| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2025-03-21 | |||
| 15:23:13 | MengyangZhang[m]1 | sorry about it, let me know if I can resend it now. don't want to interfere the current conversation on other topics | |
| 15:23:46 | dansmith | MengyangZhang[m]1: go ahead, we're done | |
| 15:24:22 | bauzas | I think his question was whether we will have cross-project sessions in the PTG | |
| 15:24:32 | sean-k-mooney | specificaly with cidner | |
| 15:24:53 | bauzas | for sure, Uggla and myself will ask Neutron, Cinder and Watcher communities if they want | |
| 15:24:54 | sean-k-mooney | MengyangZhang[m]1 has a number of cinder adjcent topics that affect nova | |
| 15:25:19 | bauzas | like every PTG, we should have one hour per x-p session, nothing booked yet | |
| 15:25:56 | MengyangZhang[m]1 | What we want is to have a uniform QoS interface, across nova, cinder and neutron, that supports per-project control and allow live updates to existing resources — whether its Cinder volumes or Neutron ports. | |
| 15:25:56 | MengyangZhang[m]1 | Regarding this blueprint: https://blueprints.launchpad.net/nova/+spec/enhanced-granularity-and-live-application-of-qos, would it be possible to discuss it with cinder, nova, and neutron in PTG? We've created three separate blueprints to proposal the same idea to different projects. | |
| 15:26:36 | sean-k-mooney | the main commonality is you want to live apply qos polices (for diffent resoces) to a vm | |
| 15:26:45 | sean-k-mooney | so volumes, ports ectra | |
| 15:26:50 | MengyangZhang[m]1 | and per-project control | |
| 15:27:06 | MengyangZhang[m]1 | neither cinder nor neutron support that now | |
| 15:27:13 | sean-k-mooney | nor those nova | |
| 15:27:26 | sean-k-mooney | we do not provide api to work on set of vms | |
| 15:28:02 | sean-k-mooney | out side of multi create and list apis all our api only work on a single vm | |
| 15:28:27 | sean-k-mooney | so the concep of applying a change to all resocue in a given project is a very large depatrue to how nova works today | |
| 15:28:41 | sean-k-mooney | s/concep/concept/ | |
| 15:32:02 | MengyangZhang[m]1 | understood, the per project control of qos utlimatly should be implemented on cinder and neutron side. we want to bring in nova is because cinder frontend qos relies on nova libvirt driver to pass the qos parameters to qemu and also per our earlier discussion that how cinder currently supports updating qos policy to volumes in use is a bug to nova. | |
| 15:33:27 | opendevreview | sean mooney proposed openstack/nova master: wrap wsgi_app.init_application with latch_error_on_raise https://review.opendev.org/c/openstack/nova/+/945028 | |
| 15:34:15 | sean-k-mooney | MengyangZhang[m]: its qustionable if it should be implented at all | |
| 15:34:32 | sean-k-mooney | if we have the live update it might be ok | |
| 15:35:01 | sean-k-mooney | but nova does not supprot the changing of the qos policy on port or volumes today while its attached to a vm | |
| 15:35:41 | sean-k-mooney | so there is a depency first we need live update supprot then if they want orchstration api to apply those updates to a set of resouce they need to be able to notify nova of those changes | |
| 15:37:02 | sean-k-mooney | MengyangZhang[m]1: any change in cinder or neutron that would change the xml that we have to generated should only be allowed in neutorn/cidner if there is an integration with nova to make that take effect | |
| 15:37:09 | sean-k-mooney | otherewise its an api bug on there side | |
| 15:38:52 | sean-k-mooney | so yes i agree the current behavior in cinder is a bug | |
| 15:39:03 | sean-k-mooney | or at least not supprot in nova | |
| 15:39:13 | sean-k-mooney | nova does not supprot all neutron api extintions | |
