Wednesday, 2022-04-27

mnasersean-k-mooney, bauzas: ricolin on our side will help drive this and i'll also be helping in shaping it from our side.  we don't have a whole *lot* of background knowledge about it but we can do a bunch of testing about the vIOMMU stuff03:13
opendevreviewMerged openstack/nova stable/yoga: Fix eventlet.tpool import  https://review.opendev.org/c/openstack/nova/+/83678803:38
opendevreviewMerged openstack/nova stable/ussuri: Fix the vGPU dynamic options race  https://review.opendev.org/c/openstack/nova/+/83152405:03
gibigood morning07:14
bauzasgibi: morning 07:16
bauzasmnaser: ack, very much appreciated07:16
bauzaselodilles: I made my homework I promised07:26
bauzasmelwitt: sent to the gate07:26
elodillesbauzas: cool, thanks \o/08:12
opendevreviewAndrew Bonney proposed openstack/nova master: Fix segment-aware scheduling permissions error  https://review.opendev.org/c/openstack/nova/+/83936109:25
opendevreviewRajat Dhasmana proposed openstack/nova master: Add support for volume backed server rebuild  https://review.opendev.org/c/openstack/nova/+/82036810:00
opendevreviewRajat Dhasmana proposed openstack/nova master: Add conductor RPC interface for rebuild  https://review.opendev.org/c/openstack/nova/+/83121910:00
opendevreviewRajat Dhasmana proposed openstack/nova master: Add API support for rebuilding BFV instances  https://review.opendev.org/c/openstack/nova/+/83088310:00
ricolinbauzas: for vIOMMU, I will work on BP asap for further discussion. Also I'm not that familar with nova future develop, are these also require to raise a meeting topic, or BP is enough for first step10:21
sean-k-mooneyricolin: you should ideally draft a spec10:33
sean-k-mooneyrather then just a blueprint10:33
ricolinsean-k-mooney: agree10:39
sean-k-mooneyare you familar enough with that process https://github.com/openstack/nova-specs/blob/master/specs/zed-template.rst10:43
sean-k-mooneythe poc patch is a good start10:44
sean-k-mooneyand i left most of my feed back in that patch already from a design point of view10:44
sean-k-mooneyso for the spec really waht i am looking for is to defien the scope of the chagne, the interfac (extra_specs and image proerties) and the usecasses we want to adress.10:45
sean-k-mooneythere is also a schduling question e.g. can we rely based on min libvirt version on all host suport this or do we need a compute capablity trait and prefileter for schduling10:45
sean-k-mooneyi havent checked thet min version to determin that10:46
sean-k-mooneythe vIOMMU has been aroudn a while so i think our min version support it but the adress space option is newer so not sure about that part10:47
gibibauzas: you probably interested in https://review.opendev.org/c/openstack/nova/+/839361 as it fixes a bug in the routed net pre-filter11:27
sean-k-mooneygibi: i was looking at that11:39
sean-k-mooneyim not sure why we would need admin rights11:39
sean-k-mooneyi would have expect the segment id to be viabel to a normal user11:39
gibibecause neutron does not return the segment_id for non admins in the GET /subnets response11:40
gibiI tried in a devstack, it only returns it for an admin11:40
sean-k-mooneyhum that soudn like a design issue to me on the neutron side11:40
sean-k-mooneyi wonder why that desision was made11:40
gibiof course other solution is to change neutron to return that for non-admins11:40
gibiI haven't tracked down the reason of this on the neutron side11:41
gibiI had many similar bugs in the qos work11:41
gibithere the resource_request field being admin-only caused similar problems11:41
sean-k-mooneyright but i dont think using admin is the right move in general11:41
gibiin general no11:41
sean-k-mooneyyep i was thinking of that and there is no reason that shoudl be admin-only11:41
sean-k-mooneyit shoudl not be writable form the api but it shoudl be readable by the owner11:42
sean-k-mooneyi think the segment id shoudl be the same readable by project_reader11:43
sean-k-mooneybut not writabel by project_member11:43
gibiI'm OK to pursue this with the neutron team but I think we can merge the above fix as changing API behavior like that might no be backportable 11:45
sean-k-mooneyits currently project_admin by policy https://github.com/openstack/neutron/blob/master/neutron/conf/policies/subnet.py#L54-L67= its a correct fix i guess11:50
gibihm, that is subnet creation11:51
gibibut yes the GET is project admin too segment_id11:51
gibihttps://github.com/openstack/neutron/blob/6db61d070483df60ffbf992d57cef2c1903fe355/neutron/conf/policies/subnet.py#L98-L10811:51
sean-k-mooneyya its there in both11:52
opendevreviewWenping Song proposed openstack/os-traits master: Add nova and cyborg owner traits for resources  https://review.opendev.org/c/openstack/os-traits/+/83912811:52
sean-k-mooneywhat troubling is actully https://github.com/openstack/neutron-lib/blob/master/neutron_lib/api/definitions/subnet_segmentid_writable.py11:52
sean-k-mooneyapparently you can update the segment id of a subnet11:53
sean-k-mooneyi dont know if the vm/schdilign implciations fo that were considered11:54
gibiI don't want to know what happens with the IP pool during such update11:54
gibiand yes, you are right that can go against the currently scheduled host of the VMs11:54
sean-k-mooneyor if it went form not a routed netowk to a routed one11:54
