Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-06-22
15:05:35 cfriesen_ when detaching a volume, does nova/libvirt wait for an ack from the guest the way we do for PCI devices? (If so, is it possible for the guest to do something to cause nova to wait forever?)
15:09:49 openstackgerrit Artom Lifshitz proposed openstack/nova master: Plumbing for tagged nic attachment https://review.openstack.org/415622
15:09:52 openstackgerrit Artom Lifshitz proposed openstack/nova master: Delete nic metadata when detaching interface https://review.openstack.org/435387
15:10:02 openstackgerrit Artom Lifshitz proposed openstack/nova master: Delete disk metadata when detaching volume https://review.openstack.org/435082
15:10:52 openstackgerrit Artom Lifshitz proposed openstack/nova master: API support for tagged device attachment https://review.openstack.org/376021
15:10:52 openstackgerrit Artom Lifshitz proposed openstack/nova master: Plumbing for tagged volume attachment https://review.openstack.org/391941
15:17:30 cfriesen_ mriedem: you've been reviewing the cinder stuff recently...is there a way to tell nova to just rudely detach a cinder volume without worrying about corrupting it (since it's going to be deleted right away, for example)
15:17:54 cfriesen_ mriedem: we're hitting a case where a heat stack delete is hanging because the guest is heavily using a cinder volume and so nova can't detach it
15:19:05 mriedem cfriesen_: not from the nova side
15:19:13 mriedem there is a force detach api in the cinder side i think
15:19:24 mriedem which is just to change the state on the volume from in-use to available so you can delete it
15:19:28 mriedem but,
15:19:52 mriedem you'll likely have issues deleting the instance then because nova will get volume not found - which may be ignored with a warning, not sure
15:20:50 cfriesen_ mriedem: so basically we need to quiesce the guest before detaching a cinder volume?
15:23:56 openstackgerrit Alex Szarka proposed openstack/nova master: Transform instance-live_migration_abort notification https://review.openstack.org/420453
15:24:29 mriedem cfriesen_: i think the only time we quiece the guest when a volume is attached is when taking a snapshot
15:25:11 cfriesen_ mriedem: I mean that more generally, like having the end user shut things down in the guest.
15:26:08 mriedem oh
15:26:15 mriedem does nova fail or just hang?
15:26:18 mriedem nova/os-brick
15:27:05 cfriesen_ mriedem: nova tries 7 times to detach the device and then gives up
15:27:54 cfriesen_ mriedem: not sure if nova returns an error or if heat ignores it
15:28:44 mriedem oh, well, it's a cast
15:28:49 mriedem so it's all async from a heat pov
15:29:00 mriedem so heat would have to timeout waiting for the volume to be detached
15:30:40 cfriesen_ mriedem: it's kind of too bad we can't do an unclean teardown if we're going to be deleting everything anyway
15:32:16 openstackgerrit Alex Szarka proposed openstack/nova master: Transform instance.resize.error notifications https://review.openstack.org/417882
15:39:25 bauzas jaypipes: just a rebase for https://review.openstack.org/#/c/469037/ given you already +2d it
15:39:44 bauzas mriedem: dansmith: possibly some others could look at it ^
15:40:11 openstackgerrit Sylvain Bauza proposed openstack/nova master: Fix ChanceScheduler returning a wrong node https://review.openstack.org/466725
15:41:08 jaypipes bauzas: done
15:41:30 bauzas cools
15:45:23 openstackgerrit Ildiko Vancsa proposed openstack/nova master: WIP/POC Implement new attach Cinder flow https://review.openstack.org/330285
15:49:07 mkucia Hi! I wonder if someone know what is the story behind `PCI device pools` in nova. Were those introduced for performance reasons? Why not plain list of devices?
15:51:22 cfriesen_ andreykurilin: are you around?
15:51:54 andreykurilin cfriesen_: hi! a bit (will need to go in 10 minutes)
15:57:15 cfriesen_ andreykurilin: about https://review.openstack.org/#/c/435141/, if I move the test into functional/v2/legacy/test_servers,py, it seems like those would run with the 2.1 API version, not 2.47
15:57:46 cfriesen_ andreykurilin: did you want me to copy it to the legacy tests so it gets run in both places?
15:58:40 andreykurilin cfriesen_: you can create a base class in the legacy module and inherit from it in your new TestCase
16:00:03 cfriesen_ andreykurilin: you're thinking a new base class with just the one test?
16:00:11 andreykurilin yup
16:00:23 cfriesen_ andreykurilin: okay, can try that
16:01:01 sfinucan gibi, Kevin_Zheng: I assume [1] needs to be rebased to work with [2] now? [1] https://review.openstack.org/#/c/459493/ [2] https://review.openstack.org/#/c/463001/
16:01:39 sfinucan mkucia: I should know that. Lemme dig out notes
16:03:14 mkucia sfinucan: That would be great, thanks!
16:06:44 openstackgerrit Rodolfo Alonso Hernandez proposed openstack/os-vif master: [WIP] Add abstract OVSDB API https://review.openstack.org/476612
16:06:57 sfinucan mkucia: Looks to be something to do with performance
16:07:02 sfinucan https://github.com/openstack/nova/blob/master/nova/pci/stats.py#L39-L44
16:07:23 sfinucan bauzas, jaypipes or sean-k-mooney might be able to offer more info, but that would be my guess
16:07:55 sfinucan What's that now, ralonsoh? ;)
16:08:06 jaypipes mkucia: pci device pools are stupid, IMHO. they need to be killed off.
