Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2021-06-01
10:55:18 sean-k-mooney lyarwood: ya when i use it its oftehn to fix an unitential rebase with git reivew
10:55:25 sean-k-mooney gibi: yep
10:55:51 sean-k-mooney gibi: nova has a full copy of the openwrt sorce tree in it somewhere
10:55:57 gibi at some point I had placement back in nova :D
10:56:36 sean-k-mooney at least the gerrit version of it had at one point we may have eventually git gc that out of the public repos
10:57:13 gibi ohh so the central copy of nova had openwrt? nice!
10:57:48 lyarwood sounds like some kind of go project repo
10:58:18 sean-k-mooney yep thats something jaypipes told me a long time ago. i assume someone pushed a review where tehy acindtally commited it locally
10:58:40 sean-k-mooney but gerrit would keep that around forever as a result in the gerrit copy of the repo
10:59:20 sean-k-mooney you would have to manually purge the review ref to remove it
11:08:39 opendevreview Stephen Finucane proposed openstack/nova master: Use neutronclient's port binding APIs https://review.opendev.org/c/openstack/nova/+/706295
11:13:03 opendevreview Stephen Finucane proposed openstack/nova master: docs: Drop references to non-filter scheduler drivers https://review.opendev.org/c/openstack/nova/+/773645
11:13:04 opendevreview Stephen Finucane proposed openstack/nova master: tests: Merge 'test_utils', 'test_scheduler_utils' https://review.opendev.org/c/openstack/nova/+/773646
11:13:04 opendevreview Stephen Finucane proposed openstack/nova master: scheduler: Merge driver into manager https://review.opendev.org/c/openstack/nova/+/773644
11:13:05 opendevreview Stephen Finucane proposed openstack/nova master: conf: Remove deprecated aliases https://review.opendev.org/c/openstack/nova/+/773647
11:38:57 opendevreview norman shen proposed openstack/nova master: Saving security group to info_cache https://review.opendev.org/c/openstack/nova/+/786348
11:42:15 stephenfin dead simple Python 3.10 prep patch here if anyone has 2 mins https://review.opendev.org/c/openstack/nova/+/790405
11:45:15 gibi stephenfin: done
11:45:18 stephenfin ta
12:15:20 opendevreview Lee Yarwood proposed openstack/nova stable/wallaby: hardware: Use image_meta.id within get_mem_encryption_constraint https://review.opendev.org/c/openstack/nova/+/793956
12:16:29 opendevreview Lee Yarwood proposed openstack/nova stable/victoria: hardware: Use image_meta.id within get_mem_encryption_constraint https://review.opendev.org/c/openstack/nova/+/793957
12:17:15 opendevreview Lee Yarwood proposed openstack/nova stable/ussuri: hardware: Use image_meta.id within get_mem_encryption_constraint https://review.opendev.org/c/openstack/nova/+/793958
13:07:29 ganso Hi nova folks! Not sure if you're familiar with this behavior, whether it is a known bug or limitation (I didn't find any launchpad entry for it), but it is very easy to reproduce: if you try attach a volume with hw_disk_bus property that is different from the root disk's hw_disk_bus one, it gets ignored. Example, root disk's is ide, volume's is virtio, and it gets attached as ide.
13:08:16 ganso using pure kvm through virt-manager, it is possible to have mixed ide+virtio disks, but nova doesn't seem to be allowing that
13:08:32 sean-k-mooney ganso: yes that is expected behavior
13:08:44 sean-k-mooney ganso: the image metadta is only use form the root disk
13:08:57 sean-k-mooney we do not use image metadta form any other disk
13:09:11 sean-k-mooney ganso: you might be able to enable this using the block device mappings api
13:09:39 ganso sean-k-mooney: is there a reason why you would want to keep it this way?
13:10:28 sean-k-mooney well if the bus is not already present in the vm at a minitum we woudl have to also attach a contole for that bus be it ide/stata whatever
13:10:38 sean-k-mooney and then all attach the volume to that new contoler
13:10:55 sean-k-mooney that could fail for a number of reasons
13:11:24 sean-k-mooney in general image metadata was intened to only be used form teh root disk
13:11:56 sean-k-mooney when not using cinder it does not really make sense to have multipel image metadta soruce
13:12:14 ganso sean-k-mooney: hmmm I see, so this is like a new feature, to handle independent disk_bus values for each new volume, something nova doesn't do today
13:12:34 sean-k-mooney if we allowed it to work for cinder there is an issue with confilt between volumens that we would have to determin how to handel
13:12:40 ganso sean-k-mooney: I was able to hack the code just to test it, I replaced the variable with "virtio" here: https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L2018
13:13:04 sean-k-mooney ganso: yes it would basiclaly be a new feature
13:13:06 ganso and it worked, upon attach it added a virtio disk beside an ide one (the root disk)
13:13:28 sean-k-mooney yes there is a virtio contoler present by defualt for the nic
13:13:37 sean-k-mooney if you put stata it will fail
13:13:55 sean-k-mooney /stata/sata
13:14:53 ganso sean-k-mooney: so, adding a controller during attach is something risk you say? would it be 100% safe if it requires the VM to be shutoff?
13:15:15 sean-k-mooney it may or may not fail
13:15:43 sean-k-mooney and it may depend on the version fo libvirt if the contoler is hot pluggable or not
13:16:20 ganso sean-k-mooney: I understand. I will do some testing around this, thank you for the clarification!
