| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2024-05-27 | |||
| 13:20:58 | opendevreview | sean mooney proposed openstack/nova master: Test live migration between hosts with differnet cpu_shared_sets https://review.opendev.org/c/openstack/nova/+/913744 | |
| 13:20:59 | opendevreview | sean mooney proposed openstack/nova master: enable numa live migration in the ceph job https://review.opendev.org/c/openstack/nova/+/913842 | |
| 13:40:16 | opendevreview | sean mooney proposed openstack/nova master: enable zswap in nova ci jobs. https://review.opendev.org/c/openstack/nova/+/905791 | |
| 13:43:38 | opendevreview | sean mooney proposed openstack/nova master: enable zswap in nova ci jobs. https://review.opendev.org/c/openstack/nova/+/905791 | |
| 15:21:47 | sean-k-mooney | bauzas: you can baseically see the ci change in the last 5-6 pataches https://github.com/openstack/nova/commits/master/.zuul.yaml | |
| 15:21:57 | bauzas | thanks | |
| 15:54:10 | gibi | sean-k-mooney: I looked at the 1 second sleep in https://review.opendev.org/c/openstack/nova/+/919979 and I have no idea why the test fails without it. So far I was not able to reproduce such failure locally. As this sleep goes away after the fix I'm fine to temporary merge it | |
| 15:55:39 | sean-k-mooney | ack ya i spent like 6 hours gettign that repoducer to work 2 weeks ago and when it finally passed locally it was flaky in the gate. adding the time.sleep makes it stable for me locally and in the gate so its the best i coudl come up with | |
| 17:31:56 | opendevreview | sean mooney proposed openstack/nova master: retry write_sys call on device busy https://review.opendev.org/c/openstack/nova/+/920203 | |
| #openstack-nova - 2024-05-28 | |||
| 02:47:38 | opendevreview | Kien Nguyen proposed openstack/python-novaclient master: support get and list server's metadata https://review.opendev.org/c/openstack/python-novaclient/+/915977 | |
| 08:38:15 | bauzas | tkajinam: I'm doing some spec paperwork for reviews but I don't see any blueprint created for https://specs.openstack.org/openstack/nova-specs/specs/2024.2/approved/libvirt-stateless-firmware.html | |
| 08:38:25 | bauzas | tkajinam: have you created one or no ? | |
| 08:48:53 | tkajinam | bauzas, there is the one https://blueprints.launchpad.net/nova/+spec/libvirt-stateless-firmware | |
| 08:49:01 | bauzas | excellent | |
| 08:49:07 | bauzas | so we only need to update the spec | |
| 08:49:43 | tkajinam | Ugh/. I thought I added the link but didn't. let me push a change so that you can +2 it | |
| 08:50:43 | bauzas | cool, I'll even +W it | |
| 08:52:54 | opendevreview | Takashi Kajinami proposed openstack/nova-specs master: Fix missing blueprint link https://review.opendev.org/c/openstack/nova-specs/+/920612 | |
| 08:54:57 | bauzas | +Wd | |
| 08:59:13 | tkajinam | thanks ! | |
| 09:07:15 | opendevreview | Merged openstack/nova master: libvirt: Ensure both swtpm and swtpm_setup exist for vTPM support https://review.opendev.org/c/openstack/nova/+/908541 | |
| 09:11:05 | opendevreview | Merged openstack/nova-specs master: Fix missing blueprint link https://review.opendev.org/c/openstack/nova-specs/+/920612 | |
| 09:11:55 | zigo | What's the directive that tells nova-compute to use Ceph as a backend for the system disk of instances? | |
| 09:12:23 | zigo | I see in the logs that nova-compute is reporting the local storage space instead of what's in Ceph, so I'm wondering if it's set properly ... | |
| 10:01:55 | frickler | zigo: "[libvirt] images_type = rbd" iirc | |
| 10:21:09 | sean-k-mooney | yep just images_type=rbd | |
| 10:21:28 | sean-k-mooney | with that config every comptue node will report the total ceph pool stoage as its local storage instead | |
