| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-11-19 | |||
| 14:57:13 | stephenfin | (switching DevStack to Python 3) | |
| 14:57:18 | gmann | i do not prefer nova or any project keep py2 job. but did not thought about min condition on setup.cfg | |
| 14:57:21 | efried | are we objecting to re-adding a py2 job because we're worried about gate resources? | |
| 14:57:25 | dansmith | that's like turning off gating for a bit because we'll be careful | |
| 14:57:34 | efried | ^ this | |
| 14:57:56 | dansmith | stephenfin: I still don't really understand.. gmann is saying we can't switch the default because a lot of jobs don't specify and would break on py3? | |
| 14:58:10 | mriedem | efried: we don't gate on ceph or ironic and yet people care about those right? | |
| 14:58:11 | sean-k-mooney | for what its worth that is why iw as suggesting fixing py3 on centos7 instad of disabling cento7 job untill cento 8 works | |
| 14:58:20 | gmann | yeah but it is deadlock seems and need to tradeoff with either one | |
| 14:58:35 | stephenfin | dansmith: I think so, yeah | |
| 14:58:47 | efried | mriedem: Agreed that we can't test everything in the world. But a minimal canary to make sure devstack doesn't just shit all over itself seems appropriate. | |
| 14:58:47 | dansmith | well, the tradeoff, is, IMHO, what efried said, which is add back a single py2 devstack job | |
| 14:59:00 | mriedem | and when can that be removed? | |
| 14:59:07 | dansmith | mriedem: when the default switches | |
| 14:59:09 | mriedem | when the tc fairy shows up in january to tell us we're ok? | |
| 14:59:11 | efried | when devstack is py3 by default and we can do the minimum thing. | |
| 14:59:23 | dansmith | mriedem: I don't care about the date, I care about the default | |
| 14:59:30 | stephenfin | Can we just make devstack py3 by default? | |
| 14:59:32 | efried | ("the minimum thing" => restore the py3 minimum in setup.cfg) | |
| 15:00:02 | stephenfin | We know there's going to be some degree of pain here. By reverting anything in nova, we're kicking the can down the road | |
| 15:00:05 | efried | stephenfin: That's mriedem's patch, yes? I rechecked that a bit ago to see if it works yet. If so, maybe we can just pull the trigger. | |
| 15:00:05 | mriedem | then just efried and dansmith should approve ianw's revert | |
| 15:00:07 | gmann | default switch was planned for ~Jan | |
| 15:00:21 | gmann | ok, let's move devstack to py3 and then if anyone need py2 testing (only lib and testing tools) can explicitly mention the py2 need | |
| 15:00:23 | efried | though the predecessor also would need to go. | |
| 15:00:26 | stephenfin | Switching to Python 3 by default is at least forward-looking pain | |
| 15:00:29 | dansmith | gmann: right so if the default flip is coming, I don't see what the big deal is | |
| 15:00:32 | gmann | except swift everyone has to switch to py3-only in ussuri | |
| 15:00:45 | mriedem | dansmith: that devstack patch of mine has been on top of another patch that has been around since april | |
| 15:00:49 | stephenfin | then lets do that now | |
| 15:01:03 | dansmith | stephenfin: I would agree except I don't know how many things will insta-fail on projects with few people working on them | |
| 15:01:04 | gmann | i have patches upnfor all the projects (except clieent, testing tool and lib) so project can merge thsoe | |
| 15:01:13 | dansmith | mriedem: I know | |
| 15:01:23 | dansmith | mriedem: I'm not a py3 bigot, so I'm really in no rush :) | |
| 15:01:34 | stephenfin | dansmith: those projects are going to insta-fail at some point though | |
| 15:01:59 | dansmith | stephenfin: yep, so if the TC wants to schedule that, and it's in Jan, I would be fine waiting and moving then as long as it happens | |
| 15:02:14 | dansmith | but, also fine now, I just don't want the disjoint default.. like I said, the date doesn't matter to me | |
| 15:02:57 | gmann | dansmith: but it can be blocked other projects too and two times effort to get rid of py2. so i am ok to move devstack first | |
| 15:03:23 | mriedem | dansmith: then just approve this https://review.opendev.org/#/c/694891/ | |
| 15:03:55 | gmann | because if nova or any other project say min version is py3 then it is fair because reno said the same. | |
| 15:04:22 | dansmith | mriedem: I'm trying to reach consensus like everyone else, not force-merge anything | |
| 15:04:50 | dansmith | if gmann is going to flip the devstack default, then we don't need to do anything, AFAICT | |
| 15:05:02 | stephenfin | yup | |
| 15:05:19 | stephenfin | gmann: just so I'm 100% clear, what was the rationale for the proposed January switchover? | |
| 15:05:35 | stephenfin | I missed whatever email discussed that so a link is fine :) | |
| 15:05:36 | dansmith | January is when the actual upstream support stops, AFAIK | |
