Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2023-06-05
16:25:34 bauzas so operators would see to use the next Antelope .z release if they want to use it
16:26:05 dansmith maybe you'll need to make a release before your demo so people can actually use it :D
16:28:39 bauzas yeah, I'll prepare a release
16:28:52 bauzas and I'll tell it loudly
16:28:58 bauzas as I'm very sad :(
16:30:31 bauzas also we maybe should have some test that verify that we don't forget a privsep decorator
16:30:52 bauzas like we do with objects or other stuff
16:31:35 bauzas but that's a bit difficult
16:47:21 dansmith yeah, not sure how to do that
17:42:45 opendevreview Sylvain Bauza proposed openstack/nova master: cpu: fix the privsep issue when offlining the cpu https://review.opendev.org/c/openstack/nova/+/885293
17:43:00 bauzas dansmith: added a relnote for making it more public
17:43:25 bauzas and again, sorry for this bug :(
17:43:38 dansmith how could permissions cause a FileNotFound/
17:43:44 dansmith that doesn't seem right o me
17:43:50 bauzas dansmith: see the bug report
17:44:04 bauzas I added the log
17:44:20 dansmith the bug report *says* there's a FileNotFound but I don't see *why* that would happen
17:44:25 dansmith users can see the file
17:44:29 bauzas this is a permission error as the base exception
17:45:01 dansmith oh I see, we're translating
17:45:05 bauzas yeah
17:45:37 dansmith okay, so the reno reads a bit strange to me because of the "if you want".. can I suggest alternative wording?
17:45:45 bauzas dansmith: sure, do this
17:46:03 bauzas tbc, I prefer -1s rather than +2s :)
17:49:21 dansmith commented
17:50:20 dansmith bauzas: if you're using the offlining the scaling being missing from /sys isn't a problem right? Could we enable this in nova-next or something so we see some coverage?
17:53:27 bauzas dansmith: you mean about the fact that we miss scaling_governor in my instance ?
17:53:49 bauzas I can find a way to say 'meh, fair enough' if you wan
17:53:51 bauzas want
17:54:06 bauzas so in this case, we would only support to ask to offline/online cpus
17:55:42 dansmith don't you choose modes, either to use scaling or cpu offlniing?
17:56:39 dansmith if you're using offlining, I would expect the missing cpufreq support wouldn't be a problem and that we could just enable this in nova-next
17:58:38 bauzas dansmith: yeah, we have a config option for saying which strategy you want https://docs.openstack.org/nova/latest/configuration/config.html#libvirt.cpu_power_management_strategy
17:58:56 dansmith right, so can we configure nova-next for offlining mode so we get some CI coverage?
17:59:02 bauzas we could so
17:59:49 bauzas okay, so I'll create another bugfix for making sure that if you want to offline, it's fine if we don't verify the scaling governor
18:00:01 bauzas unless you ask for it by the config opt
18:00:09 dansmith oh, is that not currently working?
18:00:35 dansmith because we should not depend on a working cpufreq implementation if we're using offlining
18:00:54 dansmith I can imagine some of the other platforms (maybe arm or power) don't always have cpufreq enabled
18:01:02 bauzas unfortunately no, because when we restart we check both the governor and the state here : https://github.com/openstack/nova/blob/master/nova/virt/libvirt/cpu/api.py#L143
18:01:12 dansmith and some server hardware if you enable platform-based scaling I think they don't expose it to the OS
18:01:13 bauzas so I'll change this
18:01:18 dansmith bauzas: okay yeah, that'd be good
18:01:42 bauzas dansmith: ++ your concern is awesome btw.
18:02:08 bauzas dansmith: because indeed I wasn't thinking about the other platforms but indeed you're right
18:02:12 dansmith yeah
18:02:25 bauzas +++ definitely, will fix it too and will backport it
18:02:37 dansmith cpufreq is pretty common these days, but if you unload a module or don't have it built into your kernel, we shouldn't fail
18:03:04 dansmith (shouldn't fail to allow offlining I mean)
18:03:38 bauzas indeed
18:05:03 dansmith bauzas: so I didn't look, but I assume the scaling approach already has the privsep helper?
18:05:09 dansmith and it was just the online helper that was missing it?
