| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2024-03-18 | |||
| 14:45:46 | artom | bauzas, I think gibi's saying that as long as it gets backported to Antelope, he's happy, regardless if it's merged on master in C or D | |
| 14:45:55 | bauzas | my concern is that we couldn't backport https://review.opendev.org/c/openstack/nova/+/903706/2/nova/objects/migrate_data.py#246 | |
| 14:46:09 | gibi | artom: yes, after RC1 we will have master open for D, but I guess bauzas wants to limit the changes on master until final RC to make sure any last minute change to C is easy to backport from masterj | |
| 14:46:25 | artom | bauzas, right, that's why there's a backportable hack below it in https://review.opendev.org/c/openstack/nova/+/877773/9 | |
| 14:46:38 | bauzas | I just +2d both of the changes | |
| 14:47:17 | artom | bauzas, err, want to jump on a call for me to explain stuff? | |
| 14:48:08 | bauzas | guys, that's like 4 days i told about rc1 | |
| 14:48:12 | bauzas | I wonder what you guys want | |
| 14:49:44 | gibi | I want to enable powermgmt in antelope :) | |
| 14:51:18 | gibi | backporting the fix from C or D does not matter for mej | |
| 15:13:55 | dansmith | bauzas: I'm trying to decide if that cpu shared set thing is okay or not | |
| 15:14:12 | dansmith | I don't love deleting a non-versioned property and replacing it with a versioned one like it's doing | |
| 15:14:29 | dansmith | I'm trying to decide if it's actually a problem, or just _very_ confusing (the latter is definitely true, IMHO | |
| 15:16:35 | dansmith | ah, basically what sean-k-mooney said on PS1 | |
| 15:18:06 | bauzas | dansmith: gibi: artom: Uggla: okay, so now I understand more | |
| 15:18:14 | artom | dansmith, heh, bauzas and I were just on a call where I explained the nuances | |
| 15:18:22 | artom | I can do the same for you if you wish :) | |
| 15:18:44 | dansmith | bummer | |
| 15:18:58 | bauzas | so, basically, we have an already bug with live-migration for cpu_shared_set | |
| 15:19:16 | dansmith | yeah I get that | |
| 15:19:51 | dansmith | how far back does it need to be backported? | |
| 15:19:52 | bauzas | now with power management, you can live-migrate an instance that will use a wrong cpu which is a dedicated one, like before, but which will be off | |
| 15:20:14 | bauzas | my point is that I don't think it's a RC1 needed bugfix | |
| 15:20:56 | bauzas | what we can rather do is to merge https://review.opendev.org/c/openstack/nova/+/877773 in Dalmatian and then backport it down to Antelope if we want | |
| 15:21:09 | bauzas | https://review.opendev.org/c/openstack/nova/+/903706/ won't be backported | |
| 15:21:20 | bauzas | (as it's a new object version) | |
| 15:21:58 | bauzas | so I'm just asking Uggla to modify https://review.opendev.org/c/openstack/nova/+/877773 to make sure it fixes the wrong functional test which is https://github.com/openstack/nova/blob/master/nova/tests/functional/regressions/test_bug_1899835.py | |
| 15:22:24 | bauzas | (the last assert is wrong in the regression bug) | |
| 15:22:58 | bauzas | (school) | |
| 15:28:46 | gibi | elodilles: could you look at this b | |
| 15:28:57 | gibi | ... this backport series https://review.opendev.org/c/openstack/nova/+/913198 | |
| 16:03:47 | bauzas | gibi: dansmith: we need approval for the prelude patch https://review.opendev.org/c/openstack/nova/+/912973 | |
| 16:03:47 | elodilles | gibi: ACK, i'll try to look into it when i get there | |
| 16:04:10 | dansmith | ack | |
| 16:04:30 | bauzas | thanks | |
| 16:07:17 | gibi | elodilles: thanks | |
| 16:22:47 | dansmith | bauzas: done | |
| 16:22:54 | bauzas | cool thanks ! | |
| 16:26:14 | opendevreview | Sylvain Bauza proposed openstack/nova master: Add a Caracal prelude section https://review.opendev.org/c/openstack/nova/+/912973 | |
| 16:26:21 | bauzas | dansmith: ^ (when you have time) | |