16:08:14 mkucia I see. Thanks for the tip on the people.
16:09:11 mkucia jaypipes: Agree, there are a frew things in nova PCI handling that seems a bit off.
16:09:25 sfinucan mkucia: Careful - you'll set him off ;)
16:09:35 mkucia :O
16:09:50 sfinucan (jaypipes is a not a fan of the PCI or NUMA code in nova, heh)
16:11:04 jaypipes bauzas, dansmith, mriedem: speaking of PCI devices... this would be great to get in. https://review.openstack.org/#/c/469147/
16:12:16 ralonsoh sfinucan: hey
16:12:38 ralonsoh sfinucan: you mean https://review.openstack.org/476612?
16:15:06 sfinucan ralonsoh: Indeed
16:15:13 sfinucan Good to see. I'll keep an eye on that
16:15:56 ralonsoh sfinucan: perfect! This is a WIP. Then I'll need to: 1) add unit tests, 2) add native implementation and 3) probably add functional tests
16:16:02 sfinucan I was trying to something kind of similar with the linux-bridge stuff, but got annoyed by how "porcelain'y" pyroute2 is (https://review.openstack.org/#/c/451433/)
16:16:21 sfinucan ralonsoh: Sounds like a plan to me. I assume we can kill the non-native version then?
16:16:52 ralonsoh sfinucan: we'll keep both (for Windows) and configurable via conf file
16:16:58 ralonsoh sfinucan: like in Neutron
16:17:18 sfinucan Ah, Windows. Gotcha
16:17:26 ralonsoh sfinucan: I'll take a look at https://review.openstack.org/#/c/451433/
16:17:31 sfinucan That'll do nicely. Ping you when it's ready for eyes :)
16:17:39 sfinucan ralonsoh: Don't bother. It's not even close to complete
16:17:42 ralonsoh sfinucan: ok!
16:17:45 sfinucan More of a hobby project, really
16:21:29 sfinucan jaypipes: Ahem https://docs.openstack.org/contributor-guide/writing-style/openstack-components.html https://review.openstack.org/#/c/475810/1/doc/source/scheduling.rst
16:21:44 sfinucan But I can go with 'Nova' if you _reallllly_ want it
16:22:14 jaypipes sfinucan: I stand corrected.
16:23:04 sfinucan jaypipes: Thoughts on that diagram tool in general? Seems like a lower barrier to entry than Inkscape etc., even if the output is less pretty?
16:23:21 jaypipes sfinucan: yup, I like it a lot.
16:23:27 sfinucan sweet
16:23:46 sfinucan with that, /me is off to tag (rugby)
16:47:54 openstackgerrit Rodolfo Alonso Hernandez proposed openstack/os-vif master: Rehome unit tests to ``tests\unit`` folder https://review.openstack.org/476623
16:52:34 openstackgerrit Matt Riedemann proposed openstack/nova master: Use uuid for id in os-services API https://review.openstack.org/464280
16:52:34 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Handle uuids in os-hypervisors API https://review.openstack.org/471899
16:52:35 openstackgerrit Matt Riedemann proposed openstack/nova master: Support paging over compute nodes with a uuid marker https://review.openstack.org/476625
16:52:36 mriedem dansmith: page-o-rama ^
16:52:39 mriedem dtp: ^
16:55:59 openstackgerrit Chris Friesen proposed openstack/python-novaclient master: 2.47: Show flavor info in server details https://review.openstack.org/435141
17:08:34 openstackgerrit Ed Leafe proposed openstack/nova master: Extract custom resource classes from flavors https://review.openstack.org/473627
17:08:50 openstackgerrit Jay Pipes proposed openstack/nova master: move resources_from_request_spec() to utils https://review.openstack.org/476629
17:08:50 openstackgerrit Jay Pipes proposed openstack/nova master: placement: support GET /allocation_candidates https://review.openstack.org/475448
17:08:51 openstackgerrit Jay Pipes proposed openstack/nova master: placement: scheduler uses allocation candidates https://review.openstack.org/476631
17:08:51 openstackgerrit Jay Pipes proposed openstack/nova master: Add uuid to scheduler's HostState object https://review.openstack.org/476630
17:08:52 openstackgerrit Jay Pipes proposed openstack/nova master: WIP - placement: claim resources in scheduler https://review.openstack.org/476632
17:12:51 openstackgerrit Artom Lifshitz proposed openstack/nova master: Plumbing for tagged nic attachment https://review.openstack.org/415622
17:12:53 openstackgerrit Artom Lifshitz proposed openstack/nova master: Delete nic metadata when detaching interface https://review.openstack.org/435387
17:13:02 openstackgerrit Artom Lifshitz proposed openstack/nova master: Delete disk metadata when detaching volume https://review.openstack.org/435082
17:13:13 openstackgerrit Artom Lifshitz proposed openstack/nova master: Plumbing for tagged volume attachment https://review.openstack.org/391941
17:13:14 openstackgerrit Artom Lifshitz proposed openstack/nova master: API support for tagged device attachment https://review.openstack.org/376021
17:13:48 melwitt mriedem: yes, those are ready for review. however I wanted to run it by jaypipes because, I'm interested in the "right way" to solve the BFV 0 disk problem in placement but I have no idea what to do or if it's possible right now
17:14:21 dtp mriedem want me to look those over?
17:19:58 mriedem dtp: i want everyone and their dog to look it over

Earlier   Later