18:05:25 bauzas dansmith: yeah, we only forgot it for set_offline
18:05:36 dansmith okay cool, just checking
18:05:41 bauzas https://github.com/openstack/nova/blob/master/nova/virt/libvirt/cpu/api.py is the private API
18:05:53 bauzas and https://github.com/openstack/nova/blob/master/nova/virt/libvirt/cpu/core.py is the internal code
18:06:13 bauzas that's where we decorate the callers
18:06:41 dansmith I believe you, just making sure
18:07:40 bauzas dansmith: no worries, was just for explaining you what we do internally since you didn't had time to review the series
18:07:48 dansmith decorators are easy to miss and git rebase/merges sometimes snip them out if they're used in front of all the functions
18:08:02 bauzas https://review.opendev.org/q/topic:bp%252Flibvirt-cpu-state-mgmt
18:08:13 bauzas that's the series if you wanna peak it
18:08:36 dansmith yeah, I found it last week, I had missed it for sure
18:08:43 bauzas and https://review.opendev.org/c/openstack/nova/+/868237/12/nova/virt/libvirt/driver.py is where we offline/online
18:08:54 bauzas the closest I found when we spawn
#openstack-nova - 2023-06-06
07:57:20 gibi bauzas, dansmith: nice findings. I'm OK with the privsep fix and you can ping me with the governor check fix when it is available
08:00:42 gibi regarding asserting the use of the decorator during testing we can build a list of filesystem.write(path, data) calls that we know require the privsep decorator and then check in the test that when those calls happen the func has _ENTRYPOINT_ATTR set.
08:04:14 gibi *filesyste.write_sys
08:09:53 songwenping sean-k-mooney:hi, does live migration pass filters if assigned the host?
08:16:22 opendevreview Balazs Gibizer proposed openstack/nova master: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/873216
08:17:30 gibi sean-k-mooney: This ^^ was already approved but needed a rebase and a unit test fix due the base changes. Could you check it please?
08:17:56 gibi I did not wait for the author with the rebase and went ahead and fixed up the patch
08:22:09 opendevreview Balazs Gibizer proposed openstack/nova stable/2023.1: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885343
08:30:14 opendevreview Balazs Gibizer proposed openstack/nova stable/zed: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885344
08:33:42 bauzas gibi: sorry for the late reply, but thanks
08:36:19 opendevreview Balazs Gibizer proposed openstack/nova stable/yoga: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885345
08:45:17 opendevreview Balazs Gibizer proposed openstack/nova stable/xena: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885347
08:46:55 songwenping gibi:morning, does live migration pass filters if assigned the host?
08:47:41 gibi songwenping: it depends. See the doc of the host and the force option in https://docs.openstack.org/api-ref/compute/?expanded=live-migrate-server-os-migratelive-action-detail#id131
08:53:02 songwenping we use rocky version, and nova-conductor donot find new destination.
08:53:50 songwenping and there is a problem, if the vm has affinity, it can be migarated to other host.
08:57:29 opendevreview Balazs Gibizer proposed openstack/nova stable/wallaby: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885348
08:57:48 songwenping then if we use the same affinity strategy to create vms, these vms scheduled to different hosts. gibi, is this reasonable?
08:59:00 gibi if you use affinity strategy then you cannot move the VM. Execpt if you disable the scheduler via the force flag and old enough microversion. But if you disable the scheduler then the affinity will not be honored.
08:59:23 gibi If you need both affinity and move operations then you should use soft-affinity
09:08:07 opendevreview Balazs Gibizer proposed openstack/nova stable/victoria: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885349
09:09:40 songwenping gibi, got it, thanks^^
09:57:28 opendevreview Sylvain Bauza proposed openstack/nova master: cpu: make governors to be optional https://review.opendev.org/c/openstack/nova/+/885352
10:04:41 opendevreview Balazs Gibizer proposed openstack/nova stable/ussuri: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885353
10:19:16 opendevreview Balazs Gibizer proposed openstack/nova stable/train: Fix failed count for anti-affinity check https://review.opendev.org/c/openstack/nova/+/885355
10:22:57 sean-k-mooney gibi: its workign on https://github.com/openstack-k8s-operators/nova-operator/pull/400 by the way
10:23:17 sean-k-mooney which ran at 2023-06-05 14:50:20
10:23:30 sean-k-mooney so it passed yesterday
10:23:48 gibi ack
10:24:14 gibi you oddly switched from downstream slack to upstream irc to write that though :)
10:24:58 sean-k-mooney oh right donwstream is normlaly on the bottom half of my screen and upstream is the top
10:25:04 sean-k-mooney this windows is in the wrong place
10:25:20 sean-k-mooney fixed :)
10:25:32 gibi :)

Earlier   Later