| 15:05:53 | dansmith | so I imagine it's aligned there so that the TC's deadline has weight | |
| 15:05:56 | gmann | stephenfin: to keep testing py2 on client lib and lib, testing tool which were planned to drop py2 in jan | |
| 15:06:31 | gmann | plan is like, consumers drop py2 first which is openstack services and then lib or testing tool | |
| 15:06:34 | sean-k-mooney | gmann: it was planned for m2 in febuary | |
| 15:06:38 | stephenfin | We should probably switch over sooner than that then | |
| 15:06:39 | sean-k-mooney | well by m2 | |
| 15:07:02 | stephenfin | We need to keep clients and libraries on python 2 for longer to let the services switch everything over | |
| 15:07:05 | dansmith | honestly, switching the default in devstack at m2 seems crazy late to me | |
| 15:07:08 | dansmith | just for our cycle | |
| 15:07:25 | stephenfin | so we should be switching devstack now to let the services fix whatever they need to fix | |
| 15:07:34 | stephenfin | IMO, of course | |
| 15:07:36 | dansmith | that makes more sense to me for sure | |
| 15:07:38 | dansmith | break early | |
| 15:07:40 | gmann | stephenfin: yeah and all services has to switch before m-2 | |
| 15:07:52 | stephenfin | dansmith: we also have Xmas to worry about | |
| 15:07:53 | gmann | +1. | |
| 15:08:00 | dansmith | stephenfin: not nova's X-mas! :P | |
| 15:08:09 | dansmith | but yeah I get that for sure | |
| 15:08:18 | dansmith | so, just tobe clear, | |
| 15:08:40 | dansmith | even if we push the devstack default, we need the partial revert so that py2-only projects can still install nova in their jobs | |
| 15:08:54 | sean-k-mooney | dansmith: yes | |
| 15:08:58 | openstackgerrit | Mark Goddard proposed openstack/nova master: Add functional regression test for bug 1853009 https://review.opendev.org/695012 | |
| 15:08:58 | openstack | bug 1853009 in OpenStack Compute (nova) "Ironic node rebalance race can lead to missing compute nodes in DB" [Undecided,In progress] https://launchpad.net/bugs/1853009 - Assigned to Mark Goddard (mgoddard) | |
| 15:08:58 | dansmith | so we could push the default in devstack now, then revert our can-use-py2 part at m2 as the hard deadline yeah? | |
| 15:09:12 | gmann | stephenfin: are you rebasing the mriedem patch and remove neutron-grenade job also in same patch to pass the gate | |
| 15:09:14 | stephenfin | sean-k-mooney: why? | |
| 15:09:43 | dansmith | stephenfin: so that trove (example) can still run py2 devstack jobs with nova | |
| 15:09:51 | sean-k-mooney | why? so that if neutron need to keep a job on py2 for a day or two to fix it they can | |
| 15:09:57 | dansmith | right | |
| 15:10:02 | gmann | dansmith: they has to move as soon as any dependencies fail their gate | |
| 15:10:42 | stephenfin | But unless they're setting USE_PYTHON3=False in jobs, they'll inherit the default (now python3), no? | |
| 15:10:44 | dansmith | yeah | |
| 15:10:45 | dansmith | that's my point, if nova can't run on py2 anymore (default or not) then they can't run jobs | |
| 15:10:52 | slaweq | sean-k-mooney: what py2 job in neutron You are talking about? | |
| 15:11:03 | dansmith | stephenfin: right, I'm sayingsome project that needs another month to be able to run devstack in default py3 | |
| 15:11:12 | stephenfin | ohhhh | |
| 15:11:15 | stephenfin | gotcha | |
| 15:11:20 | dansmith | if we don't revert the minimum change, they physically can't install nova in unsupported py2 | |
| 15:11:23 | sean-k-mooney | slaweq: none it was just an example of a big project with many jobs that often need nova in them | |
| 15:11:26 | stephenfin | okay | |
| 15:11:35 | slaweq | sean-k-mooney: ahh, ok :) | |
| 15:11:37 | gmann | for client lib which will happen later make sense to keep nova install-able on py2 , for service not | |
| 15:11:46 | stephenfin | then I'm also onboard with that | |
| 15:12:05 | dansmith | stephenfin: that all hinges on gmann being willing to pull the trigger | |
| 15:12:20 | sean-k-mooney | by the way os-vif is currently broken as we have not swaped all the jobs to python 3 | |
| 15:12:22 | stephenfin | so, revert the change to nova, merge the Python 3-ification patch to DevStack, wait a month and re-revert that change to nova | |
| 15:12:39 | dansmith | stephenfin: just the half-revert | |
| 15:12:40 | sean-k-mooney | we can trivially but we need to wait as a lib project | |
| 15:12:42 | gmann | revert only setup.cfg and req right ? | |
| 15:12:51 | gmann | yeah what dansmith mentioned | |
| 15:13:04 | dansmith | stephenfin: revert the min, default devstack to py3, later un-revert the minimum | |
| 15:13:12 | gmann | +1 | |
| 15:13:31 | stephenfin | what about your concerns that we could accidentally break Python2 support in that time? | |
| 15:13:36 | stephenfin | don't they still hold? | |