Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2025-06-19
16:31:38 gibi sean-k-mooney: FYI I think this will be the first use of the round robin placement a_c strategy out in the field https://bugs.launchpad.net/nova/+bug/2114947
16:37:23 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for images APIs https://review.opendev.org/c/openstack/nova/+/952284
16:37:24 opendevreview Stephen Finucane proposed openstack/nova master: api: Separate volume, snapshot and volume attachments https://review.opendev.org/c/openstack/nova/+/952347
16:37:24 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for volumes APIs https://review.opendev.org/c/openstack/nova/+/952348
16:37:25 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for snapshots APIs https://review.opendev.org/c/openstack/nova/+/952349
16:37:26 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for volume attachments APIs https://review.opendev.org/c/openstack/nova/+/952350
16:37:26 opendevreview Stephen Finucane proposed openstack/nova master: tests: Use valid UUIDs for cinder resources https://review.opendev.org/c/openstack/nova/+/952935
16:37:27 opendevreview Stephen Finucane proposed openstack/nova master: api: Only apply "soft" additionalProperties validation to requests https://review.opendev.org/c/openstack/nova/+/952936
16:41:55 opendevreview Stephen Finucane proposed openstack/nova master: api: Correct expected errors https://review.opendev.org/c/openstack/nova/+/951640
16:46:09 sean-k-mooney gibi: di it result in a bug or is it the fix for the bug
16:47:07 sean-k-mooney Uggla: i can set it as approved if you want in launchpad. at least in this case
16:47:48 sean-k-mooney updated https://blueprints.launchpad.net/nova/+spec/automatic-memballoon-freeing
16:48:23 Uggla sean-k-mooney, no it is ok, when I will be able to do it, I will check all approved SLBP and set them accordingly.
16:51:35 gibi sean-k-mooney: they just hit the bug, I suggested to tune placement, we will see if this will help them
16:51:35 sean-k-mooney gibi: ah you saying it can be fixed by it, i think
16:51:52 gibi the reporter is pretty responsive so hopefully we get feedback
16:52:10 sean-k-mooney cool
16:52:32 sean-k-mooney im not entirly sur if this is the same since it appears to hvae allcoated a device with the worng? resouce class
16:52:45 gibi which reminds me that we still not have upstream testing with breadth-first strategy due to the blocked devstack patch I have to go back to
16:53:05 sean-k-mooney oh they are not doing it properly
16:53:07 gibi sean-k-mooney: originally they had a config issue not having pci_in_placement enabled
16:53:16 sean-k-mooney wel
16:53:17 gibi now they enabled it and hitting timeout
16:53:25 sean-k-mooney they seam to have both a cutom trait and resouce class
16:54:03 sean-k-mooney device_spec = { "vendor_id":"10de", "product_id":"26b1", "address":"0000:4F:00.4", "resource_class":"CUSTOM_NVIDIA_RTX6000_ADA_48Q", "traits":"CUSTOM_NVIDIA_RTX6000_ADA_48Q", "managed":"no" }
16:54:19 sean-k-mooney they shoudl just be useign the reouces class not the triat
16:54:35 sean-k-mooney it wont break anything to have both
16:54:57 gibi yeah
16:55:05 gibi we can follow up with that once they see it working
16:55:41 sean-k-mooney but the other thing they are trying to do that kind fo unsupproted
16:56:05 sean-k-mooney is allcoatign mutlipel vFs
16:56:20 sean-k-mooney so these are gpus
16:56:33 sean-k-mooney and they are suing pci passthough to use the VFs
16:56:53 sean-k-mooney which i guess means they are trying to use the new way of doing vgpu that we added last cycle
16:57:07 sean-k-mooney with managed = no
16:57:24 sean-k-mooney but whiel you shoudl be able to have mulitpel devices in one vm
16:57:28 sean-k-mooney we have not actully tested that
16:58:10 sean-k-mooney i guess there is no really reason form a nova perspecitve why that woudl not work
