| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2021-01-26 | |||
| 10:22:28 | alexe9191 | @lyar | |
| 10:23:50 | bauzas | sapd1: some folks use nvidia T4s for testing | |
| 10:23:53 | alexe9191 | @lyarwood: Thank you, I was looking for a way to hack the code locally so that I can use the openstack command line tools. But that does not seem to be the smartest course of action here. | |
| 10:23:58 | elod | lyarwood: the patch looks ok to me! | |
| 10:24:04 | elod | +2'd | |
| 10:26:21 | sapd1 | bauzas, Thanks, I will check it. | |
| 10:27:03 | bauzas | sapd1: fwiw, Intel gvt-g cards also work :p | |
| 10:27:24 | MrClayPole | admin0: "virsh list" shows no instances. I can boot the VM, run "dumpxml instance-" it shows http://paste.openstack.org/raw/801977/ and as it boots the 2 shared storage disks so they should be OK? The migration fails with "Error: Failed to perform requested operation on instance " |
|
| 10:27:24 | MrClayPole | requested operation. (HTTP 500) (Request-ID: |
|
| 10:27:31 | bauzas | sapd1: but.... what I can say ? :) | |
| 10:27:39 | sapd1 | bauzas, could you give some recommendations? Is it required license ? | |
| 10:28:03 | bauzas | sapd1: which usecase would you like to do for vGPUs ? | |
| 10:28:05 | admin0 | virsh list --all | |
| 10:28:13 | admin0 | to also show stopped instances | |
| 10:29:45 | bauzas | sapd1: if you want to see my face... https://www.youtube.com/watch?v=sQCgyo2BRe4 | |
| 10:30:29 | bauzas | and for the cards, well... as I said, that depends on which kind of use case you'd like to have | |
| 10:30:38 | bauzas | t4s are cheap | |
| 10:31:10 | sapd1 | bauzas, some workloads such as ML, video transcode.. | |
| 10:31:27 | MrClayPole | Well don't I feel stupid, Thanks for pointing out the --all. This is the output when powered off http://paste.openstack.org/raw/801982/ I've taken a look and can't see anything that would block the migration | |
| 10:31:37 | bauzas | sapd1: then, test with a T4 | |
| 10:33:02 | sapd1 | bauzas, haha. Is it required license for testing ? | |
| 10:33:18 | bauzas | again, there are nvidia docs for this :p | |
| 10:33:40 | lyarwood | MrClayPole: look for req-3fc432cf-4d5c-4d9b-b16e-d06d912e93e7 in your cinder-api / cinder-volume logs | |
| 10:34:30 | lyarwood | MrClayPole: and yeah if all of your data is in these volumes it might be easier to recreate the instance | |
| 10:34:33 | bauzas | sapd1: https://docs.nvidia.com/grid/latest/grid-licensing-user-guide/index.html#license-editions-nvidia-grid | |
| 10:34:47 | lyarwood | MrClayPole: I'm assuming this is the first attempt to cold migrate that failed? | |
| 10:34:54 | lyarwood | MrClayPole: resize sorry | |
| 10:35:47 | bauzas | sapd1: and https://docs.nvidia.com/grid/latest/grid-vgpu-user-guide/index.html#virtual-gpu-types-grid | |
| 10:36:12 | bauzas | and for a T4 https://docs.nvidia.com/grid/latest/grid-vgpu-user-guide/index.html#vgpu-types-tesla-t4 | |
| 10:36:47 | MrClayPole | lyarwood: Yeah, running a cold migration. I'm temped to see this one through rather than go for the recreate as this'll be a good chance for me to start learning the internals of openstack. | |
| 10:37:33 | bauzas | sapd1: if it's just transcoding, a vApps license with a T4-1A vGPU type seems reasonable to me | |
| 10:38:00 | lyarwood | MrClayPole: cool well if you can trace that request-id in the cinder logs and pastebin them I can take a look and walk you through this | |
| 10:38:01 | bauzas | for ML, well, good luck, that highly depends on your training needs | |
| 10:39:08 | lyarwood | elod: do you know if anyone has capped get-pip.py given py27 support has now been dropped? Borked stable/queens AFAICT at least. | |
| 10:40:01 | bauzas | sapd1: but you can run for a short period of time a server without a licensed vGPU https://docs.nvidia.com/grid/latest/grid-licensing-user-guide/index.html#software-enforcement-grid-licensing | |
