Earlier  
Posted Nick Remark
#openstack-nova - 2019-01-16
15:01:18 mriedem if it's optional, then it could just be a very targeted thing that gets populated in the db api if that field is requested in columns_to_join
15:01:30 mriedem *expected_attrs
15:01:51 mriedem dansmith might have an opinion on that
15:02:45 dansmith yep, like bdms, makes sense
15:03:55 efried mriedem: Do I have my time wrong, or are we supposed to be doing the delete-placement-from-nova hangout right now?
15:04:20 gibi efried: isn't it 17:00 UTC?
15:04:38 efried d'oh, that's my bad. Thanks gibi
15:04:58 efried gibi: and the bw one is also 1700 UTC, on friday?
15:05:03 gibi efried: yes
15:05:27 mriedem ShilpaSD: i left some comments in the patch
15:06:51 ShilpaSD mriedem: will check, thank you for explaination and comments
15:07:30 openstackgerrit Sundar Nadathur proposed openstack/nova master: WIP: Add Cyborg device profile groups to spec obj. https://review.openstack.org/631243
15:07:30 openstackgerrit Sundar Nadathur proposed openstack/nova master: WIP: Add utility function to get Cyborg client. https://review.openstack.org/631242
15:07:31 openstackgerrit Sundar Nadathur proposed openstack/nova master: WIP: Get resolved Cyborg ARQs and add PCI BDFs to VM's domain XML. https://review.openstack.org/631245
15:07:31 openstackgerrit Sundar Nadathur proposed openstack/nova master: WIP: Create and bind Cyborg ARQs. https://review.openstack.org/631244
15:14:36 mriedem melwitt: will your max volume attach series change the device name of existing attached bdms if the user reboots their server? i believe device tags are linked to bdms via device name, so was wondering if we could possibly have an upgrade impact with the device tag <> bdm.device_name getting out of whack
15:15:02 mriedem artom: ^ you might know off the top of your head on the device tag stuff
15:15:46 artom mriedem, it's been a while, lemme check the code
15:18:12 artom mriedem, yeah, it's by name
15:24:58 mriedem artom: do you remember why we used bdm.device_name as the correlation key for that stuff? maybe bdm uuid didn't exist when you wrote that
15:25:46 mriedem or the volume id for that matter
15:26:02 artom mriedem, there was attempt, I think
15:26:07 artom Lemme trawl through the patches
15:26:32 dpawlik mriedem: hi, if you have free time pls check https://review.openstack.org/#/c/628128/ https://review.openstack.org/#/c/628956/
15:26:36 dpawlik thanks
15:27:26 openstackgerrit Matt Riedemann proposed openstack/nova stable/queens: Removing pip-missing-reqs from default tox jobs https://review.openstack.org/628956
15:30:55 jaypipes sean-k-mooney: did you see Numan's comments on https://review.openstack.org/#/c/602384/ causing intermittent failures in tempest runs?
15:32:04 sean-k-mooney reading it now.
15:33:05 sean-k-mooney ok so the nova patch i abandoned to chagne the libvirt type to ethernet woudl adress there issue
15:33:21 sean-k-mooney i can un abandon it and see if that helps.
15:33:57 sean-k-mooney idelaly i dont want libvirt involed in configuring ovs
15:35:03 sean-k-mooney jaypipes: i also wonder if ovn support multiple port bidning.
15:37:39 jaypipes sean-k-mooney: since Numan is another RHatter, can you get in touch with him about this?
15:38:14 mriedem dpawlik: done
15:38:18 sean-k-mooney yep ill bring it up internally
15:39:18 openstackgerrit sean mooney proposed openstack/nova master: libvirt: delegate ovs plug to os-vif https://review.openstack.org/602432
15:39:22 mriedem kashyap: so on https://review.openstack.org/#/c/629627/ i guess you forgot my comment about referring to the new doc from the kvm live migration doc and/or the 'security' doc?
15:39:31 kashyap mriedem: Hi
15:39:52 kashyap mriedem: I did comment in the abandoned follow-up change I guess
15:39:54 kashyap Let me see
15:40:07 jaypipes sean-k-mooney: danke
15:40:12 mriedem kashyap: you said you'd follow up,
15:40:12 kashyap mriedem: (And thanks for your time!)
15:40:14 mriedem and then abandoned the follow up
15:40:25 mriedem so now i guess you need to re-follow-up
15:40:52 artom mriedem, ah, so when we detach a tagged BDM, we remove its metadata using its serial/volume id
15:40:53 kashyap mriedem: Heh, so the "strategy" was:
15:41:07 artom mriedem, but when we build the metadata, we go by device name
15:41:27 kashyap mriedem: (1) Squash the first follow-up into the main doc as we needed a re-spin; (2) Do the re-follow-up with all the hyperlinking goodness.
