| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-09-22 | |||
| 18:47:16 | sdague | ? | |
| 18:47:28 | sdague | but it would do magic for guests that had been booted with old versions? | |
| 18:47:32 | cburgess | Correct | |
| 18:47:36 | cburgess | To all of the above. | |
| 18:47:42 | mriedem | i think it's a pretty safe assumption. i think most of the code assumes that anything we do with qemu on the host is the same version that libvirt is using. | |
| 18:47:42 | sdague | ok, in which case, this code would totally work for you | |
| 18:48:25 | sdague | and, honestly, anything more esoteric than what you are doing I would expect people to have to hack the code | |
| 18:48:34 | sdague | s/are/were/ | |
| 18:49:42 | sdague | I want to be careful here making the main path potentially more fragile for an edge world that we aren't really sure exists anywhere | |
| 18:49:53 | cburgess | OK | |
| 18:49:59 | cburgess | Just felt... wrong. | |
| 18:50:29 | sdague | cburgess: I did have a pause until I looked up that qemu-img and qemu come from the same source tarball | |
| 18:50:48 | sdague | in which case I'm not going to assume you can mix / match | |
| 18:50:54 | cburgess | You can.. | |
| 18:50:58 | cburgess | Its perfectly legal to do so. | |
| 18:51:07 | cburgess | It just happens to come in the same source build. | |
| 18:51:12 | sdague | the qemu team tells you you can? | |
| 18:51:13 | cburgess | Well legal-ish. | |
| 18:51:17 | sdague | and that they'll support it? | |
| 18:51:33 | sdague | like "it didn't blow up for us" | |
| 18:51:37 | cburgess | This is the first time I'm aware of a flag like this that requires synced version. | |
| 18:51:38 | sdague | is not the same as upstream supported | |
| 18:51:51 | sdague | cburgess: sure, that caught us off guard for sure | |
| 18:52:09 | cburgess | sdague I don't know what the "official" support policy is from upstream. | |
| 18:52:28 | cburgess | sdague I only know what the "make it work" policy is. :P | |
| 18:52:36 | sdague | if you find a piece of evidence that the qemu team says "+1 we support mix and match" I'd change my tune :) | |
| 18:52:57 | openstackgerrit | Merged openstack/nova master: Remove method "_get_host_ref_from_name" https://review.openstack.org/504796 | |
| 18:52:58 | sdague | but until then, this is pretty robust version detection, a lot more than parsing strings of stderr | |
| 18:53:37 | openstackgerrit | Merged openstack/nova master: enhance api-ref for os-server-external-events https://review.openstack.org/504263 | |
| 18:53:38 | sdague | and we could always change it later if people really needed the mix & match | |
| 18:53:44 | sdague | and came forward | |
| 18:54:10 | cburgess | sdague I removed my objection. | |
| 18:54:11 | sdague | cburgess: while you are at terminal - https://review.openstack.org/#/c/454323/ - live snapshot by default | |
| 18:54:15 | sdague | cburgess: cool | |
| 18:54:24 | mriedem | we could parse the error and retry the command with the new flag, but... | |
| 18:54:29 | mriedem | error message parsing isn't fun either | |
| 18:54:31 | sdague | mriedem: yeh | |
| 18:54:36 | cburgess | I agree... its ugly. | |
| 18:55:05 | sdague | and given that we got a new required flag we were not expecting, I'm really wary of considering error messages or even version string dumps contract | |
| 18:55:43 | cburgess | sdague Yeah so to be clear.. we never deployed libvirt 1.2.2 because it was bad. We just skipped that version which is why we never hit that live snapshot bug. But yeah we have run with it since we merged the code upstream. | |
| 18:56:01 | sdague | cburgess: ok, cool | |
| 18:56:39 | cburgess | sdague rmk added that code... umm... 4 or 5 years ago and we have been running with it enabled ever since. | |
| 18:57:30 | sdague | cburgess: yeh, I knew you all had been using that for a long time, I just realized at PTG we hadn't come around and flipped that back since getting past 1.2.2 | |
| 18:58:13 | cburgess | sdague cool.. its +1 from me. | |
| 19:00:16 | mriedem | leakypipes: superdan: https://review.openstack.org/#/c/505417/6 | |
| 19:00:18 | sdague | for the land of fun... qemu-img --version | head -1 on centos7, ubuntu 16.04, ubuntu 17.04 | |