| 15:39:43 | sean-k-mooney | it may be valid to use them in oter contexts (zun or ironic) but for example nova does not supprot the extion that allow you to change the mtu of a neuton netowrk | |
| 15:40:02 | sean-k-mooney | or the multi provider extention that allows a neturon network to ahve more then one phsynet | |
| 15:40:43 | sean-k-mooney | that can be valid when using ironic but it does not mean its valid when using nova libvirt virt driver | |
| 15:41:26 | sean-k-mooney | the line between bug and driver dependent feature is soemtiem a grey areay | |
| 15:41:52 | sean-k-mooney | backend qos is likely fine to allow change at any time in cinder | |
| 15:42:36 | sean-k-mooney | frontend qos is not implemtned by cinder as you said its impletned by nova vai libvirt. that where the impednce mismatch comes form | |
| 15:44:23 | opendevreview | sean mooney proposed openstack/nova master: wrap wsgi_app.init_application with latch_error_on_raise https://review.opendev.org/c/openstack/nova/+/945028 | |
| 15:45:25 | MengyangZhang[m]1 | yes, all makes sense, that's not the case for neutron since qos there is implemented by the underlying network driver. Can we just leave this to the PTG meeting coz my team will be there and we can all reach an agreement and close this topic | |
| 15:47:02 | sean-k-mooney | ack | |
| 15:57:44 | opendevreview | ribaudr proposed openstack/nova master: FUP: Improve libvirt fixture for hostdevs https://review.opendev.org/c/openstack/nova/+/945232 | |
| 15:57:44 | opendevreview | ribaudr proposed openstack/nova master: FUP improve and add integration tests for PCI SR-IOV servers https://review.opendev.org/c/openstack/nova/+/944106 | |
| 18:14:46 | dansmith | vmware ci has been broken for quite some time it feels | |
| 18:15:17 | dansmith | the run I just checked seems to be failing with volume backed rebuild, which has been a thing for what two cycles? | |
| 20:10:17 | dansmith | melwitt: this works for me, if I push it up would you +W if I +2 even though it's my own work? https://termbin.com/cpvj | |
| 20:13:53 | sean-k-mooney | dansmith: i dont think they every got that to work | |
| 20:14:07 | dansmith | sean-k-mooney: what? | |
| 20:14:10 | dansmith | who? | |
| 20:14:10 | sean-k-mooney | dansmith: i know we have not merged all the patches they submited to fix thing | |
| 20:14:17 | sean-k-mooney | dansmith: the vmware ci | |
| 20:14:20 | dansmith | ah | |
| 20:15:10 | sean-k-mooney | i belive the rebuild thign is one of the pendign patches but ill admin i have not looked at the vmware patches in like 4 months | |
| 20:15:28 | dansmith | okay | |
| 20:19:11 | melwitt | dansmith: I wasn't saying there's anything wrong or need to be changed fwiw, just pointing it out and have bogdan ack it if he agrees | |
| 20:19:55 | melwitt | (if I understood his comment right anyway) | |
| 20:21:52 | dansmith | melwitt: okay was just trying to keep this moving, but fair enough | |
| 20:22:10 | dansmith | I'm glad you were able to distill something out of that because I was highly confused | |
| 20:23:55 | melwitt | ack | |
| 20:25:57 | sean-k-mooney | dansmith: https://review.opendev.org/c/openstack/nova/+/941095 i think is the fix for https://bugs.launchpad.net/cinder/+bug/2097771 | |
| 20:26:17 | sean-k-mooney | or https://bugs.launchpad.net/nova/+bug/2058225 | |
| 20:26:21 | sean-k-mooney | depending on you perspective | |
| 20:26:57 | sean-k-mooney | but its failing tempest with vmware so http://openstack-ci-logs.global.cloud.sap/openstack-nova-941095-4ptpq/tempest.html | |
| 20:35:06 | dansmith | hrm | |
| 20:35:37 | dansmith | so that's been broken and WIP for like a year | |
| 20:36:47 | sean-k-mooney | i think the current issue is just the key/passwrod | |
| 20:37:04 | sean-k-mooney | i know that they had some neutron networking issues for a while | |
| 20:50:29 | sean-k-mooney | dansmith: im kind of torn, on one hand we didnt spend a lot of time reviewign the vmware fixes | |