gibiyeah that too11:54
* sean-k-mooney lesson is to stop looking at what extensions neutorn optionall support11:55
sean-k-mooneyi hope they only allow this before ports are added to the subnet11:55
* gibi closes the browser tab and goes to wash his hands11:55
sean-k-mooneygibi: added +2 but now +w to see if bauzas wants to review. we can kick it into the gate if bauzas  is busy and does not review by tomorrow11:58
gibisure. thanks11:59
gibibtw the author is a first time nova contributor :)11:59
sean-k-mooneyi guess https://bugzilla.redhat.com/show_bug.cgi?id=1761903 woudl fail qe without thsi so once that lands we should backport it for them i guess12:01
gibiyeah this fix needs to be backport12:06
gibiup until wallaby upstream12:08
opendevreviewWenping Song proposed openstack/os-traits master: Add nova and cyborg owner traits for resources  https://review.opendev.org/c/openstack/os-traits/+/83912812:20
*** amoralej is now known as amoralej|lunch12:30
gibisean-k-mooney: the OWNER os-traits patch now looks good to me https://review.opendev.org/c/openstack/os-traits/+/839128 let's land it and propose a os-traits release12:57
sean-k-mooneyyep i see the added a release not. i guess that does not hurt but im not sure how many people actully consume that12:58
sean-k-mooneyill take a look again shortly12:58
gibiyeah reno does not hurt12:59
sean-k-mooneyi need to finsih reviewing your pci patches again too12:59
gibithanks for that!12:59
sean-k-mooneythe code looks fine on the first patch but i have not found a block of uninterupted time in the last few days to trace through the unit tests and line them up12:59
sean-k-mooneyim sure they are fine but that all i have left to do on the first patch13:00
gibino worries13:07
*** amoralej|lunch is now known as amoralej13:12
opendevreviewMerged openstack/os-traits master: Add nova and cyborg owner traits for resources  https://review.opendev.org/c/openstack/os-traits/+/83912813:26
bauzasgibi: sean-k-mooney: eventually accepted https://review.opendev.org/c/openstack/nova/+/83936113:28
bauzasgibi: sean-k-mooney: that said, I'm sad we need to be an admin for knowing the API extensiion13:29
sean-k-mooneywell we can know if the extension is aviabel without being an admin but we cant consume the data form the extention without being an admin13:31
sean-k-mooneyi guess the assumetion here is that routed networks are provider networks and requies an admin to create13:32
sean-k-mooneybut you then dont need to be an admin to create ports on that network13:32
sean-k-mooneyso i really think it shoudl be readonly13:32
sean-k-mooneyfor project_memeber13:32
sean-k-mooneyand project_admin to create the subnet and affinitese it to a segment13:32
sean-k-mooneyso we likely shoudl add neutron to the bug and ask them to change the default policy for get but not create to project_reader13:33
*** dasm|off is now known as dasm13:48
bauzassean-k-mooney: yeah, we should ask the neutron team to change the policy14:19
sean-k-mooneyshould we just tag them in the bug and ask them to update and or propsoe a patch to do that?14:27
sean-k-mooneywe dont consider policy change to be api changes14:27
sean-k-mooneyso presumable that would be backportable14:27
gibiI'm not sure neutron will think this is backportable, but let see14:47
melwittbauzas: thanks ^_^14:59
opendevreviewDan Smith proposed openstack/nova master: DNM: Run against performance.json patch  https://review.opendev.org/c/openstack/nova/+/83893415:45
*** amoralej is now known as amoralej|off15:57
gibibauzas: I reported a gate-failure https://bugs.launchpad.net/nova/+bug/1970642 it is not super frequent but I saw it multiple times17:00
opendevreviewArtom Lifshitz proposed openstack/nova master: Reproduce live migration rollback w/o multi port bindings error  https://review.opendev.org/c/openstack/nova/+/83909118:53
opendevreviewArtom Lifshitz proposed openstack/nova master: Fix LM rollback w/o multi port bindings extension  https://review.opendev.org/c/openstack/nova/+/83922718:53
*** artom__ is now known as artom18:53
*** dasm is now known as dasm|off22:19
opendevreviewGhanshyam proposed openstack/nova master: DNM: test https://review.opendev.org/c/openstack/tempest/+/839647  https://review.opendev.org/c/openstack/nova/+/83964822:53
opendevreviewGhanshyam proposed openstack/nova stable/yoga: DNM: test https://review.opendev.org/c/openstack/tempest/+/839647  https://review.opendev.org/c/openstack/nova/+/83965322:56
opendevreviewGhanshyam proposed openstack/nova stable/xena: DNM: test https://review.opendev.org/c/openstack/tempest/+/839647  https://review.opendev.org/c/openstack/nova/+/83965422:56
opendevreviewGhanshyam proposed openstack/nova master: DNM: test https://review.opendev.org/c/openstack/tempest/+/839647  https://review.opendev.org/c/openstack/nova/+/83964823:02
opendevreviewGhanshyam proposed openstack/nova stable/yoga: DNM: test https://review.opendev.org/c/openstack/tempest/+/839647  https://review.opendev.org/c/openstack/nova/+/83965323:02
opendevreviewGhanshyam proposed openstack/nova stable/xena: DNM: test https://review.opendev.org/c/openstack/tempest/+/839647  https://review.opendev.org/c/openstack/nova/+/83965423:03

Generated by irclog2html.py 2.17.3 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!