Monday, 2026-07-27

jgilabersean-k-mooney, melwitt can you take a look at the nova patches related to mdevs https://review.opendev.org/q/topic:%22cyborg-vgpu-support%22 when you have some time? thanks12:44
jgilaberI'd like to ensure they are in a good place before leaving on PTO12:45
jgilaberI tested them end to end in https://review.opendev.org/c/openstack/cyborg/+/997364 12:46
sean-k-mooneyjgilaber: sure ill try and take a look at them this week13:09
melwittjgilaber: sure, will do13:58
jgilaberthanks!14:04
opendevreviewGihong Lee proposed openstack/cyborg master: Migrate accelerator drivers to shared lspci helpers  https://review.opendev.org/c/openstack/cyborg/+/99881714:50
melwittsean-k-mooney: fyi this is the KNOWN_NICS patch https://review.opendev.org/c/openstack/cyborg/+/997782 that came out of the smartnic spike https://gist.github.com/melwitt/bb7979d9b1e06cba4a0702e90b4211ab16:49
sean-k-mooneyyes i saw it brifely last week16:49
melwittah ok cool16:50
sean-k-mooneylong term i think we shoudl look to deprecat that driver and the nic driver type in general16:50
sean-k-mooneybut for now that is cool16:50
sean-k-mooneywere you able to test this end to end?16:50
melwittyeah I was thinking that too and briefly discussed during the last cyborg meeting16:51
melwittyes the end to end test worked16:51
melwitt(it's in that github gist)16:51
sean-k-mooneywiht neutron or just with reqing it via a device profile16:51
sean-k-mooneyso there are 2 wasy to consume this right now, you can create a neutron port of type acclerator-direct16:52
melwittwith neutron16:52
sean-k-mooneyand the sriov nic agent shoudl be able to manage it and you shoudl be able to use the vf via a neutron network16:52
sean-k-mooneyok cool16:52
sean-k-mooneyit shoudl also work via the flavor16:52
melwittthere is a minimalistic patch I could do to duplicate the logic from the pci driver (I think?) to use a whitelist conf instead of the KNOWN_NICS list in the shorter term since removing the nic driver would be a larger change16:53
sean-k-mooneymelwitt: so looking at https://review.opendev.org/c/openstack/cyborg/+/997782 i think its ok16:53
sean-k-mooneyit may be nice to ether add docs to this patch or to have a seperate followup16:53
sean-k-mooneymelwitt: well so if you jsut used teh whitelist tha woudl work with the pci driver today without code changes16:54
sean-k-mooneythe biter question is will that still work with neutron16:54
sean-k-mooneythe intel driver in theory allwos use to encode info about the pshenet as traits on the RP16:55
sean-k-mooneyi think you can do that with the generic driver manually 16:55
sean-k-mooneythe larger question i had was does neutron or nova care16:55
sean-k-mooneyi.e. is anythign today enfocing the VF and phsnet match end to end16:55
melwittsean-k-mooney: you mean like "how to" kind of docs?16:56
melwittI can add that, just wasn't sure if I should or if the support matrix is all that was wanted16:57
sean-k-mooneyyes a how to use the intel driver to provide cyborg manged neutron ports16:57
sean-k-mooneyand if it worked with the generic drive we coudl add that later16:57
sean-k-mooneyi assume the generic pci driver will fail in nova16:58
sean-k-mooneybecuase in teh nic driver we encode the physnet info in the attachmetn handel https://github.com/openstack/cyborg/blob/master/cyborg/accelerator/drivers/nic/intel/sysinfo.py#L236-L24316:58
melwitthm ok, I can give that a try16:58
sean-k-mooneyso assume that is checked somewhere16:59
sean-k-mooneybut maybe it is not16:59
melwittgotcha16:59
sean-k-mooneymelwitt: basiclly what i want to confirm is if we have 2 nic ports on phsynet A and B17:00
sean-k-mooneydo we actully passh the phsyent info end to end and enforce it17:00
melwittyeah, last meeting we were chatting about the idea of "moving" the nic driver stuff into the pci driver and just have a little bit of "if device type" to handle them both17:00
sean-k-mooneyor was that missed when this was orgianlly devleoped17:00
sean-k-mooneymelwitt: ya that could work17:00
sean-k-mooneyor like in nova we coudl add teh physical_network tag to the whitelist config option17:01
melwittok I can look into that with the physnet17:01
sean-k-mooneycool for now i think we can proceed with your small patch17:01
sean-k-mooneyadn then we can followup with docs and more testing17:01
sean-k-mooneymelwitt: does that work for you?17:02
melwittcool sounds good17:02
sean-k-mooneymelwitt: jgilaber incase this passed ye by this is an interisting patch https://review.opendev.org/c/openstack/cyborg/+/99872617:09
sean-k-mooneyi was going to look at this as part fo large rework next cycle17:09
sean-k-mooneybut even without that the cleanup form conslitdating woudl be nice to have17:09
sean-k-mooneymy incliation is to allow that to prpoceed with a fairly narrow scope this cycle and build on it next cycle to intoduce a PCIBusManager with real objects instead of just strings17:10
melwittoh yeah, that's cool17:10
melwittmakes sense17:11
sean-k-mooneyamoralej: so while the spec was proposed in yoga the keyston bootstrap command only created the manager role form bobcat17:27
sean-k-mooneyhttps://github.com/openstack/keystone/commit/bbbbd10e482a67b53187a93daa58f8bee770852f17:27
sean-k-mooneyamoralej: so ya htat why it works upstream and fails with keystone form antelope17:27
opendevreviewMerged openstack/cyborg master: Add Intel XXV710 25GbE backplane NIC to known NIC list  https://review.opendev.org/c/openstack/cyborg/+/99778218:10
jgilabersean-k-mooney, thanks I saw the patch and added it to my review list, but have not really looked much18:19
sean-k-mooneyjgilaber: no worries i am more higlighing it in case it impact your other work and jsut incase you found it interesting18:42
sean-k-mooneyits proably more impactfult ot chandan actully 18:42
jgilaberyes I don't think it'll conflict with my patches18:43

Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!