| 16:58:30 | opendevreview | Merged openstack/nova master: vgpu: Allow device_addresses to not be set https://review.opendev.org/c/openstack/nova/+/902084 | |
| 17:18:46 | zigo | and yet another one another one (twice "another one") | |
| 17:18:46 | zigo | bauzas: Typ3 on the last 3 lines: | |
| 17:19:29 | zigo | Live-mig of vGPU is a nice feature, thanks! | |
| 17:20:02 | zigo | Also, good thing for SEV. We were looking at implementing it, now I'll make sure we do that over Caracal! :) | |
| 18:35:42 | opendevreview | Merged openstack/nova stable/2023.2: Add cpuset_reserved helper to instance NUMA topology https://review.opendev.org/c/openstack/nova/+/913193 | |
| 18:35:48 | opendevreview | Merged openstack/nova stable/2023.2: Reproducer for not powering on isolated emulator threads cores https://review.opendev.org/c/openstack/nova/+/913194 | |
| 18:59:52 | opendevreview | Sylvain Bauza proposed openstack/nova master: Add a Caracal prelude section https://review.opendev.org/c/openstack/nova/+/912973 | |
| 19:00:45 | bauzas | dansmith: melwitt: could you please quickky approve now the prelude above $ so I could provide a new RC1 release patch version tonight? | |
| 19:10:29 | dansmith | bauzas: done | |
| 19:17:02 | bauzas | dansmith: melwitt: ta folks, as soon as it merges, I'll update the RC1 patch | |
| 19:46:11 | opendevreview | Merged openstack/nova master: Add a Caracal prelude section https://review.opendev.org/c/openstack/nova/+/912973 | |
| 22:15:44 | opendevreview | Merged openstack/nova stable/zed: Reproducer for bug 2003991 unshelving offloaded instance https://review.opendev.org/c/openstack/nova/+/888093 | |
| #openstack-nova - 2024-03-19 | |||
| 10:20:57 | opendevreview | OpenStack Release Bot proposed openstack/nova stable/2024.1: Update .gitreview for stable/2024.1 https://review.opendev.org/c/openstack/nova/+/913631 | |
| 10:21:00 | opendevreview | OpenStack Release Bot proposed openstack/nova stable/2024.1: Update TOX_CONSTRAINTS_FILE for stable/2024.1 https://review.opendev.org/c/openstack/nova/+/913632 | |
| 10:21:05 | opendevreview | OpenStack Release Bot proposed openstack/nova master: Update master for stable/2024.1 https://review.opendev.org/c/openstack/nova/+/913633 | |
| 10:35:26 | opendevreview | Rocky proposed openstack/nova master: Fix the command to list hw_machine_type unset instances https://review.opendev.org/c/openstack/nova/+/913652 | |
| 10:48:21 | bauzas | guys, please welcome our master branch that's now a Dalmatian :) | |
| 10:48:46 | bauzas | so long Caracal and thanks for the fish | |
| 10:49:14 | bauzas | (RC1 tag is set and caracal branch created) | |
| 10:53:25 | gibi | bauzas: \o/ | |
| 11:15:46 | sean-k-mooney | :) | |
| 12:00:53 | opendevreview | Fabian Wiesel proposed openstack/nova master: Do not close returned image-chunk iterator & get_verifier early https://review.opendev.org/c/openstack/nova/+/909474 | |
| 13:18:57 | fwiesel | bauzas: I've reopened the bug #2055700 as #2058225. In my mind _rebuild_default_impl is implemented libvirt specific. I can implement driver.rebuild for vmwareapi to work around the bug, but I don't think that would be a great way forward. | |
| 13:21:03 | opendevreview | Merged openstack/nova stable/2023.2: Power on cores for isolated emulator threads https://review.opendev.org/c/openstack/nova/+/913195 | |
| 13:40:59 | bauzas | fwiesel: cool | |
| 13:44:52 | fwiesel | But I do not know enough how it is supposed to work in libvirt to make a good suggestion on how to implement it differently. I mean, I assume it makes sense that it is implemented the way it is. | |
| 13:53:08 | bauzas | fwiesel: you're referring to https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L4132-L4137 | |
| 14:00:25 | bauzas | fwiesel: _rebuild_default_impl is a virt-agnostic solution for rebuild which does delete the old guest and then create a new one https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L3648-L3731 | |
