Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2023-04-24
16:11:44 bauzas so we may ask for a cap
16:11:45 dansmith I said some of our packages we install from bindep to avoid compiling things like mysqlclient (AFAIK)
16:11:54 sean-k-mooney ah
16:12:06 sean-k-mooney ya sorry you did mention mysql
16:12:11 sean-k-mooney not uwsgi
16:12:16 dansmith I think the uwsgi python module *is* tied directly to the python version though
16:12:27 dansmith so we'd need a python3.9-uwsgi-module-python3 (or whatever the name is)
16:12:32 sean-k-mooney at least in the disto package i think that is correct
16:13:16 dansmith uwsgi-plugin-python3
16:13:28 dansmith this ^ is compiled directly against the base python3, not 3.9
16:14:46 dansmith see the last answer here: https://stackoverflow.com/questions/68413988/building-a-uwsgi-plugin-for-python-3-9-failed-for-older-version-it-works-are-th
16:14:55 sean-k-mooney for the devstack venv path i think you can override that with https://review.opendev.org/c/openstack/devstack/+/558930/26/files/apache-horizon.template
16:14:55 dansmith might be able to pip install uwsgi with the 3.9 python
16:15:08 sean-k-mooney WSGIPYTHONHOME
16:15:14 dansmith I'
16:15:17 dansmith I am pretty sure not
16:15:30 dansmith uwsgi runs the python interpreter internally, AFAIK
16:15:44 dansmith that's how it provides its phantom importable modules, etc
16:15:50 sean-k-mooney i know i was doing somethign to get it too work in that seriess
16:15:55 sean-k-mooney but i dotn recall
16:16:31 sean-k-mooney oh it was this https://review.opendev.org/c/openstack/devstack/+/558930/26/functions-common#1605
16:16:38 sean-k-mooney anywya that wont really help here
16:16:54 sean-k-mooney for the ceph job we whould rever back to 3.8 i guess
16:17:05 sean-k-mooney if we are keeping it on focal
16:18:28 sean-k-mooney mixing the workaround to run with uwsgin for a venv and the ceph job is not helpful i tought there was a simple fix in that but no
16:19:17 bauzas stupid question but I guess we can't block a specific tooz version in our own requirements.txt file ?
16:19:45 sean-k-mooney we can but we are not ment too
16:19:53 sean-k-mooney we can do tooz!=whatever
16:20:08 sean-k-mooney that will cause use to downgrade if its already installed
16:20:12 bauzas and tooz<=version ?
16:20:22 sean-k-mooney we dont want to cap
16:20:24 dansmith right but we'll still install the initial version first, then downgrade it,
16:20:31 sean-k-mooney but we can block know broken ones
16:20:38 dansmith and if someone that runs after us has >=4.0 it will be re-re-installed
16:20:40 dansmith so that's not the best solution, IMHO
16:21:00 bauzas I was just trying to buy us some time :)
16:21:19 dansmith yeah, it might be a quick fix
16:31:18 bauzas fwiw https://review.opendev.org/c/openstack/requirements/+/881329
16:31:27 bauzas feel free to vote (loudly)
16:34:24 sean-k-mooney dansmith: was the same done for castalan? for glance?
16:34:48 sean-k-mooney its currently castellan===4.1.0
16:34:49 dansmith we just dropped the 38 jobs
16:34:53 sean-k-mooney ah ok
16:41:55 bauzas so, before I leave, https://review.opendev.org/c/openstack/requirements/+/881329 should unblock the gate
16:42:28 bauzas this will give us time, ideally to investigate the uwsgi 3.9 support
16:42:52 bauzas that being said, I'm OK with dropping 3.8 support for the functional tests
16:54:55 sean-k-mooney i thihnk we should just see if we can deploy devstack with ceph on 22.04 using cephadm
16:55:19 sean-k-mooney if we can we can swap the nova ceph jobs to that
16:55:37 sean-k-mooney i can push a DNM patch to test that but i proably wont have time to actully try it myself
16:57:00 dansmith there's a patch started already IIRC
16:59:11 dansmith ah that was an old DNM I was thinking about
21:59:11 opendevreview Lin Yang proposed openstack/os-traits master: CPU: add traits for new X86 feature "AMX" https://review.opendev.org/c/openstack/os-traits/+/868149
22:18:12 dansmith um, neutron now requires >=3.9 as well? https://zuul.opendev.org/t/openstack/build/c7a96046411c4e05a669003bb67b299d
22:18:48 dansmith yup: https://review.opendev.org/c/openstack/neutron/+/881333
22:19:22 clarkb looks like they did it in response to the tooz update
22:23:02 opendevreview Dan Smith proposed openstack/nova master: Stop ignoring missing compute nodes in claims https://review.opendev.org/c/openstack/nova/+/880633
22:23:02 opendevreview Dan Smith proposed openstack/nova master: Remove silent failure to find a node on rebuild https://review.opendev.org/c/openstack/nova/+/880632
22:23:03 opendevreview Dan Smith proposed openstack/nova master: Make focal job non-voting https://review.opendev.org/c/openstack/nova/+/881409
22:23:10 dansmith clarkb: yeah, not sure if they were trying to help with that or not...