16:58:20 sean-k-mooney we know it did o r does work for generic vfs
16:58:42 sean-k-mooney so the only reaosn for it not to work with gpus woudl eb due to a hardware/driver limitaiton
16:59:09 sean-k-mooney oh ok they are doign bad things
16:59:14 sean-k-mooney openstack flavor create 8xRTX-ADA-48Q --private \
16:59:16 sean-k-mooney --ram 4096 --vcpu 4 --disk 0 \
16:59:18 sean-k-mooney --property "resources:CUSTOM_NVIDIA_RTX6000_ADA_48Q"=1 \
16:59:20 sean-k-mooney --property "trait:CUSTOM_NVIDIA_RTX6000_ADA_48Q"="required" \
16:59:22 sean-k-mooney --property "pci_passthrough:alias"="rtx6000-ada-48q:8"
16:59:24 sean-k-mooney openstack flavor set --project admin 8xRTX-ADA-48Q
16:59:45 sean-k-mooney they are requesting 8 of the gpus VF via the alisa btu also orverwriting the resouce: and traits request
17:00:05 sean-k-mooney they shoudl not be settign
17:00:07 sean-k-mooney --property "resources:CUSTOM_NVIDIA_RTX6000_ADA_24Q"=1 \
17:00:09 sean-k-mooney --property "trait:CUSTOM_NVIDIA_RTX6000_ADA_24Q"="required" \
17:00:22 sean-k-mooney or well the relevent lines form the same flavor
17:16:52 opendevreview Merged openstack/nova stable/2024.2: [tool] Fix backport validator for non-SLURP https://review.opendev.org/c/openstack/nova/+/951969
17:16:55 gibi yepp they should not set the rc and trait on the flavor just in the alias
17:24:28 sean-k-mooney i commented on the bug to that effect
21:50:33 opendevreview sean mooney proposed openstack/os-resource-classes master: Add VCPU_SHARES resource class for CPU performance tiering https://review.opendev.org/c/openstack/os-resource-classes/+/952951
22:51:17 opendevreview Merged openstack/nova master: Fix disable memballoon device https://review.opendev.org/c/openstack/nova/+/945621
#openstack-nova - 2025-06-20
01:17:19 opendevreview Takashi Kajinami proposed openstack/nova master: Drop explicit setting of iommu for AMD SEV support https://review.opendev.org/c/openstack/nova/+/909635
05:28:12 opendevreview Takashi Kajinami proposed openstack/nova master: Drop explicit setting of iommu for AMD SEV support https://review.opendev.org/c/openstack/nova/+/909635
06:46:32 Uggla sean-k-mooney, gibi "but whiel you shoudl be able to have mulitpel devices in one vm", "we have not actully tested that", in fact I tested that. It works well if the vgpu type supports multiple gpus. There is an xml file provided by nvidia with a propserty defining if multiple gpus are supported.
07:14:57 opendevreview Rajesh Tailor proposed openstack/nova stable/2025.1: Fix disable memballoon device https://review.opendev.org/c/openstack/nova/+/952959
07:16:01 opendevreview Rajesh Tailor proposed openstack/nova stable/2024.2: Fix disable memballoon device https://review.opendev.org/c/openstack/nova/+/952960
07:17:20 opendevreview Rajesh Tailor proposed openstack/nova stable/2024.1: Fix disable memballoon device https://review.opendev.org/c/openstack/nova/+/952961
08:10:36 opendevreview Merged openstack/nova stable/2024.1: [tool] Fix backport validator for non-SLURP https://review.opendev.org/c/openstack/nova/+/951970
09:30:02 opendevreview Balazs Gibizer proposed openstack/nova master: Fix neutron client dict grabbing https://review.opendev.org/c/openstack/nova/+/952966
09:30:29 gibi bauzas: sean-k-mooney: ^^ this fix is needed to make nova work in py3.13
09:30:55 gibi cc jkulik: zigo: ^^
09:34:40 opendevreview Balazs Gibizer proposed openstack/nova master: Fix neutron client dict grabbing https://review.opendev.org/c/openstack/nova/+/952966
09:39:50 opendevreview Balazs Gibizer proposed openstack/nova master: Fix neutron client dict grabbing https://review.opendev.org/c/openstack/nova/+/952966
10:36:30 sean-k-mooney gibi: so i wondered about doing something like that as well but i expected the behaivor to be the same __getattr__ is more convetional so that nice
10:36:51 sean-k-mooney have you tested this in devstack?