15:41:32 mriedem artom: and if i'm looking at the libvirt driver code correctly, we only build the metadata on spawn, attach and resize
15:41:53 mriedem so my fear about screwing up the linkage during reboot with different device names is probably unfounded
15:42:05 artom mriedem, detach as well
15:42:09 artom Wait
15:42:24 artom No, you're right, only attach
15:42:43 artom But I think we rebuild the whole metadata when we attach any device
15:42:43 mriedem i saw attach volume and then the two places where we build a config drive
15:42:45 mriedem spawn and resize
15:43:11 mriedem yeah looks like it
15:43:17 mriedem re rebuild the whole metadata
15:45:09 artom mriedem, we're talking about https://review.openstack.org/#/c/616777/8/nova/compute/utils.py, right?
15:45:39 artom I'm still parsing it, but... does it only apply the new naming scheme to newly attached/booted-with bdms?
15:45:56 mriedem artom: yeah
15:46:09 mriedem yeah in that is the patch i'm talking about
15:46:24 mriedem i don't know about the latter, that's what i'm wondering
15:46:32 artom So then I don't think that's a problem? Because while the scheme changes, the device name will still be consistent between the libvirt XML and the bdm object in the db
15:46:37 artom Which is what we care about
15:47:07 mriedem for libvirt this is really the code that matters, not compute.utils https://review.openstack.org/#/c/616777/8/nova/virt/libvirt/blockinfo.py
15:47:14 mriedem since libvirt is special and has it's own device naming thing
15:48:47 artom mriedem, as long as that bit happen before the XML is generated, it shouldn't be a problem
15:49:31 mriedem once the bdm.device_name is originally set, i also don't think it will change. _prep_block_device in the compute manager calls _add_missing_dev_names but that noops if a bdm already has a device_name set
15:50:06 artom mriedem, well, if we want to really be sure, could we add some tagged devices to rally tests?
15:50:42 artom Or grenade? What's the thing that does upgrades in the gate?
15:51:36 dansmith heh
15:51:39 dansmith grenade
15:51:42 dansmith oopsie
15:51:59 dansmith guess I never dropped my cape
15:53:29 mriedem artom: grenade does attach a volume on the old side before the upgrade, but it doesn't tag it - and even if it did, to verify the guest can read the tag properly on the new side (via tempest) would require running a test that does that ssh into the guest, which wouldn't be a "smoke" test and those are the only kinds of tests that grenade jobs run
15:53:31 mriedem plus,
15:53:46 mriedem the chances of the device name changing when you only have one volume attached is pretty slim
15:53:55 mriedem since i think it'd just be vda or vdb in that case
15:54:30 mriedem anyway, if the root answer is "the device naming schema is unchanged, but fixed to allow more than 26 names" then i tihnk we're ok
15:54:36 mriedem i.e. consistent naming schema, but unbounded
15:54:56 mriedem i think that's what the commit message in https://review.openstack.org/#/c/573066/ is saying
15:55:01 dansmith mriedem: just ensure the first 26 are the same pattern after the change, right?
15:55:12 dansmith only the 27th should be different and then you're probably fine yeah?
15:55:30 artom mriedem, I thought the question was "does the new schema retroactively get applied to existing devices?"
15:56:10 mriedem i think the real question is, why am i worrying about this and wasting my time on it
15:56:40 artom And actually not even that - "does the new schema retroactively get applied to *only* the database or *only* the XML?"
15:57:16 artom mriedem, when experiencing bouts of existential nihilism I usually turn to alcohol
15:57:30 mriedem if anything it would be only the xml
15:57:37 mriedem pretty sure we don't change bdm device_name once set
15:57:48 mriedem b/c that hoses up some db constraints and stuff we passed to cinder during the attach
15:58:07 artom Err, then we actually might have a problem :(
15:58:45 mriedem like dan said, if the old pattern and new pattern are the same for the first 26 then we're fine
15:59:33 artom Yeah, screw those people with more than 26 disks, what were they thinking anyways
15:59:54 sean-k-mooney artom: you could not have more then 26 disks before
16:00:09 artom sean-k-mooney, ohhhh, hah!
16:00:12 artom Yeah, we're fine then
16:00:56 artom And maybe mriedem's right, fix that bug if/when it comes up
16:01:09 artom Seems like a pretty edgy case

Earlier   Later