: This morning I saw py27 failing on several totally unrelated patches on hacking test_check_doubled_words | |
| 14:14:50 | efried | e.g. http://logs.openstack.org/49/664449/1/check/openstack-tox-py27/4ad64aa/testr_results.html.gz | |
| 14:16:13 | dansmith | artom: yeah, I dunno. utils feels a little generic.. what about in neutronapi or thereabouts? | |
| 14:16:23 | efried | nothing about that test has changed. No new hacking rules have been added recently (months). The only thing around hacking that's changed is when stephenfin updated hacking reqs, but that was like 6 weeks ago. | |
| 14:16:44 | stephenfin | efried: Scroll up | |
| 14:16:45 | artom | efried, https://bugs.launchpad.net/nova/+bug/1804062 I think | |
| 14:16:46 | openstack | Launchpad bug 1804062 in nova (Ubuntu Eoan) "test_hacking fails for python 3.6.7 and newer" [High,Triaged] | |
| 14:16:55 | kashyap | efried: No idea about that one, afraid. | |
| 14:16:57 | stephenfin | ||
| 14:17:01 | stephenfin | efried: ^ | |
| 14:17:06 | stephenfin | (or down) | |
| 14:17:15 | stephenfin | I'm looking at it | |
| 14:17:18 | efried | stephenfin: But all the failures I saw today were in py27 | |
| 14:17:48 | artom | efried, <lyarwood> artom: no missed that thanks, however this is on 2.7 oddly enough. | |
| 14:17:49 | openstack | Launchpad bug 1804062 in nova (Ubuntu Eoan) "test_hacking fails for python 3.6.7 and newer" [High,Triaged] | |
| 14:17:49 | artom | ||
| 14:17:50 | stephenfin | Hmm, Canonical hardly backported the change, did they? | |
| 14:18:12 | artom | dansmith, hrmm, neutronapi does make sense | |
| 14:18:23 | artom | And it's something both compute manager and driver use already | |
| 14:18:27 | artom | *libvirt driver | |
| 14:18:34 | dansmith | artom: mmhmm :) | |
| 14:19:02 | artom | It's still kinda weird that suddenly neutronapi will understand about compute manager and libvirt | |
| 14:19:08 | artom | But... that was going to be weird regardless | |
| 14:19:13 | dansmith | well, it doesn't reall,y | |
| 14:19:31 | dansmith | it just declares "late" and "early" events or something | |
| 14:19:42 | dansmith | it doesn't need to be revert specific, or even compute/virt specific | |
| 14:19:44 | artom | dansmith, ah I see - "bind-time" vs "plug-time" | |
| 14:19:49 | dansmith | something like that yeah | |
| 14:19:56 | artom | Well, except it's *very* revert specific | |
| 14:20:12 | artom | Because IIUC it doesn't happen unless the instance is present on the host, shut off | |
| 14:20:15 | dansmith | the fact that it's revert specific is a nova detail | |
| 14:20:27 | dansmith | it's specific to re-binding without changing the host, or whatever the eff it is | |
| 14:20:58 | artom | Updating the binding to a host that already has the instance | |
| 14:20:58 | efried | stephenfin: looks like it may have been backported, yeah. | |
| 14:21:20 | dansmith | artom: there's also nova/network/somethingelse | |
| 14:21:24 | dansmith | if you prefer | |
| 14:21:26 | prometheanfire | nova tests flakey? | |
| 14:21:42 | efried | prometheanfire: We think we're hitting https://bugs.launchpad.net/nova/+bug/1804062/comments/4 | |
| 14:21:43 | openstack | Launchpad bug 1804062 in nova (Ubuntu Eoan) "test_hacking fails for python 3.6.7 and newer" [High,Triaged] | |
| 14:21:49 | artom | dansmith, events.py? | |
| 14:21:56 | artom | neutronapi is fine | |
| 14:21:58 | dansmith | artom: but the detail here is pretty neutron specific, as I understand it | |
| 14:22:03 | artom | dansmith, oh yeah | |
| 14:22:04 | efried | stephenfin: I'm going to propose a patch to disable in 2.7.15, unless you're already on that. | |
| 14:22:27 | stephenfin | I was just going to fix the test | |
| 14:22:35 | efried | stephenfin: oh, what's the fix? | |
| 14:22:45 | stephenfin | No idea. I'm working on that :) | |
| 14:22:52 | dansmith | artom: if you meant nova/network/events.py as a new modiule.. maybe but that's pretty heavy.. I dunno, I don't care that much, I just don't want it in virtapi | |
| 14:22:53 | stephenfin | Wait, is this fixed in 2.7.16? | |
| 14:22:59 | stephenfin | i.e. is it a Python bug? | |
| 14:22:59 | dansmith | artom: I bet mriedem has an opinion | |
