| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-09-16 | |||
| 13:20:32 | bauzas | what sean-k-mooney said | |
| 13:20:45 | openstack | Launchpad bug 1893904 in OpenStack Compute (nova) "Placement is not updated if a VGPU is re-created on a new GPU upon host reboot" [Undecided,New] | |
| 13:21:14 | stephenfin | Okay, that's what I'm thinking too. However, glance doesn't place the same restrictions on image metadata properties | |
| 13:21:21 | sean-k-mooney | correct | |
| 13:21:31 | sean-k-mooney | but nova will drop all nonstandard image metadata | |
| 13:21:39 | sean-k-mooney | well no | |
| 13:21:44 | sean-k-mooney | there is a config option for glance | |
| 13:21:50 | stephenfin | So do we just document things, or should we store stuff that's namespaced? | |
| 13:21:51 | sean-k-mooney | that uses the metadefs for validation | |
| 13:22:12 | sean-k-mooney | stephenfin: there are no namespaces in glance metadata | |
| 13:22:22 | sean-k-mooney | at least there was not untill we added trits | |
| 13:22:34 | stephenfin | right, there are traits now | |
| 13:22:38 | stephenfin | i.e. we store anything prefixed with 'aggregate_metadata_' | |
| 13:22:42 | sean-k-mooney | they decied to use prefixes not namespaces | |
| 13:22:54 | stephenfin | namespace, prefix, whatever | |
| 13:23:08 | stephenfin | but just '<namespace>key=value' | |
| 13:23:11 | sean-k-mooney | they are different technically | |
| 13:23:31 | stephenfin | sure, but let's gloss over that for now | |
| 13:23:42 | sean-k-mooney | but ya am i think we shoudl jsut close this as not a bug | |
| 13:23:49 | sean-k-mooney | well invalid | |
| 13:23:58 | stephenfin | wontfix | |
| 13:24:08 | stephenfin | it was a behavior change | |
| 13:24:11 | sean-k-mooney | ah we have that then ya | |
| 13:24:20 | gibi | stephenfin: I agree to close it with some notes about not supporting non-standard props | |
| 13:24:25 | sean-k-mooney | yes but it was not supproted before | |
| 13:24:29 | sean-k-mooney | it just happened to work | |
| 13:24:52 | stephenfin | Okay, I'll do that so. I have a functional test to prove it too. I assume I can just drop that? | |
| 13:25:31 | sean-k-mooney | you could keep it as a negitve regression | |
| 13:25:45 | sean-k-mooney | to ensure we dont acidentaly roll back and enable it | |
| 13:27:14 | gibi | stephenfin: drop it | |
| 13:28:10 | gibi | you could add a warning in the doc of AggregateImagePropertiesIsolation if you wish | |
| 13:29:26 | gibi | sean-k-mooney: btw, do you familiar with this periodic log https://zuul.opendev.org/t/openstack/build/470edd644f704a578226c55c7d92ba39/log/controller/logs/screen-n-cpu.txt#8179 ? For me it seems like a new and pretty useless thing to log | |
| 13:29:28 | openstackgerrit | Stephen Finucane proposed openstack/nova master: docs: Improve 'AggregateImagePropertiesIsolation' docs https://review.opendev.org/752236 | |
| 13:29:43 | stephenfin | bauzas, gibi: Real easy doc-o bug there so ^ | |
| 13:29:55 | bauzas | kk | |
| 13:30:50 | gibi | stephenfin: +2 | |
| 13:30:51 | bauzas | stephenfin: you closing the bug, right? | |
| 13:30:56 | sean-k-mooney | stephenfin: glance has a allow_additional_image_properties option which when set to false means you can only use standard image properties https://docs.openstack.org/glance/latest/admin/useful-image-properties.html#using-image-properties | |
| 13:31:01 | stephenfin | bauzas: yup, writing a reply atm | |
| 13:31:08 | bauzas | ta | |
| 13:31:11 | sean-k-mooney | https://docs.openstack.org/glance/latest/configuration/glance_api.html#DEFAULT.allow_additional_image_properties | |
| 13:31:35 | stephenfin | gibi: I'm not sean-k-mooney but I suspect that's just a difference between how that modules treats DEBUG and how we do | |
| 13:31:47 | stephenfin | we can easily configure logging levels to ignore that | |
| 13:31:48 | gibi | stephenfin: so we just need a filter | |
| 13:31:53 | stephenfin | yes | |
| 13:32:19 | sean-k-mooney | gibi one sec ill take a look | |
| 13:32:24 | stephenfin | there's a config option that we use to set the logging level for e.g. oslo.log | |
| 13:33:32 | sean-k-mooney | gibi that is because we swapped the default backend | |
| 13:33:34 | openstack | Launchpad bug 1887377 in OpenStack Compute (nova) "nova does not loadbalance asignmnet of resources on a host based on avaiablity of pci device, hugepages or pcpus." [Undecided,New] - Assigned to sean mooney (sean-k-mooney) | |
