Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-11-19
13:53:22 mriedem so do whatever
13:53:31 openstackgerrit Alexandre arents proposed openstack/nova master: Do not update root_device_name during guest config https://review.opendev.org/670000
13:53:42 mriedem let the tc work on getting devstack to use py3 by default
13:54:48 corvus mriedem: if nova runs all devstack jobs with USE_PYTHON3 set to true, then the same thing can happen again
13:55:21 corvus if you don't want to revert that whole change, can you at least keep one py2 job running as long as py2 is the devstack default?
13:55:58 mriedem the same thing == not being able to install nova with py2, that's wrong, right? as long as we don't screw up setup.cfg,
13:56:11 mriedem tempest tests could break if nova lands some code that doesn't work with py2, i get that
13:56:23 mriedem but frankly i don't see that happening real soon, and if it does it can be fixed,
13:56:32 mriedem and it's also motivation to get downstream projects to move to py3 as well
13:56:38 corvus mriedem: right, i'm saying it's a testing hole. someone could introduce a change that doesn't work with py2.
13:56:42 mriedem which is what i thought everyone wanted in ussuri as a major community goal
13:56:58 corvus my only goal here is that "./stack.sh" always work for new users.
13:57:51 mriedem "always works" is pretty subjective
13:57:53 mriedem bugs are bugs
13:58:08 mriedem i agree that if nova doesn't install it's a problem
13:58:11 mriedem so revert the setup.cfg change
13:58:11 tosky mriedem: please approve the revert - all the other projects are broken because nova does not start
13:58:25 mriedem some other core can approve if they care
13:58:40 mriedem i'd like the community to shit or get off this py3 pot
13:58:43 corvus mriedem: yeah, but there's no testing of nova in the default devstack configuration.
13:58:49 mriedem christ we're talking about py3.8 gating in the tc threads these days already
13:59:28 mriedem do whatever stephenfin and efried want
14:00:42 tosky mriedem: I'm not sure why you are saying or implying that the community does not want to move - I think we want to move, just without blocking the work
14:02:10 mriedem i've already said i'm good with fixing setup.cfg so nova can install again,
14:02:34 mriedem i don't think we should be gating on py2 though and if something slips through that's a py2 bug then we fix it
14:02:53 mriedem otherwise just let the qa/tc/infra whoever team that cares about this manage it directly imo
14:03:09 mriedem since the projects obviously can't orchestrate this themselves
14:03:15 tosky fine, but then probably you also need to revert the requirment change
14:03:22 tosky not just setup.cfg
14:03:57 mriedem do whatever you guys want
14:04:33 stephenfin So someone needs to go first, and I thought it could be us since we're a service, not a library, but obviously that isn't the case
14:04:51 stephenfin What's the hold up with DevStack switching to Py3 first though? Can't we just merge that?
14:05:06 tosky I don't know
14:05:12 mriedem someone has to care
14:05:14 corvus please have at least one test on nova that runs devstack in the default configuration. otherwise, there is a very high risk that changes will merge to nova which will break it.
14:05:53 stephenfin okay, makes sense. So can I go switch DevStack to use Python 3 by default
14:06:28 corvus stephenfin: https://review.opendev.org/649097 is relevant
14:07:21 sean-k-mooney yes so it look lke we need to merge those 3 patches
14:07:31 efried sean-k-mooney: sorry, what's the third?
14:07:54 efried oh, the predecessor of --^ ?
14:07:58 corvus mriedem: do you want to remove the wip setting on that change?
14:07:59 sean-k-mooney https://review.opendev.org/#/c/688799/5 https://review.opendev.org/#/c/649097/10 and https://review.opendev.org/#/c/649278/5
14:08:45 sean-k-mooney mriedem's change should be the last of the 3
14:08:54 sean-k-mooney the other two can merge in any order i think
14:09:11 sean-k-mooney https://review.opendev.org/#/c/688799/5 already has 1 +2
14:10:08 stephenfin sean-k-mooney: Now that CentOS 8 is out, why wouldn't you just use that?
14:10:51 stephenfin IMO if we're switching to Python 3-only we should be testing on Python 3-supporting distros, namely CentOS 8
14:10:54 tosky stephenfin: we were discussing about CentOS 8/devstack support with frickler earlier today - it's not there yet
14:10:56 sean-k-mooney you could but devstack may not work on it
14:11:26 sean-k-mooney im pretty sure you will need eple or ius repos to make it work
14:11:28 stephenfin tosky: can we disable CentOS-based jobs so?
14:11:40 tosky stephenfin: there is a review for it
14:11:42 sean-k-mooney stephenfin: we can run python 3 on centos 7
14:12:02 sean-k-mooney but we need pyton36 form eple or ius
14:12:04 tosky stephenfin: https://review.opendev.org/#/c/694967/
14:12:40 kashyap (On RPM-based distros, DevStack generally "works for me" on Fedora. Never tried on CentOS, though.)