| 10:45:01 | sapd1 | bauzas, nice. I will request for buying it and try later. :D | |
| 10:45:22 | bauzas | sapd1: ask maybe first a nvidia rep | |
| 10:45:32 | bauzas | I'm not a sales person :) | |
| 10:45:47 | bauzas | and see which kind of license matches your need | |
| 10:46:18 | sapd1 | bauzas, yep. I need time to research on it. If I have any issues, I will ping you. thanks :D | |
| 10:46:21 | frickler | lyarwood: elod: there's some patches up for devstack about get-pip | |
| 10:46:48 | frickler | most of them pushed by elod in fact ;) | |
| 10:47:24 | bauzas | sapd1: again, neither working for nvidia, nor a sales person, but if I can help... | |
| 10:47:35 | sapd1 | bauzas, yeah :D | |
| 10:48:17 | lyarwood | frickler: I had a feeling :D | |
| 10:48:44 | lyarwood | elod++ | |
| 10:48:57 | elod | lyarwood: frickler: yes, I noticed that amotoki had a fix for that and backported it till pike :] | |
| 10:50:09 | lyarwood | do we need a bug? I just wrote https://bugs.launchpad.net/devstack/+bug/1913283 before finding this | |
| 10:50:11 | openstack | Launchpad bug 1913283 in devstack "get-pip.py no longer supports py27" [Undecided,New] | |
| 10:50:12 | elod | it will need merging in reverse order for some patches (where there's grenade) | |
| 10:51:12 | elod | these are the patches btw: https://review.opendev.org/q/topic:%22get-pip-py27%22 | |
| 10:53:30 | lyarwood | seems a little over the top, can't we just set PIP_GET_PIP_URL on the py27 jobs? | |
| 10:59:51 | frickler | lyarwood: then people will complain that their local stuff fails. I personally am all for dropping py2 support completely, but it doesn't seem to be possible to get consensus on that yet | |
| 11:00:28 | lyarwood | frickler: ah fair | |
| 11:33:26 | gibi | stephenfin: hi! if you need me to look at the placement test isolation issue then could you summarize what is the problem? is there a simple reproduction of the problem I can do locally to start with? | |
| 11:40:44 | khomesh24 | #openstack-nova Hello, can someone check the bug https://bugs.launchpad.net/nova/+bug/1913016? | |
| 11:40:45 | openstack | Launchpad bug 1913016 in OpenStack Compute (nova) "nova api os-resetState should not reset the state when VM is shelved_offloaded" [Undecided,New] - Assigned to Khomesh Thakre (khomesh24) | |
| 11:42:28 | sean-k-mooney | gibi: i think the issue is each of the gabby test files are written such that they need to be run indepently | |
| 11:42:47 | lyarwood | khomesh24: that looks valid to me, did you have a patch? | |
| 11:43:53 | sean-k-mooney | for example https://github.com/openstack/placement/blob/master/placement/tests/functional/gabbits/shared-resources.yaml | |
| 11:44:08 | sean-k-mooney | needs to be run with its own placment instance | |
| 11:46:18 | sean-k-mooney | although that is proably not the best example since it wont conflict with others since other dont create inventoies of CUSTOM_MAGIC | |
| 11:47:27 | stephenfin | gibi: I just figured it out | |
| 11:47:31 | stephenfin | https://storyboard.openstack.org/#!/story/2008556 | |
| 11:48:01 | stephenfin | pushing a fix shortly | |
| 11:50:27 | gibi | sean-k-mooney, stephenfin: thanks, then I will not dig into it | |
| 11:51:26 | sean-k-mooney | stephenfin: i assume that is because we are using a global object at some point and not mockign it propertly in the tests | |
| 11:52:26 | stephenfin | yes, the RuleDefault objects provided in placement.policies.* | |
| 11:59:24 | gibi | sean-k-mooney: replied in https://review.opendev.org/c/openstack/nova/+/757109 I think you missed the @retries decorator | |
| 12:00:00 | khomesh24 | lyarwood, no i wanted to valid first | |
| 12:07:08 | lyarwood | khomesh24: kk for me it's valid, I've marked the bug as confirmed | |
