Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-03-06
16:13:53 mriedem stephenfin: comments inline
16:14:40 mriedem stephenfin: i'm less concerned about the excessive warning logs, but we need to mention the impact to novncproxy_base_url when using novnc >= 1.0.0
16:14:54 openstackgerrit Merged openstack/nova stable/queens: unquiesce instance on volume snapshot failure https://review.openstack.org/545961
16:15:01 openstackgerrit Merged openstack/nova stable/queens: Add the ability to get absolute limits from Cinder https://review.openstack.org/546157
16:15:42 stephenfin mriedem: Fair. I'll address those now
16:47:31 mriedem jgriffith: really small cleanups for https://review.openstack.org/#/c/548427/ - otherwise looks ready
16:47:36 mriedem jgriffith: i can hit those if you're busy
16:49:02 openstackgerrit sahid proposed openstack/nova-specs master: libvirt: add support for virtio-net rx/tx queue sizes https://review.openstack.org/539605
16:50:15 dansmith mriedem: did we agree on a specless blueprint for the db purge stuff?
16:51:03 mriedem dansmith: yeah i think so
16:51:03 sahid mriedem: if you can add this in your queue, https://review.openstack.org/#/c/485522/ - we have the code ready so i think it would be nice to make progress
16:51:19 dansmith cool
16:51:21 mriedem sahid: yeah i need to go over that
16:51:32 sahid cool thanks
16:54:45 openstackgerrit Eric Berglund proposed openstack/nova master: PowerVM Driver: DiskAdapter parent class https://review.openstack.org/549053
16:55:19 openstackgerrit Eric Berglund proposed openstack/nova master: WIP: PowerVM Driver: Localdisk https://review.openstack.org/549300
17:01:34 openstackgerrit Merged openstack/nova stable/queens: Add resource_class to fields in ironic node cache https://review.openstack.org/546085
17:04:34 openstackgerrit Matt Riedemann proposed openstack/nova master: Check for multiattach before removing connections https://review.openstack.org/548427
17:04:53 mriedem jgriffith: i updated it ^ +2 now, thanks a ton
17:05:12 mriedem cores ^ and the patch below it are needed to fix a pretty severe issue with multiattach to connections on the same host
17:05:21 mriedem gonna need to bakcport that to queens and cut a patch release
17:07:18 gibi Kevin_Zheng: If you have time and interest then you can help me with https://blueprints.launchpad.net/nova/+spec/add-request-id-to-instance-action-notifications
17:08:04 gibi Kevin_Zheng: we can reassing the bp to you so you can propose the implementation and then I will go and review it
17:09:11 mnaser hey, small riddle: you create a vm with nova, it creates ports in a neutron network, you detach those ports using the nova interface detach api .. what happens to those ports?!
17:10:46 cfriesen mnaser: I'd assume they hang around in neutron
17:10:49 bauzas mriedem: looking but a bit complex
17:11:01 mnaser cfriesen: except nova deletes them
17:11:41 mnaser so i did a little research and the preserve_on_delete for neutron ports gets set only if the ports are manually assigned on boot (or later attached)
17:11:54 mnaser if nova creates the port, preserve_on_delete=False so detaching the port via the api destroys it in nova
17:11:54 cfriesen mnaser: I'm guessing someone thought it made sense for nova to delete them since nova created them. not sure I agree in this case.
17:12:21 mnaser well, interface *detach* should be an explicit detach. especially if i'm manually detaching it from a running instance
17:12:35 mnaser i'm trying to understand where i can start fixing this (probably by writing a bug)
17:13:14 openstackgerrit Dan Smith proposed openstack/nova master: Add simple db purge command https://review.openstack.org/550171
17:13:23 bauzas mriedem: mmmm, that's a big stuff https://review.openstack.org/#/c/548427/9/nova/virt/libvirt/driver.py
17:13:50 bauzas mriedem: so, when disconnecting the volume, we should ask Cinder and look at all the instances ? huh
17:14:14 bauzas I understand why we do this
17:14:16 mriedem bauzas: we have to do connection counting for that volume and instances on the same host
17:14:21 mriedem but only for multiattach volumes
17:14:28 bauzas yeah I understand that
17:14:29 mriedem this patch is also written to be backportable
17:14:37 bauzas but wow, it's a big call, right?
17:14:42 mriedem what is?
17:15:12 openstackgerrit Stephen Finucane proposed openstack/nova master: conf: Deprecate 'keymap' options https://review.openstack.org/483994
17:15:14 bauzas every time we would disconnect the volume, we would have to call Cinder *and* verify all the instances
17:15:19 stephenfin mriedem: Done, good sir ^
17:15:20 bauzas for the host
17:15:26 mriedem bauzas: for multiattach volumes yes
17:15:55 mriedem bauzas: the attached host for the volumes is in the cinder API,
17:16:15 mriedem but as discussed in the nova/cinder session last thursday, cinder is going to add an admin-only policy rule to not return that attached host information
17:16:17 mriedem so we can't rely on it here
17:16:33 cfriesen mnaser: write up a bug and poke mriedem to look at it since he's the neutron contact in the wiki.
