| opendevreview | melanie witt proposed openstack/devstack master: WIP Fix IPv6 support for multinode devstack deployments https://review.opendev.org/c/openstack/devstack/+/984605 | 00:19 |
|---|---|---|
| *** vhari_ is now known as vhari | 03:12 | |
| opendevreview | Amit Uniyal proposed openstack/devstack master: Check kernel support before setting lvm_type=auto https://review.opendev.org/c/openstack/devstack/+/986049 | 08:05 |
| opendevreview | Amit Uniyal proposed openstack/devstack master: Check kernel support before setting lvm_type=auto https://review.opendev.org/c/openstack/devstack/+/986049 | 08:08 |
| opendevreview | Amit Uniyal proposed openstack/devstack master: Check kernel support before setting lvm_type=auto https://review.opendev.org/c/openstack/devstack/+/986049 | 08:21 |
| sean-k-mooney | clarkb: fungi where is the best place to talk about new images and what feature our providers supprot? | 11:18 |
| sean-k-mooney | here #openstack-infra or on matix | 11:18 |
| sean-k-mooney | context is in the nova and cybrog ssessions we were wondering about the possiblity of testing with vms that have a specific set of nova feature enabel to allow emulating sriov capbale nics | 11:19 |
| sean-k-mooney | hw_vif_model=igb + hw_machine_type=q35 + hw_viommu_model=auto + hw_mem_page_size=small possibale als hw_memory_locked=true | 11:21 |
| sean-k-mooney | i could have typoed those since im doint that form memory | 11:21 |
| sean-k-mooney | igb is an intel nic form the mid 2000s i belive that supprot up to 7 vf, the machine type and iommu option allow the igb nic to be in its own iommu group | 11:22 |
| sean-k-mooney | so that the vfs acna be passthough through and i think but i migh tbe wrong the hw_mem_pagesize might be required to make the viommu | 11:23 |
| sean-k-mooney | work | 11:23 |
| sean-k-mooney | dan mention there was a patch to make the viommu aviable in the past so we cna locally test and confirm the details but i just was wondering how we wwould go about chekcing if any of our provider are new enough to be able to suprpot testing in this way | 11:24 |
| fungi | sean-k-mooney: you just need the flags list from /proc/cpuinfo in each provider? | 12:11 |
| sean-k-mooney | no | 12:43 |
| sean-k-mooney | this has no depency on the hardware | 12:43 |
| sean-k-mooney | this has a depency on the nova/qemu versions used on the provider | 12:43 |
| sean-k-mooney | fungi: so https://review.opendev.org/c/openstack/project-config/+/945715 added a viommu-ubuntu-noble lable in the past | 12:45 |
| sean-k-mooney | but we would need to set a few more flags on that to fully test sirov. what there now may work in a limite set of cases | 12:47 |
| sean-k-mooney | im just wondering acan we tweak it to use the igb nic type so we can test with virutal function not just full device passthough | 12:47 |
| sean-k-mooney | we are also looking into if the rdo third pary ci could do this or if we can cobble together another ci system outside the firwall to do soem testing with real hardware as a redhat third party ci | 12:48 |
| sean-k-mooney | but ovbiouly if we can test this isn the first party ci that even better | 12:48 |
| frickler | essentially I think we could do anything where you can provide a patch for. without seeing what changes in our config are needed, this is very difficult to discuss | 12:49 |
| fungi | okay, so the example was adding image metadata on our glance uploads, and you want to add more of those, but need to know which metadata signals would actually work in which providers? | 12:50 |
| *** ykarel_ is now known as ykarel | 14:35 | |
| sean-k-mooney | fungi: sorry busy day. ya so it woudl be addign extra metadata on an image upload (ideally a second copy of the image to avoid affecting other jobs) on provider tha tsupprot that feature | 19:25 |
| sean-k-mooney | but obviously we need ot 1 figure out the exact set fo properites, i can do that with a devstack and make it concreat | 19:25 |
| sean-k-mooney | i can also lookup which release intoduced the requried featues | 19:26 |
| sean-k-mooney | its been a coulple of years since they were added but there is obvioulsy a lot of variance | 19:26 |
| clarkb | is there any way for us to knwo if a cloud supports that functionality? | 20:45 |
| clarkb | like we can set flags on images but how do we know if it is safe to do that or if we'll get the expected result back out? | 20:45 |
| sean-k-mooney | am so in practice nvoa will just ignore image metadta that it does nto supprot | 20:48 |
| sean-k-mooney | we stop supproting custom image metadta beforfe we sarted supproting python 3 | 20:48 |
| sean-k-mooney | in practice you coudl boot it an see | 20:48 |
| sean-k-mooney | but the closet proxly woudl be to look at ther reporte api microversion and infer | 20:49 |
| clarkb | got it so it doesn't make the image unbootable if the metadata is unknown. I think that is my main concern | 20:49 |
| clarkb | so worst case you still boot just without the functionality in the instance actually working | 20:49 |
| sean-k-mooney | the reaon i suggested a seperate image is the viommu does impact performace | 20:49 |
| sean-k-mooney | so i woudl not want to add it by defualt | 20:49 |
| sean-k-mooney | although i need to double check if that is true only when its used | 20:50 |
| sean-k-mooney | clarkb: the igb nic case is more probelmatic as that is extened an image proeprty | 20:51 |
| sean-k-mooney | and in that case it will prevent it booting if the cloud does not supprot it | 20:52 |
| sean-k-mooney | if we knwo about the iamge proeprlty and you pass a value we dont know about/supprot then that is a hard error | 20:52 |
| sean-k-mooney | vexxhost-ca-ymq-1 apprently supprot viommu based on the old patch but i am also not entirly sure if that patch really worked | 20:52 |
| sean-k-mooney | https://opendev.org/opendev/zuul-providers | 20:54 |
| sean-k-mooney | does not have that lable/iamge | 20:54 |
| sean-k-mooney | so it was obviously removed at aome point | 20:54 |
| sean-k-mooney | we did follow up ot add q35 which was also needed https://review.opendev.org/c/openstack/project-config/+/948335/1/nodepool/nodepool.yaml | 20:55 |
| sean-k-mooney | its late here but the simplest way might just be to ask mnaser if this feature can be used in vexhost | 20:56 |
| clarkb | sean-k-mooney: the config all moved to the opendev/zuul-providers repo | 20:58 |
| clarkb | none of the nodepool config has been used since like july | 20:58 |
| clarkb | which may be where things got lost in the shuffle | 20:58 |
| sean-k-mooney | yep i was just looking at the new repo | 20:59 |
| sean-k-mooney | provided its still possibel to set flag like that with teh new mode we can likely upload the same image twice with a diffent name | 21:00 |
| sean-k-mooney | and then create a lable for it | 21:00 |
| sean-k-mooney | i guess ill look into this next week. i dont see any example in the proted repo of adding metadta to images | 21:02 |
| sean-k-mooney | so we need to confirm if zuul can still do this in the new mode or if we lost htat in the converion | 21:03 |
| sean-k-mooney | clarkb: it looking like this might be a regression in the new way of doing image building for what it sworth | 21:08 |
| sean-k-mooney | so we may need to talk to the zuul folks to reimplemnt that if we want to use it agin | 21:08 |
| clarkb | I think it was a pretty direct port of the nodepool code; maybe it is just lacking documetnation | 22:01 |
Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!