| 12:07:22 | lyarwood | khomesh24: it doesn't make sense to allow an admin to reset to active if the instance is offloaded | |
| 12:08:56 | khomesh24 | lyarwood, ack, i will start working on the patch then, thank you | |
| 12:12:21 | gibi | lyarwood, khomesh24: do we then make os-resetState smart for other non valid reality-db state combinations too? | |
| 12:13:14 | gibi | for me os-resetState is a blunt instrument, and the implementation also suggest that (it validates nothing) | |
| 12:13:40 | lyarwood | gibi: I wouldn't be opposed to it tbh, anything we can do to avoid obvious PEBKAC is good with me | |
| 12:14:19 | gibi | lyarwood: so when the admin reset to ACTIVE then we should call down to the compute to validate that the instance is really running happily? | |
| 12:14:54 | lyarwood | gibi: no I'd just sanity check the vm_state | |
| 12:15:07 | lyarwood | gibi: SHELVED_OFFLOADED to ACTIVE obviously isn't valid | |
| 12:16:20 | lyarwood | gibi: https://docs.openstack.org/api-ref/compute/?expanded=reset-server-state-os-resetstate-action-detail#reset-server-state-os-resetstate-action the docs are also wrong, you can set any valid vm_state AFAICT | |
| 12:16:21 | gibi | lyarwood: is it not valid as ACTIVE would need instance.host set but we assume that in SHELVED_OFFLOADED the instance.host is None? | |
| 12:16:46 | lyarwood | gibi: correct, any further action would just fail | |
| 12:17:07 | lyarwood | gibi: I don't want to block admins from fixing actual issues with instances | |
| 12:17:27 | lyarwood | gibi: but I can't see how allowing SHELVED_OFFLOADED to ACTIVE would help in any situation tbh | |
| 12:17:47 | lyarwood | gibi: lets wait for a patch to emerge and we can thrash this out on the review | |
| 12:17:52 | gibi | lyarwood: the api schema restricts the new state to enum': ['active', 'error'] | |
| 12:17:56 | lyarwood | ah | |
| 12:18:12 | lyarwood | thanks | |
| 12:18:35 | gibi | lyarwood: OK, then I would not depend on SHELVED_OFFLOADED but depend on instance.host is not None, when resetting to ACTIVE | |
| 12:18:55 | gibi | that would catch ERROR -> ACTIVE for VM that was never scheduled to a host | |
| 12:19:05 | gibi | (and other edge cases, I guess) | |
| 12:20:03 | gibi | lyarwood: also I would add warning to the API ref that resetting state is dangerous | |
| 12:25:01 | openstackgerrit | Stephen Finucane proposed openstack/placement master: policy: Suppress policy deprecation warnings https://review.opendev.org/c/openstack/placement/+/771852 | |
| 12:25:02 | openstackgerrit | Stephen Finucane proposed openstack/placement master: Implement secure RBAC for resource providers https://review.opendev.org/c/openstack/placement/+/760240 | |
| 12:25:03 | openstackgerrit | Stephen Finucane proposed openstack/placement master: Implement secure RBAC for aggregates https://review.opendev.org/c/openstack/placement/+/760235 | |
| 12:25:05 | openstackgerrit | Stephen Finucane proposed openstack/placement master: Implement secure RBAC for allocations https://review.opendev.org/c/openstack/placement/+/760236 | |
| 12:25:05 | openstackgerrit | Stephen Finucane proposed openstack/placement master: Implement secure RBAC for allocation candidates https://review.opendev.org/c/openstack/placement/+/760237 | |
| 12:25:07 | openstackgerrit | Stephen Finucane proposed openstack/placement master: Implement secure RBAC for inventories https://review.opendev.org/c/openstack/placement/+/760238 | |
| 12:25:08 | openstackgerrit | Stephen Finucane proposed openstack/placement master: Implement secure RBAC for resource classes https://review.opendev.org/c/openstack/placement/+/760239 | |
| 12:25:09 | openstackgerrit | Stephen Finucane proposed openstack/placement master: Implement secure RBAC for traits https://review.opendev.org/c/openstack/placement/+/760241 | |