14:12:40 stephenfin sean-k-mooney: we can, but we can also do it on CentOS 8. Both involve some work, and it seems obvious to me that the effort should spent on the latter option
14:13:02 mriedem corvus: removing the WIP doesn't make it work, as you can see from the job results
14:13:03 sean-k-mooney stephenfin: we can do both
14:13:12 kashyap stephenfin: Just chiming in randomly, why can't one just stick to a Fedora-only job? (IIRC, there was one; need to check if it's still maintained, or is rotting)
14:13:16 mriedem as sean pointed out there are a series of changes so someone has to align that stuff
14:13:22 mriedem the one under my WIP has been sitting since april
14:13:29 mriedem so clearly not a community priority
14:14:10 mriedem kashyap: as far as i can recall the fedora jobs in openstack have been a garbage dump since forever
14:14:36 mriedem i don't know of any projects in openstack that gate on fedora, maybe tripleo or some deployment projects
14:14:38 kashyap mriedem: Right, need bodies.
14:14:39 stephenfin kashyap: Fedora is too much of a moving target for CI, I'm afraid
14:14:41 tosky mriedem: maybe devstack jobs - tripleo was able to have f28 jobs
14:14:43 sean-k-mooney mriedem: well that because fedora and ci is basically an anti pattern. it chagne too much and keeps breaking
14:14:47 stephenfin Ask sean-k-mooney - they know all about it
14:14:50 stephenfin jinx :D
14:14:50 kashyap The only one I know who worked on Fedora jobs was/is ianw
14:15:00 stephenfin anyway, that's a different discussion
14:16:03 kashyap Yeah, agreed on Fedora being a "moving target"; if CentOS-8 is more manageable, by all means.
14:16:42 kashyap sean-k-mooney: "Anti-pattern" for a cloud software probably; libvirt and QEMU upstream test on Fedora jobs just fine.
14:16:49 sean-k-mooney kashyap: i think the main issue with centos 8 will be a lack of packages
14:16:58 sean-k-mooney kashyap: libvirt and qemu are too of the main issues
14:17:33 sean-k-mooney we hit temproy brakages or chagnes in the output of things like qemu-img
14:17:38 kashyap (Also, bear in mind -- most / all upstream kernel / Virt issues are fixed on Fedora first -- as most kernel / Virt upstream devs use it.)
14:18:24 sean-k-mooney ya the kernel and gcc kept causing issues for compiling dpdk and ovs too
14:18:36 sean-k-mooney anyway thats a different topic
14:18:38 stephenfin corvus, tosky: Can we combine https://review.opendev.org/#/c/649097/ and https://review.opendev.org/#/c/694967/ and just merge that?
14:18:54 kashyap Anyway, this is all largely about having enough warm bodies that are incentivized to care.
14:19:06 stephenfin So switch DevStack to Python 3 and drop the CentOS 7 job, which will no longer work
14:19:48 mriedem stephenfin: those non-ubuntu jobs aren't voting anyway so they don't matter
14:19:54 stephenfin True
14:20:09 mriedem if a job isn't voting it's not enough of a priority for people to care about it
14:20:09 tosky stephenfin: maybe, I don't know; I'd wait for gmann too, even though I have some objections to his comment in https://review.opendev.org/#/c/649097/ as I wrote there
14:20:46 stephenfin I don't want to inflict pain on people, but as mriedem notes someone needs to start driving this and if I'm doing that from nova's side, then a straight up revert is what I want
14:20:48 stephenfin It sounds like we'd be kicking a can down the road. I'd rather help people pick up the pieces than do that
14:21:14 sean-k-mooney mind if i just rebase the 3 devstack patch into the right order
14:21:25 stephenfin sean-k-mooney: before you do, what are the three?
14:21:33 tosky stephenfin: it's too big right now - if you don't want to fully revert https://review.opendev.org/#/c/694891/, are you going to accept at least a revert of setup.cfg (the condition) and *requirements.txt ?
14:21:57 sean-k-mooney Handle uninstalled python3 -> Handle localrc early enough in stackrc -> Change USE_PYTHON3=True by default
14:22:25 sean-k-mooney that fixes centos 7 then enable python 3 early then set it to the defaul
14:22:29 stephenfin sean-k-mooney: again, I don't think we need "Handle uninstalled python3". Let's use CentOS 8 instead
14:22:38 sean-k-mooney we can do both
14:22:51 stephenfin we can, but it doesn't mean we should
14:23:16 stephenfin tosky: We can, but that means we'll no longer test against Python 2.7
14:23:36 sean-k-mooney if we want to drop all centos testing until centos 7 works we could but why when its a trivial patch

Earlier   Later