10:37:56 sean-k-mooney ill quickly rebase my test patch but it was faillng on a weired mysl isse so its proably goign to fail for unrelated reasons
10:41:47 gibi sean-k-mooney: yepp I tested it in devstack locally
10:42:03 opendevreview sean mooney proposed openstack/nova master: [DNM] testing py3.13 eventlet bug workaround https://review.opendev.org/c/openstack/nova/+/951749
10:42:37 gibi the cpython bug is triggered by having two refs on the internal __dict__ so __getattr__ removes that
10:42:39 sean-k-mooney gibi: ok i was tryign to test it with ^ but for some reason i was getting mysql issues iwth the iniall root passward config
10:43:02 sean-k-mooney gibi: i see how did you find that
10:43:47 sean-k-mooney did you find the cpython bug through searching the web
10:44:10 gibi it wasn't me. It was jkulik who managed to create the small reproduction and then engaged with the python community
10:44:42 gibi the discussion happened on eventlet-removal channel as first it was thought that this is yet another eventlet induced bug
10:45:29 opendevreview Stephen Finucane proposed openstack/nova master: api: Only apply "soft" additionalProperties validation to requests https://review.opendev.org/c/openstack/nova/+/952936
10:45:30 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for volumes APIs https://review.opendev.org/c/openstack/nova/+/952348
10:45:30 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for snapshots APIs https://review.opendev.org/c/openstack/nova/+/952349
10:45:31 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for volume attachments APIs https://review.opendev.org/c/openstack/nova/+/952350
10:45:31 opendevreview Stephen Finucane proposed openstack/nova master: tests: Use valid UUIDs for cinder resources https://review.opendev.org/c/openstack/nova/+/952971
10:45:32 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for floating IP APIs https://review.opendev.org/c/openstack/nova/+/952972
10:45:32 opendevreview Stephen Finucane proposed openstack/nova master: api: Add response body schemas for security group APIs https://review.opendev.org/c/openstack/nova/+/952973
10:45:49 sean-k-mooney right i spent 4 or 5 hours at the weekend tyring to create a repoducer for this
10:47:12 sean-k-mooney i even through a bunch of AIs at it including gemini deep reasearch https://docs.google.com/document/d/10U6wnSO2vUOZjjHJQvd0Mfei2Kep-uho1LxJIPC_0VQ/edit?usp=sharing
11:04:54 gibi that is a long doc :)
11:56:37 elodilles hi o/ i know it's Friday, but may i get a core review to this trivial gate fix (python-novaclient)? o:) (docs job is broken in Zuul without this) https://review.opendev.org/c/openstack/python-novaclient/+/947690
12:04:56 elodilles got a +W, thanks! o:)
12:28:27 sean-k-mooney elodilles: the patch is fine even if we have existing tech debt there
12:28:36 sean-k-mooney we ar enot properly tagging things in novaclient in general
12:28:49 sean-k-mooney i.e. using doc and pdf-docs
12:28:59 sean-k-mooney even in nova we coudl do a better job of that
12:30:04 sean-k-mooney we have a bunch of things that are not tagged or using the wrogn tag in novas bindep https://review.opendev.org/c/openstack/nova/+/951749/9/bindep.txt
12:43:33 elodilles sean-k-mooney: true. to be honest, since we just extending the bindep here and python-novaclient is not that project that gets tons of patches nowadays regularly i was just happy that the gate is working with the fix o:) but you are right, probably it would be better to just tag libpcre with 'doc' tag to only get it installed in docs job
12:43:58 sean-k-mooney yep an i +2'd it because the chagne is fine
12:44:13 sean-k-mooney i was jsut pointing that out as a nit mroe then anything else to acknowlage the tech debt

Earlier   Later