| 14:23:13 | dansmith | artom: and it's probably contrary to my opinion | |
| 14:23:35 | artom | dansmith, mriedem, well you guys can oil-wrestle it out | |
| 14:23:42 | artom | I'm gonna go write code ;) | |
| 14:23:45 | dansmith | artom: I think he likes to disagree with me because he thinks it makes him look smart, whereas I like to think agreeing with him makes _me_ look smart | |
| 14:25:27 | efried | cdent: Do you understand the issue underlying https://bugs.launchpad.net/nova/+bug/1804062/comments/4 ? | |
| 14:25:29 | openstack | Launchpad bug 1804062 in nova (Ubuntu Eoan) "test_hacking fails for python 3.6.7 and newer" [High,Triaged] | |
| 14:27:10 | efried | (whoah, that's a pretty serious regex in that hacking check) | |
| 14:27:29 | prometheanfire | efried: thanks for the pointer | |
| 14:28:08 | cdent | weird. I have no memory of any of this | |
| 14:29:11 | sean-k-mooney | cdent: efried do vmware or powervm have a similar feature to libvirt auto converge livemigation optio where the vm is slowed if the migration is taking a long time to allow it too complete | |
| 14:29:56 | efried | sean-k-mooney: I don't recall any such thing, but that doesn't mean it doesn't exist. | |
| 14:30:27 | sean-k-mooney | i dont think any driver other then libvirt support it in nova but i was wondering if it was supported by other hyperviors in general | |
| 14:30:36 | cdent | efried: I don't know if the underlying issue much matters? Basically if we take off the skip and things work, groovy, if not, we still stuck? Or are you after "can we fix the regex"? | |
| 14:30:38 | efried | actually, I think powervm live migrate might actually quiesce the vm while it's migrating | |
| 14:31:23 | cdent | sean-k-mooney: the mysteries of the DRS are beyond me. If you can think it, it probably does it if you set the right setting. | |
| 14:31:49 | kashyap | yonglihe: I've also asked the QEMU post-copy migration author/maintainer to look at the spec, too. (I myself need to re-look at it, just been buried) | |
| 14:31:59 | kashyap | (s/need/"wanted to"/) | |
| 14:32:10 | sean-k-mooney | cdent: ya i assumed that its proably done under the hood in vspher or whatever | |
| 14:32:15 | efried | cdent: Well, stephenfin is looking for a way to fix whatever we're doing in the test, but so far we're stymied on that. If it's not possible (or if it takes us longer than a few more minutes to figure out) I'd like to just disable the test. But since you touched it, and your code change implies some kind of understanding of the root cause of the breakage, I though I would ask you if you know what's going on there. | |
| 14:32:52 | cdent | my understanding was based on chasing many links through the interwebs, not understanding the tokenizer | |
| 14:33:06 | efried | okay | |
| 14:34:57 | efried | I'm trying to repro on bionic. Not happening at 2.7.15rc1 | |
| 14:35:18 | efried | aha, but does reproduce on latest | |
| 14:35:29 | stephenfin | yeah, I can't reproduce on Fedora 20 with 2.7.16 | |
| 14:35:30 | efried | which lists self as 2.7.15+ -- not helpful :( | |
| 14:42:21 | stephenfin | or 3.6.8. Hmm | |
| 14:44:23 | stephenfin | I always forget how damn big the nova repo is | |
| 14:44:37 | stephenfin | Definitely should have shallow cloned | |
| 14:52:54 | kashyap | sean-k-mooney: In that PostCopy spec, this naming sounds scary to me: "*_permit_reboot_risk" | |
| 15:03:53 | efried | stephenfin: It's looking like it's only broken at 3.6.7 (some releaselevels - at least 'final') and 2.7.15 (likewise). So we could skip those and move on with life. | |
| 15:04:10 | efried | oh, never mind, I just reproduced at 3.6.8 'final'. | |
| 15:04:13 | stephenfin | If that's the case, yup, let's do that | |
| 15:04:14 | stephenfin | damn | |
| 15:05:07 | efried | stephenfin: what's your 3.6.8 where it wasn't reproducing? | |
| 15:05:17 | efried | stephenfin: and did you remember to turn off the skip decorator? :P | |
| 15:05:39 | stephenfin | python36-3.6.8-6.fc30.x86_64 | |
| 15:05:49 | stephenfin | That only applies to 2.7, no? | |
| 15:06:00 | stephenfin | I've got it reproduced in an Ubuntu VM | |