Monday, 2025-08-25

*** mhen_ is now known as mhen01:18
gibirelease liasions please check the pending nova stable releases we have an important security hardening bug in them 09:42
gibihttps://ironicbaremetal.org/blog/coming-soon-threading/09:42
gibihttps://review.opendev.org/c/openstack/releases/+/95795109:42
gibihttps://review.opendev.org/c/openstack/releases/+/95800909:42
gibihttps://review.opendev.org/c/openstack/releases/+/94784709:42
gibiahh I see Uggla was there already. 09:45
Uggla@gibi, yes I think I have reviewed them, unless I miss one.09:46
opendevreviewKamil Sambor proposed openstack/nova master: [hacking] Improve N373 to catch also other primitives  https://review.opendev.org/c/openstack/nova/+/95840810:02
elodillesgibi: the release patches merged, things should be available on pypi and tarball.o.o, tags, etc., in some minutes10:09
elodillesUggla: ^^^10:09
gibielodilles: thanks!10:13
elodillesnp10:15
*** sean-k-mooney1 is now known as sean-k-mooney12:54
opendevreviewKamil Sambor proposed openstack/nova master: Switch nova-conductor to use ThreadPoolExecutor  https://review.opendev.org/c/openstack/nova/+/95708813:02
hayashinoCould someone review this patch? https://review.opendev.org/c/openstack/nova/+/94216713:13
opendevreviewKamil Sambor proposed openstack/nova master: Switch nova-conductor to use ThreadPoolExecutor  https://review.opendev.org/c/openstack/nova/+/95708813:52
*** DHE_ is now known as DHE14:44
bryanfraschettiHello Nova team o/16:01
bryanfraschettiHope you're all doing well. I'm just wondering the current state of this bug: https://bugs.launchpad.net/nova/+bug/2039803, where live migration fails across similar hosts due to the compareHypervisorCPU check, and the associated topic: https://review.opendev.org/q/topic:%22bp/cpu-selection-with-hypervisor-consideration%22. I noticed that there was a recent comment 16:01
bryanfraschettihttps://bugs.launchpad.net/nova/+bug/2039803/comments/9, which mentioned that replacing the baselineCPU method is one of the remaining missing puzzle pieces but I don't have a lot of context on what's been done and what's needed.16:01
bryanfraschettiIt might be worth mentioning that that we're seeing this on Ubuntu Jammy / Caracal, which contains the other two commits mentioned in that same comment.16:01
sean-k-mooneybryanfraschetti: ya so it depend on the specific cpu version you have16:32
sean-k-mooneythe root cause si effectivly this16:32
sean-k-mooneyfor some cpus the diffent ebtween the capablities api and domcapabiltys api out can cause the migration check in nova to fail16:32
sean-k-mooneythere is a workaroudn config option to disable the nova check in pre-migration and rely on the much later libvirt check16:33
sean-k-mooneythat is much more costly to revert when it fails but ti will work in more cases16:33
sean-k-mooneythe other alternitive is to use an explcit custom model instead of host-model or leaving the cpu_mode empty16:34
sean-k-mooneycpu_mode unset and host-model are almsot but not quite identical16:34
sean-k-mooneythey are ment to select teh virtual cpu model that supprot hte largest subset of the host cpu features16:35
sean-k-mooneythat should be the closest cpu model to the hsot i.e. the host-model16:35
sean-k-mooneyfor soem intel cpus in particalr but also newer amd cpus the odl and new api may differ.16:36
sean-k-mooneyusing the new api in all places coudl fix it16:36
sean-k-mooneythe only concern then woudl be makeign sure the havior fo host-model and unset remains the same16:36
sean-k-mooneybryanfraschetti: given Feature Freeze for this cycle is on thrusday this is not currently a priorty for the core team however if someone has time to work on fixign it we wont boject16:37
sean-k-mooneybut we likely wont have time to review for a week or two16:37
*** melwitt_ is now known as jgwentworth16:43
*** jgwentworth is now known as melwitt16:44
bryanfraschettiThanks @sean-k-mooney for the explanation that clarifies things a lot actually. So the issue isn't so much the new approach of comparing hypervisors but rather that there's mixed usage of new and old approaches to check the capabilities, which can sometimes disagree - especially on new intel cpus. Consistently applying the new approach should resolve the discrepancies, which makes sense.16:56