| 14:02:52 | gibi | elodilles: thanks for the reviews so far. the parent of https://review.opendev.org/c/openstack/nova/+/913196 is merged now. So when you have time please check this backport too | |
| 14:02:56 | bauzas | as you can read, we power off the guest, detach its volumes, (reimage the root disk if volume-backed), destroy the guest, attach the volumes again and spawn the guest | |
| 14:14:52 | sean-k-mooney | fwiesel: the way its currently implemtned shoudl wor for vmware too | |
| 14:15:22 | sean-k-mooney | fwiesel: the only driver that does anythign custom is ironic | |
| 14:16:09 | sean-k-mooney | the vmware dirver can provide its own solution | |
| 14:17:43 | elodilles | gibi: ACK, will do | |
| 14:18:44 | sean-k-mooney | the imporant thing to remember about rebuild is rebuild is not a move operations, evacuate is a rebuild to a diffent host but by default a normal rebuild shoudl never change host. if you change image however you do need to call the schduler with the current host adn the existin flavor and new iamge to confirm that host is valid. that should all be handeled for you in the | |
| 14:18:46 | sean-k-mooney | generic compute manager code so the only thing the driver level rebuidl needs to do is ensure the root disk is reimaged | |
| 14:19:35 | sean-k-mooney | if there is a more effiecnt way to do that in vmware then the generic approch then that is fine to use | |
| 14:20:05 | dansmith | not "should never change host" but "does not change host" | |
| 14:20:14 | dansmith | even the driver shouldn't be able to override that | |
| 14:22:08 | sean-k-mooney | right | |
| 14:22:31 | sean-k-mooney | im just tyring to say that if there was to be a vmware version fo rebuidl in the vmware driver the only thing it should od is reimage the disk | |
| 14:22:47 | sean-k-mooney | the vm must remain on the same hypervior host | |
| 14:23:14 | sean-k-mooney | i just mentioned evcautate because the driver code will have to work when called in that code path too | |
| 14:23:26 | dansmith | yep | |
| 14:27:56 | fwiesel | sean-k-mooney: The current implementation calls driver.destroy, and *then* calls detach volume. In my mind, that shouldn't work in the general case. | |
| 14:33:03 | bauzas | no | |
| 14:33:13 | bauzas | fwiesel: https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L3675-L3676 | |
| 14:33:44 | bauzas | poweroff, detach, destroy is the workflow | |
| 14:34:38 | fwiesel | Destroy: https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L3695-L3701 and then via https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L3713-L3715 comes https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L3589-L3595 | |
| 14:35:43 | fwiesel | bauzas: The root volume is being detached after the instance has been destroyed if reimage_boot_volume and is_volume_backed | |
| 14:36:26 | fwiesel | So, poweroff, detach everything except the root-disk, destroy, detach the root_disk (in the case of reimage_boot_volume and is_volume_backed) | |
| 14:36:31 | bauzas | we still detach the volumes, including the root bdm in https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L3675-L3676 | |
| 14:37:49 | fwiesel | Sure, and after the desotry the _rebuild_volume_backed_instance -> _detach_root_volume | |
| 14:40:39 | fwiesel | Well, we have `reimage_boot_volume=True` ergo `detach_root_bdm = not reimage_boot_volume = False`, so `detach_block_devices(..., detach_root_bdm=False)`. | |
| 14:51:09 | bauzas | fwiesel: ah you're right, it's a bool | |
| 14:51:25 | bauzas | okay, so the root volume isn't detached yet, my bad | |
| 14:53:24 | bauzas | and ack, we rebuild the root volume in _rebuild_volume_backed_instance() | |
| 14:55:26 | bauzas | where we detach the volume, delete the old attachment, and then reimage the volume thru cinder | |
| 14:56:13 | bauzas | interestingly enough, we destroy the guest indeed before the detach, but we skip its bdm when we call the driver destroy | |