| 11:19:11 | wncslln | morning nova o/, I'm solving this bug https://bugs.launchpad.net/nova/+bug/1994139 | |
| 11:19:19 | wncslln | I've changed the neutron client to openstacksdk client and a bunch of tests are broken with the message: "keystoneauth1.exceptions.auth_plugins.MissingAuthPlugin: An auth plugin is required to determine endpoint URL" | |
| 11:19:44 | wncslln | I need to change the way that the auth plugin is loaded too? | |
| 11:20:35 | sean-k-mooney | yep that is part of changing to the client | |
| 11:20:48 | sean-k-mooney | *form the client to the sdk | |
| 11:21:22 | sean-k-mooney | we already converged form ironic client ot neutron client last cycle | |
| 11:21:35 | JayF | ironic client -> openstacksdk | |
| 11:21:37 | sean-k-mooney | the likely needed to do the same converation ill see if i can find any patches related to it | |
| 11:21:39 | JayF | don't put neutron on me ;) | |
| 11:21:55 | sean-k-mooney | sorry yes ironic to sdk | |
| 11:22:01 | JayF | There are examples of openstacksdk to an external service in Ironic, but I don't think we preserve user context -- you all may need that for Neutron (?) | |
| 11:22:16 | sean-k-mooney | yes in general | |
| 11:22:23 | JayF | **Ironic driver | |
| 11:22:35 | sean-k-mooney | ironic is kind of special in that its kind of admin only | |
| 11:22:45 | JayF | *was admin only | |
| 11:22:48 | wncslln | hahahaha | |
| 11:22:48 | sean-k-mooney | i.e. we cant just use the users token | |
| 11:23:15 | wncslln | i have find this patch on ironic https://review.opendev.org/c/openstack/ironic/+/900265/22/ironic/common/cinder.py#94 | |
| 11:23:28 | wncslln | but i think is kind a different from nova | |
| 11:23:35 | JayF | honestly that's probably not a bad example I suspect | |
| 11:24:00 | JayF | but I was talking more about https://opendev.org/openstack/nova/src/branch/master/nova/virt/ironic | |
| 11:24:20 | JayF | (various commits over the last 2-3 cycles to migrate from openstacksdk, completed last cycle) | |
| 11:26:17 | sean-k-mooney | the delta is that nova is using keystoneauth directly in some cases | |
| 11:26:31 | sean-k-mooney | because there were gaps in neutronclient | |
| 11:26:43 | sean-k-mooney | so we needed ot use the keystoneauth client to make raw http requests | |
| 11:27:27 | JayF | Sounds ... fun | |
| 11:27:56 | sean-k-mooney | we also did it to be able to normalise the config | |
| 11:28:22 | sean-k-mooney | i.e. by using keytone auth even with the existing clients it allowed use to have effefctivly the same config section for each of the service clients | |
| 11:28:48 | sean-k-mooney | and auto lookup the urls in the keystone cataloge ectra | |
| 11:29:12 | sean-k-mooney | the question is do we have an example fo confverting that to the sdk yet | |
| 11:33:25 | sean-k-mooney | wncslln: sahid started trying to do the convertion a year ago but didnt get that far https://review.opendev.org/c/openstack/nova/+/882714 | |
| 11:33:59 | sean-k-mooney | wncslln: it might be good to push the WIP patches you ahve so we can review the test failures and advise on how to adapt them | |
| 11:35:07 | wncslln | ack. later I push the WIP and request for review | |
| 11:35:26 | sean-k-mooney | looking at the neutron module | |
| 11:35:40 | sean-k-mooney | we are only using keyston auth in two places | |
| 11:35:48 | sean-k-mooney | and im not entirly sure we need to remove t | |
| 11:36:25 | sean-k-mooney | Uggla is currently adding new usage of the sdk in there manilla series | |
| 11:36:39 | sean-k-mooney | so we shoudl look at how they are intialising the sdk connection ectra | |