bryanfraschettiQuick Q about the workaround options which skip the pre-migration check. What happens if these are enabled (eg. skip_cpu_check_on_start and skip_check_on_dest), but the hypervisors truly are incompatible and migration fails? Will this be caught by the libvirt check, and will the migration roll back - effectively not migrating, or will the migration be left in a broken state?16:56
gmaansean-k-mooney: on service role, I was checking if neutron has documented that nova service user need 'service' role or not but I could not find. Cinder has documented it so that is good.16:58
gmaansean-k-mooney: and I am now thinking many operator might have not have 'service' role and might have configured  only admin role in neutron.conf 16:59
gmaanand nova policy change to 'role:service' will break them. I am now re-thinking on keep allowing admin also for those service API at least for this cycle - https://review.opendev.org/c/openstack/nova/+/957578/comment/a097d558_fee01375/17:00
gmaanand this cycle I can update neutron doc and release notes about requiring 'service' role to nova user17:00
gmaanwhen I objected on not allowing admin, I thought of only direct user usage of service APIs which should be blocked right away but did not know that neutron did not state about 'service' role for their service users17:02
gmaanand new default enable by default makes nova change to break such neutron deployment which has only admin role on nova user17:02
gmaanI know you are ok with admin-or-service role but just want to confirm that we are on same page as I spent lot of time on service token things in tempest, neutron, oslo.policy check side which I should not have if discussed with you in advance :(17:04
sean-k-mooneygmaan: so that why i orginlly propsoed doing `service or admin` last cycle17:04
sean-k-mooneyfor upgrades17:04
sean-k-mooneyin our installer we alwasy create the users for service with both admin and service roles17:05
gmaanyeah, I know. I thought admin role given to service user was just in devstack because we implies those in our policy but agree now that operator can have same17:05
sean-k-mooneyno that how kolla etcra all do it17:06
gmaanlet me update my change and once that is merged I can update the spec document also to sync with implementation17:06
sean-k-mooneyim not sure if kolla give both admin and serivice or just admin today17:06
sean-k-mooneyi think it became more comment when we started requiring it for nova and cinder to talk to each other17:07
sean-k-mooneybut i ahve not checked in a long time17:07
gmaanyeah, I am not sure if they have service  role but maybe17:08
sean-k-mooneygmaan: the have supprote for at least ironci https://github.com/openstack/kolla-ansible/commit/600e912400ab8b52ca422f007f63da9d4fcc0a69 18:00
sean-k-mooneyhttps://github.com/search?q=repo%3Aopenstack%2Fkolla-ansible++role%3A+%5C%22service%5C%22&type=code18:01
sean-k-mooneyit looks like kolla at least does provide teh service role to several services but not to all18:02
sean-k-mooneyironic nova cidner 18:02
gmaanyeah, and for neutron too so nova will be ok18:02
JayFsean-k-mooney: I am not certain Ironic documents to use a service scope/role user18:03
JayFsean-k-mooney: even if we permit and support it18:03
sean-k-mooneyoh i missed neutron yes18:03
JayFsean-k-mooney: so we should ensure, if we're changing the shape of this, not only to ask what's possible, but ask what is likely and what path we've been asking folks to take18:03
gmaanyeah, that is what i am mainly checking because most of the place it is done like 'neutron need admin to talk to nova' which is ok but still add service role too18:04
sean-k-mooneyJayF: right so orginally i propeose haveing the policy rule be admin_or_service for 1 slurp and then service_only18:04
sean-k-mooneybut we are tryign to determin if that is actuly required or not18:05
JayFthere's no doubt in my mind the answer is "yes"18:05
sean-k-mooneyto be safe we proably shoudl but we dropted that form this release spec18:05
sean-k-mooneybecause we tought it was not required18:05
gmaanwe could avoid it if we implemeted it before we switch to enable the global new defaults 18:05
sean-k-mooneywell we already did that no?18:06
sean-k-mooneyi tougl we did that a release or two ago18:06
sean-k-mooneyas part of phase 118:06
gmaanyes I mean if new defaults were disabled by default then admin was added via old default by oslo.policy18:06
sean-k-mooneyoh18:06
sean-k-mooneyok well ya that bridge has sailed... /me bridges dont sail boats do.... 18:07
* sean-k-mooney or ships18:07
*** haleyb is now known as haleyb|out20:58

Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!