13:16:24 sean-k-mooney for example when using q35 machine type libvirt only alloctes 1 spare pcie port for hotplug devices
13:16:37 sean-k-mooney so you would not have enough to add the contoler and a volume
13:16:54 sean-k-mooney there is one way you might be able to do this today
13:17:05 sean-k-mooney but only for new vms
13:17:07 sean-k-mooney https://docs.openstack.org/api-ref/compute/?expanded=create-server-detail#create-server
13:17:25 sean-k-mooney server create allows you to specify the block device mappings block_device_mapping_v2 (Optional)
13:17:31 sean-k-mooney "block_device_mapping_v2": [{
13:17:33 sean-k-mooney "boot_index": "0",
13:17:35 sean-k-mooney "uuid": "ac408821-c95a-448f-9292-73986c790911",
13:17:37 sean-k-mooney "source_type": "image",
13:17:38 ganso sean-k-mooney: oh I see, it would depend on that as well, so depending on how I test it, I may be avoiding a problem (like testing with just 1 extra disk instead of 2 or 3, as it could run out of devices)
13:17:39 sean-k-mooney "volume_size": "25",
13:17:41 sean-k-mooney "destination_type": "volume",
13:17:43 sean-k-mooney "delete_on_termination": true,
13:17:45 sean-k-mooney "tag": "disk1",
13:17:47 sean-k-mooney "disk_bus": "scsi"}]
13:17:49 sean-k-mooney which has the disk bus
13:18:07 sean-k-mooney ganso: yes although we have a config option for how may free port to allcote
13:18:18 ganso sean-k-mooney: I will try that! could just work out for my use case! thank you!
13:18:47 sean-k-mooney ganso: if that works and you can set the disk bus per volume when creatign a new server
13:18:50 sean-k-mooney which the api imples
13:19:03 sean-k-mooney then we could consider if that would be extanded to port attach
13:20:01 sean-k-mooney that coudl be done by either alowing the disk bus to be specified on attach or possibel adding support for a subset of data to be stored in teh block device mapping form the volume metadtata
13:20:41 sean-k-mooney e.g. copy just the hw_disk_bus into the bdm that we create on attach
13:21:08 ganso sean-k-mooney: well, even if that works for the "create" operation, it is a more limited one, as you said, when creating the VM it is a safer scenario (not hotpluggable, can determine the required controllers before-hand). The attach operation is less safe in this regard and could fail in many ways, as you said
13:21:14 sean-k-mooney ganso: this is a feature that would require a spec and someone more expert in the strage layware like lyarwood to review
13:22:32 sean-k-mooney ganso: the failure could actully be handeled internally in the driver just roleing back and rejecting the attach but we would have to code with that in mind and think about the ux and error message we give to a user
13:23:05 sean-k-mooney whihc is one of the reason i think this woudl need a spec rather then trateing it as a bugfix
13:24:10 ganso sean-k-mooney: yea, just like the hotplugged error message for IDE devices, I was confused at first when the attach didn't work while the request was accepted, but I had no obvious error message
13:24:20 sean-k-mooney ganso: the driver code in general assumes there is only one value for hw_disk_bus for the entire vm so thre are proably edgcases like rescue and move operations that might break.
13:24:21 ganso sean-k-mooney: I agree! +1
13:30:10 ganso sean-k-mooney: btw, if you have a minute, that patch you previously reviewed already has a +2. I was wondering if you could please take another look at it when you have a minute, since you're already familiar with it: https://review.opendev.org/c/openstack/nova/+/784166
13:36:11 sean-k-mooney ah yes am i do not have +2 rights on nova but ill take a look now and maybe poke some of the nova cores to review when im done
13:49:14 opendevreview Merged openstack/nova master: Add unit test for importing ed25519 ssh key https://review.opendev.org/c/openstack/nova/+/789642
13:50:12 opendevreview Merged openstack/nova master: Change minversion of tox to 3.18.0 https://review.opendev.org/c/openstack/nova/+/791968
13:51:11 sean-k-mooney bauzas: you like anit-afintiy groups right :P care to review ganso's patch https://review.opendev.org/c/openstack/nova/+/784166
13:51:53 opendevreview Merged openstack/nova master: Remove unused DeleteFromSelect db api method https://review.opendev.org/c/openstack/nova/+/784528
13:52:14 bauzas sean-k-mooney: not sure it's correct to say "I like" for the instance groups but sure ;)
13:52:50 sean-k-mooney hehe that was ment to be drowning in sarcasim but i for got to anotate that :)
13:56:27 sean-k-mooney do we have quota for how may revision of a patch you are allow to have without unit tests passing...
13:57:28 sean-k-mooney i fell like if we did its less then 51 in 6 weeks
14:01:00 bauzas ganso: I provided another +2 but please look at my comment in https://review.opendev.org/c/openstack/nova/+/784166
14:01:49 bauzas once you look at it, tell me and then I'll +W your change
14:02:30 sean-k-mooney bauzas: yes adding an explaintion would be nice to avoid the tribal knoladge that live migration never reschdules
14:02:44 bauzas can be a follow-ip
14:02:46 bauzas up*
14:03:09 sean-k-mooney also yes i agree it can be a followup
14:03:19 bauzas ganso: do you want to backport your change, actually ?
14:03:38 bauzas if so, maybe we would need to have a new revision for the relnote
14:03:54 sean-k-mooney we can backport the followup too
14:04:01 sean-k-mooney just and fyi
14:04:03 bauzas or squashing both when backporting but...
14:04:05 sean-k-mooney reno is ok with that
14:04:13 sean-k-mooney no need to squash

Earlier   Later