17:16:44 mriedem if we did a kind of admin client thing for cinder, like we do for neutron port binding, we could get that information from cinder w/o going to the nova db to get the instances on this host
17:16:55 mriedem cfriesen: mnaser: it's not a bug
17:17:12 mriedem there was a spec from someone at some point about adding a delete_on_termination like flag for ports like we have for bdms
17:17:18 mriedem i can't find it anymore
17:17:20 bauzas mriedem: yeah I was thinking of that
17:17:22 bauzas but anyway
17:17:37 bauzas if you want to backport it, then it could be a problem if we ask Cinder to modify their API
17:17:48 bauzas so I think I'm +1, just need to verify the tests
17:17:54 mnaser the use-case is: user A booted an instance A, user A wants to move the port from instance A and instance B, they can't do it because a detach destroys the port and they need admin api to create ports with specific ips
17:18:57 cfriesen mnaser: currently I think you'd have to create the port first and hand it to nova when booting the instance. Then the detach should leave it around.
17:19:14 mriedem bauzas: if you want to read through the back and forth comments earlier in the review, there are several options here, but not backportable ones
17:19:17 mnaser cfriesen: yeah, but if you created an instance 2 years ago and never thought about this, it gets a bit hectic but yeah
17:19:28 mriedem bauzas: so we have this to backport to queens, it will work out of the box, no operator config needed to fix the bug,
17:19:28 bauzas mriedem: yeah, okay, anyway
17:19:37 bauzas yeah, I'm just writing that comment
17:19:46 mriedem we could optimize in the future if we can rely on cinder admin creds in nova.conf and can escalate our volume api calls like we do for neutron
17:20:12 mriedem that InstanceList.get_uuids_on_host() can also be optimized
17:20:46 bauzas ++
17:21:41 bauzas mriedem: +Wd
17:22:30 bauzas we need another core for the dependent change https://review.openstack.org/#/c/549411/1
17:23:10 openstackgerrit Stephen Finucane proposed openstack/nova master: Update docs, API samples for noVNC 1.0.0 https://review.openstack.org/550173
17:23:37 openstackgerrit Dan Smith proposed openstack/nova master: Add simple db purge command https://review.openstack.org/550171
17:28:51 mriedem bauzas: thanks
17:31:04 mriedem dansmith: heh a bit of prior art https://blueprints.launchpad.net/nova/?searchtext=purge
17:31:29 dansmith heh
17:32:47 mriedem dansmith: is what you're doing different from what we had merged (in a spec) back in mitaka? https://review.openstack.org/#/c/200224/9/specs/mitaka/approved/purge-deleted-instances-cmd.rst
17:33:45 dansmith mriedem: it's similar although simpler
17:38:10 openstackgerrit git-harry proposed openstack/nova master: Fix version cap when no nova-compute started https://review.openstack.org/549737
17:39:57 mriedem ok i marked 4 other blueprints as superseded against yours
17:40:35 dansmith I think that's pretty good evidence that we should do this thing :)
17:40:39 mriedem dansmith: we should throw it on the nova meeting agenda for thursday, per the process for specless blueprints; i don't expect issues, but thems the rules
17:40:50 mriedem yeah i know
17:40:59 dansmith aye
17:41:24 jaypipes melwitt, mriedem, dansmith: attempting to summarize scheduler/placement decisions and priorities here: https://etherpad.openstack.org/p/rocky-ptg-scheduler-placement-summary. Your feedback welcome.
17:41:29 mriedem dansmith: are there going to be more than 1 patch? in dublin i thought you were talking about a series
17:42:08 dansmith mriedem: that one patch is the meat.. I have at least one other little glue patch to make archive purge when it's done, which I'll have up in a few
17:46:01 openstackgerrit Dan Smith proposed openstack/nova master: Add --purge helper flag to archive_deleted_rows https://review.openstack.org/550182
17:46:07 dansmith mriedem: that ^
17:46:19 mriedem ok that's what i was thinking of
17:46:25 mriedem otherwise you have the 2 step dance
17:46:28 dansmith yeah
17:54:48 dansmith mriedem: this week's meeting is early so I won't be there, but I put it on the agenda
17:59:20 mriedem dansmith: yeah i can proxy
17:59:22 mriedem assuming i'm there
17:59:28 dansmith sweet
17:59:58 mriedem jaypipes: ok - mel is out this week, and i'm not sure about the relative priority of all of those things (or doability)
18:00:11 mriedem efried reviewing that would be good too but he's also on vacation this week
18:00:13 mriedem lazy bums
18:00:46 mriedem the only thing on the 'merge update_provider_tree' stuff is the whole explosion on friday about merging traits and who owns what and all that

Earlier   Later