| 13:33:34 | bauzas | sean-k-mooney: looks to me a feature https://bugs.launchpad.net/nova/+bug/1887377 | |
| 13:33:46 | sean-k-mooney | os vif now uses the native backend | |
| 13:33:51 | bauzas | sean-k-mooney: because it could be a weigher or using the placement API | |
| 13:33:54 | sean-k-mooney | instead of ovs-vsctl | |
| 13:34:07 | stephenfin | definitely a feature | |
| 13:34:10 | sean-k-mooney | gibi: i can install a log filter for that if you like and drop it | |
| 13:34:31 | stephenfin | sean-k-mooney: again, there's already a conf option for this in nova | |
| 13:34:35 | stephenfin | that oslo.log provides | |
| 13:34:44 | sean-k-mooney | bauzas: its one that i want to backport downstream and do as a preqquist to numa in placment | |
| 13:34:55 | stephenfin | I just can't remember what it's called and quick grep doesn't return anything, heh | |
| 13:35:02 | sean-k-mooney | bauzas: i would prefer to backport it upstream too but if we agree it a feature then fine | |
| 13:35:21 | sean-k-mooney | i guess its something we could not bring back but its broderlien in my view | |
| 13:36:03 | bauzas | sean-k-mooney: well, it would change the behaviour then, right? | |
| 13:36:13 | bauzas | at least if you modify the default | |
| 13:36:27 | sean-k-mooney | loadbalancing between the numa nodes on the host | |
| 13:36:31 | bauzas | surely | |
| 13:36:43 | sean-k-mooney | it would slightly yes | |
| 13:36:49 | bauzas | but operators already know about it, | |
| 13:36:55 | bauzas | so they know it's packing atm | |
| 13:37:14 | sean-k-mooney | they know that if they are using numa they should match teh numer of numa nodes that are on the host | |
| 13:37:20 | sean-k-mooney | but they often dont do that | |
| 13:37:23 | bauzas | if we change the default to rather spreading, then this would change their own capacity | |
| 13:37:24 | gibi | stephenfin: https://docs.openstack.org/oslo.log/latest/configuration/index.html#DEFAULT.default_log_levels | |
| 13:37:33 | gibi | I think this is the config | |
| 13:37:35 | stephenfin | that's the one \o/ | |
| 13:37:43 | bauzas | sean-k-mooney: for this reason, I don't want it to be a bug | |
| 13:37:52 | sean-k-mooney | bauzas: it will pesimise the spawnign of large vms | |
| 13:37:54 | stephenfin | yes, we overload that in 'nova/config.py' | |
| 13:38:02 | sean-k-mooney | but it would fix many other issues | |
| 13:38:04 | bauzas | and fwiw, everytime we added a new filter or weigher, we said "noooo sorry, not a bug" | |
| 13:38:16 | bauzas | this would be the same | |
| 13:38:19 | sean-k-mooney | bauzas: its not a new filter or weigher | |
| 13:38:22 | bauzas | I know | |
| 13:38:48 | bauzas | but spreading/packing resources are related to a weigher, even not that's the implementation way | |
| 13:39:01 | gibi | stephenfin, sean-k-mooney: I will add an extra log settings for the periodic ovs debug log in https://docs.openstack.org/oslo.log/latest/configuration/index.html#DEFAULT.default_log_levels | |
| 13:39:01 | sean-k-mooney | ok well then ill close the downstrema bz and say this wont be fixed in 16.x then | |
| 13:39:06 | bauzas | ie. you want to change how to organize the resources | |
| 13:39:14 | bauzas | for this reason, not a bug | |
| 13:39:15 | gibi | I mean in nova.config.parse_args | |
| 13:39:20 | sean-k-mooney | gibi: you could or i coudl do it in os-vif | |
| 13:39:26 | sean-k-mooney | i think os-vif makes more sense | |
| 13:39:36 | sean-k-mooney | gibi: this is coming form the ovs plugin | |
| 13:39:38 | gibi | yeah, you are right os-vif is a better place | |
| 13:39:49 | gmann | gibi: stephenfin brinzhang_ ok for me, +A on 500->400. | |
| 13:39:59 | sean-k-mooney | it woudl still be configrev via the nova.conf jsut form a os vif config section | |
| 13:40:00 | gibi | gmann: thanks | |
| 13:40:05 | sean-k-mooney | so the ux is more or less the same | |
| 13:40:09 | gmann | gibi: is it ocata is 'Unmaintained' for nova? we only need backport till pike? | |
| 13:40:16 | gmann | lyarwood: ^^ | |
| 13:40:18 | bauzas | gmann: gibi: stephenfin: are we discussing on the API issue we saw yesterday ? | |
| 13:40:33 | gmann | bauzas: yeah, server group with empty 'policies' | |