| 11:38:23 | wncslln | already have some patch with these changes on manilla? | |
| 11:38:31 | sean-k-mooney | https://review.opendev.org/c/openstack/nova/+/831194/46/nova/share/manila.py | |
| 11:39:00 | sean-k-mooney | so i have not reviewd that yet but that is where they are building the abstraction to talk to manilla using the sdk | |
| 11:39:13 | zigo | frickler: Thanks! :_ | |
| 11:39:15 | sean-k-mooney | and they should be loading auth parmters from the config in that patch i belive | |
| 11:39:57 | sean-k-mooney | wncslln: they are using utils.get_sdk_adapter('shared-file-system', check_service=True) https://review.opendev.org/c/openstack/nova/+/831194/46/nova/share/manila.py#41 | |
| 11:41:02 | sean-k-mooney | wncslln: which if i recall correctly replaces _get_session and https://github.com/openstack/nova/blob/master/nova/network/neutron.py#L261-L268 and https://github.com/openstack/nova/blob/master/nova/network/neutron.py#L70-L88 | |
| 11:41:48 | sean-k-mooney | once we have converted to callign utils.get_sdk_adapter some of the auth exceptions will change form keystoneauth exception to sdk ones i think | |
| 11:42:42 | sean-k-mooney | so we will still need to update tests for that change but that should be fine | |
| 11:42:51 | sean-k-mooney | https://github.com/openstack/nova/blob/master/nova/utils.py#L969-L995 is the defintion of get_sdk_adapter | |
| 11:43:40 | wncslln | in a first moment, I'll try to change to get_sdk_adapter and see what comes, after that I push the WIP | |
| 11:43:46 | wncslln | what you think? | |
| 11:43:54 | sean-k-mooney | sure | |
| 11:43:58 | sean-k-mooney | sound good | |
| 12:49:55 | zigo | "Error starting thread.: nova.exception.InvalidCPUInfo: Configured CPU model: EPYC and CPU Flags [] ar not compatible with host CPU. Please correct your config and try again. Unacceptable CPU info: CPU doesn't have compatibility." | |
| 12:49:55 | zigo | I'm having a hard time setting a CPU model on computes with 2x "AMD EPYC 9734 112-Core Processor" in bookworm (worked with the EPYC-Rome under bullseye). Is this a known issue ? I'm getting: | |
| 12:58:01 | zigo | Even Opteron_G1-v1 doesn't work ... :/ | |
| 13:02:44 | sean-k-mooney | so this depends on your libvirt version and the models it knwo about to some degree at least | |
| 13:02:54 | sean-k-mooney | if you do virsh capabilities | |
| 13:03:00 | sean-k-mooney | spelled correctly of couse | |
| 13:03:14 | sean-k-mooney | then it will print the closet model too the host cpu in the output | |
| 13:03:14 | zigo | <model>EPYC-Milan</model> | |
| 13:03:14 | zigo | <arch>x86_64</arch> | |
| 13:03:14 | zigo | <cpu> | |
| 13:03:14 | zigo | It tells: | |
| 13:03:15 | zigo | <vendor>AMD</vendor> | |
| 13:03:17 | zigo | <microcode version='178258451'/> | |
| 13:03:19 | zigo | <signature family='25' model='160' stepping='2'/> | |
| 13:03:38 | sean-k-mooney | ack so EPYC-Milan is what would be used if you use the default which is host_model | |
| 13:04:24 | zigo | So you'd recommend to use host-model ? | |
| 13:05:45 | sean-k-mooney | well that is the default if you dont set anything (indirectly but its effectivly the same) | |
| 13:06:15 | sean-k-mooney | the installer/package should not be setting a model by default | |
| 13:06:51 | zigo | By default, I'm setting host-passthrough, then things can be overridden on cluster level or per compute level. | |
| 15:39:27 | sean-k-mooney | zigo: i would not ship with host-passthrough, as a default in an installer or package. it will increase performance but severly limit livemigration | |
| 15:39:59 | zigo | sean-k-mooney: Yeah, I may switch to host-model as default. | |