| 19:00:24 | sdague | qemu-img version 1.5.3, Copyright (c) 2004-2008 Fabrice Bellard | |
| 19:00:32 | sdague | qemu-img version 2.5.0 (Debian 1:2.5+dfsg-5ubuntu10.16), Copyright (c) 2004-2008 Fabrice Bellard | |
| 19:00:39 | sdague | qemu-img version 2.8.0(Debian 1:2.8+dfsg-3ubuntu2.4) | |
| 19:00:47 | cburgess | Yeah you will break RHEL with that change. | |
| 19:00:51 | cburgess | Forgot to mention that. | |
| 19:00:56 | sdague | cburgess: why? | |
| 19:01:12 | cburgess | sdague Opps yeah just realized I was wrong. We don't block on 2.10 now.. we just change the args. My bad. | |
| 19:01:20 | sdague | cburgess: right | |
| 19:01:26 | cburgess | sdague Ignore me. | |
| 19:01:28 | sdague | with 2.10 right now you just explode in a ball of fire | |
| 19:01:31 | sdague | which is what the gate did | |
| 19:01:34 | cburgess | sdague LOL | |
| 19:01:42 | sdague | because the ppa for pike has 2.10 | |
| 19:01:46 | cburgess | sdague qemu-img barfs back an error at you I assume? | |
| 19:01:55 | sdague | yeh | |
| 19:02:01 | sdague | which is a stack trace | |
| 19:02:07 | cburgess | Yeah... cute. | |
| 19:02:13 | mriedem | so uh, no group rates in sydney huh? | |
| 19:03:00 | sdague | cburgess: qemu actually just gives a clean exit, but it's definitely not expected by nova code, so nova code stack traces | |
| 19:03:02 | superdan | mriedem: ack on all thanks | |
| 19:03:09 | mriedem | \o/ | |
| 19:03:29 | cburgess | sdague Also depends on which version of RHEL 7 you are on. I believe later versions of RHEL, or the OSP repos include newer versions of QEMU. On 7.4 for instance we have 2.6 | |
| 19:03:46 | superdan | mriedem: I wonder if we don't have a thing that actually tests server['fault'] from the api? | |
| 19:03:58 | sdague | cburgess: ah, yeh with OSP it might be different. I just have a centos 7 vanilla install | |
| 19:04:01 | mriedem | superdan: i'd expect a functional test somewhere, | |
| 19:04:10 | mriedem | superdan: fault testing is weird since the instance has to be deleted or in error state | |
| 19:04:10 | superdan | I would have thought | |
| 19:04:15 | mriedem | always trips me up | |
| 19:04:17 | cburgess | sdague yeah there are various additional repos you get access to with OSP. | |
| 19:04:22 | superdan | mriedem: um, what? | |
| 19:04:38 | superdan | oh for the api to show it? | |
| 19:04:41 | superdan | I see | |
| 19:04:42 | mriedem | yeah, | |
| 19:04:47 | mriedem | like these failure recreate tests we do, | |
| 19:04:50 | superdan | like your cell0 test, you could just boot things what will never schedule | |
| 19:04:57 | mriedem | i'm always wanting to just poll until the fault shows up, but most of the time that won't work | |
| 19:05:04 | openstackgerrit | Matt Riedemann proposed openstack/nova stable/pike: Fix 500 if list servers called with empty regex pattern https://review.openstack.org/506754 | |
| 19:05:08 | superdan | mriedem: gotcha | |
| 19:05:18 | mriedem | so i've become astute and the instance actions api | |
| 19:08:11 | superdan | mriedem: the args to that function aren't kwargs | |
| 19:08:23 | superdan | I can call them that way for reference, but it's wrong | |
| 19:08:36 | mriedem | i know they aren't kwargs | |
| 19:08:44 | superdan | can I comment? | |
| 19:08:45 | mriedem | but you could have some variables in the test instead or something, like you did elsewhere | |
| 19:09:02 | mriedem | i just hate having to remember what None, None, {}, None, [], [], None means | |
| 19:09:10 | superdan | yep, I can do that | |
| 19:10:01 | superdan | mriedem: I'm not seeing a functional test that actually checks fault | |
| 19:10:05 | superdan | you know we have one somewhere? | |
| 19:10:14 | mriedem | superdan: i don't | |
| 19:10:17 | mriedem | w/o searching | |
| 19:10:35 | superdan | okay | |
| 19:10:44 | mriedem | gd these sydney hotel rates | |
| 19:15:42 | stvnoyes | mriedem: going thru your review ccomments on live migrate/v3 - https://review.openstack.org/#/c/463987/18/nova/virt/libvirt/driver.py line 7077. | |
| 19:16:28 | stvnoyes | init connection should be done for the new flow so that is a problem that needs to be fixed | |