| 20:50:53 | sean-k-mooney | dansmith: on the other hand not much progress has been made since we paused the removal but it also has not got worse | |
| 20:51:03 | sean-k-mooney | so its still kind of limping along | |
| 20:52:11 | opendevreview | sean mooney proposed openstack/nova master: wrap wsgi_app.init_application with latch_error_on_raise https://review.opendev.org/c/openstack/nova/+/945028 | |
| 20:55:20 | sean-k-mooney | dansmith: i slightly tweaked the testing and fixed a comment if you can readd +2 | |
| 20:55:42 | sean-k-mooney | im goign to kick of one more run of the operator job and call it a day | |
| 20:55:52 | sean-k-mooney | we can tak a look at the resulst on monday | |
| 21:34:12 | dansmith | it's also wip, so not really a candidate anyway | |
| #openstack-nova - 2025-03-24 | |||
| 15:35:46 | dansmith | gibi: sean-k-mooney: so starting to look at placement where the capacity checks happen, for addressing the overcommit bugs... | |
| 15:36:12 | sean-k-mooney | oh fun | |
| 15:36:15 | dansmith | I'm only seeing a single test that ever asserts the capacity-exceeded scenario in all of placement's unit and functional tests | |
| 15:36:38 | dansmith | that is, like, unexpected since that's kinda the core function of placement | |
| 15:36:58 | sean-k-mooney | when you say functional did you include the gabbit tests | |
| 15:36:59 | dansmith | I know there are lots of gabbit tests I'm probably not grepping to find, but.. am I missing something? | |
| 15:37:21 | sean-k-mooney | well the gabbit test are the primarly api test/funcitonal tests in placment | |
| 15:37:44 | sean-k-mooney | i know they are closer to neutron fullstack tests then our functional tests | |
| 15:37:59 | sean-k-mooney | but i think thats where a lot of the end to end functial behaivor is verifed | |
| 15:38:24 | sean-k-mooney | https://github.com/openstack/placement/tree/master/placement/tests/functional/gabbits | |
| 15:39:01 | sean-k-mooney | there are a number of allcoation an dallcoation bug tests there for sepcifc beahivor | |
| 15:39:58 | dansmith | yeah, | |
| 15:40:13 | dansmith | but only one that explicitly tests overcapacity for allocations, AFAICT, | |
| 15:40:16 | sean-k-mooney | dansmith: i belive some of this is also enfoced at the db level althouhg i dont think they are usign db triggers or anything to actully encode it at the db level | |
| 15:40:21 | dansmith | and it's just the simplest case of asking for more than exists | |
| 15:40:32 | dansmith | sean-k-mooney: it is, and there's a single db test for one simple sccenario there | |
| 15:40:40 | sean-k-mooney | ack | |
| 15:40:50 | dansmith | I mean the db object layer in placement | |
| 15:41:00 | sean-k-mooney | i guess the assumtin is if you neever allow a simple allcotion to oversubscibe that covers the more advanced cases | |
| 15:41:03 | dansmith | there's a big complex function that checks if we're over capacity after an allocation | |
| 15:41:12 | sean-k-mooney | right | |
| 15:41:31 | sean-k-mooney | so https://github.com/openstack/placement/blob/master/placement/tests/functional/gabbits/allocations-bug-1779717.yaml#L73 is testing creating an allcotion for multiple consumers | |
| 15:41:38 | dansmith | it's not that simple though, that big function has a lot of edges that can't possibly be covered by a single test (single python test and a single gabbit test) | |
| 15:41:47 | dansmith | yep | |
| 15:42:07 | sean-k-mooney | although that is not really about capastiy | |
| 15:42:16 | sean-k-mooney | lookign at the bug itself https://bugs.launchpad.net/nova/+bug/1779717 | |
| 15:43:15 | sean-k-mooney | i guess to un pick this a littel you might want to create a repoducer test for the current evac/cold migration case | |
| 15:43:21 | sean-k-mooney | where we are doing the swap | |