22:23:32 dansmith so I imagine that's going to take a while to undo, if at all, so this ^ marks that job as non-voting
22:24:15 dansmith I'm also not sure why we have that focal job, tbh
22:24:51 dansmith ah: https://review.opendev.org/c/openstack/nova/+/861111
22:25:01 dansmith so I can probably just nuke it instead of making it n-v since it'll fail
22:27:01 opendevreview Dan Smith proposed openstack/nova master: Remove silent failure to find a node on rebuild https://review.opendev.org/c/openstack/nova/+/880632
22:27:01 opendevreview Dan Smith proposed openstack/nova master: Remove focal job for 2023.2 https://review.opendev.org/c/openstack/nova/+/881409
22:27:02 opendevreview Dan Smith proposed openstack/nova master: Stop ignoring missing compute nodes in claims https://review.opendev.org/c/openstack/nova/+/880633
22:28:13 dansmith gmann: ^
22:42:56 dansmith fml
22:42:56 dansmith oh right, the ceph jobs are focal still, which means we're blocked until neutron reverts that or we have to mark ceph jobs as n-v
23:07:55 artom dansmith, hey, just going to drop this here drive-by style: https://github.com/openstack/devstack-plugin-ceph/blob/master/devstack/plugin.sh#L5-L9
23:08:18 dansmith artom: what about it?
23:08:21 artom Apparently this cephadm thing is the New Correct Way of installing ceph, essentially bypassing all the Ubuntu Ceph RPMs
23:08:50 dansmith yeah we know :)
23:09:01 artom I'll follow up with an email tomorrow, but that's what I got out of a chat with Giulio
23:09:36 dansmith last I checked (when we discussed at the recent ptg) it either didn't work correctly, or something about what we need for our regular jobs doesn't work with that mode
23:09:46 dansmith I don't know the details, but that's what we need to get worked out
23:09:58 artom Ack - more info (and contact people) to follow tomorrow
23:10:26 dansmith I just rechecked this, which frickler refreshed recently: https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/865315
23:10:32 dansmith which I'm hoping is supposed to actually work
23:12:23 dansmith you see the jobs there are still deploying without that flag and on focal "until it works" and that was more recent than the cephadm patches
23:12:39 dansmith I mean, before the revert of course
23:13:34 dansmith I want to say it's something like installing with cephadm doesn't leave the ceph config visible which nova needs to resolve the cluster or something like that
23:40:16 dansmith artom: yeah, the cephadm-based job fails pretty early and pretty hard: https://zuul.opendev.org/t/openstack/build/cce90efc7177459e9eebdab966bab75c
23:40:44 clarkb the jammy ceph version is also new enough for nova iirc
23:41:20 dansmith clarkb: right, the reason we didn't do that before was because they didn't yet have jammy packages IIRC, but I think they do now
23:41:46 dansmith either way, it's failing pretty hard
23:42:09 clarkb I think there was actually a misunderstanding. Everyone was looking for cloud archive packages for some reason, but the base distro had packages that were quite new
23:42:10 dansmith I mean the distro-package-based one
23:42:32 dansmith clarkb: no I think people were looking for ceph.org packages, or at least that's what I remembered
23:42:47 clarkb ya that was the first issue, then they looked for UCA packages and didn't find those either
23:42:49 dansmith but perhaps that was because jammy already had them and that's just how we did it before, I dunno
23:42:53 dansmith ack okay
23:42:57 clarkb but the base distro has newish packages that can probabl ybe made to work
23:43:43 dansmith the job running on jammy is about to finish, but it had lots of tempest fail in it
23:43:52 dansmith from this: https://review.opendev.org/c/openstack/devstack-plugin-ceph/+/865315?tab=change-view-tab-header-zuul-results-summary
23:47:17 dansmith qemu: module block-block-rbd not found
23:47:20 dansmith in nova compute's log
23:47:33 sean-k-mooney[m] dansmith: i think you just need to add the ensure podman role to a pre playbook
23:48:20 dansmith sean-k-mooney[m]: could be, but my point is, I don't think that job is like super healthy and well-tested such that switching to it is the obvious choice :)
23:48:48 sean-k-mooney[m] hehe ya...

Earlier   Later