| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-02-11 | |||
| 20:31:39 | artom | It seems safer :P | |
| 20:31:50 | artom | As a first whack at it, anyays | |
| 20:31:51 | sean-k-mooney | i think we can for now. | |
| 20:31:58 | sean-k-mooney | we really should not have too | |
| 20:32:05 | sean-k-mooney | but its is simpler | |
| 20:32:16 | artom | Yeah, I thought the same when I first tried it in Rocky | |
| 20:32:23 | artom | And then went to code the XML update stuff on the source | |
| 20:32:32 | artom | And realized i_n_t isn't enough | |
| 20:33:00 | sean-k-mooney | the problem is the xml generation is mix with the buisness logic | |
| 20:33:13 | sean-k-mooney | they should have been kept seperate but they werent | |
| 20:33:25 | artom | I don't disagree :S | |
| 20:33:31 | artom | A lot of things should have been ;) | |
| 20:33:52 | sean-k-mooney | so for now i think keeping the changes minimal are fro the best adn maybe we should just decuple it fully latter | |
| 20:34:33 | sean-k-mooney | and by later i mean as a sperate activiy in train | |
| 20:35:36 | artom | We'd need a quorum of virt drivers that can support NUMA LM | |
| 20:35:48 | sean-k-mooney | i also hate that that file is like 9000 lines long | |
| 20:35:57 | sean-k-mooney | not really | |
| 20:35:57 | artom | To better understand what we need to encode in the objects/business logic | |
| 20:36:08 | artom | No? I mean, I have no idea what HyperV wants | |
| 20:36:26 | artom | I think it would help me if I had claudiub explain it to me :) | |
| 20:36:34 | artom | (If he's still around) | |
| 20:36:38 | sean-k-mooney | hyperv support numa but does not to cpu pinning the same way | |
| 20:36:59 | openstackgerrit | Eric Fried proposed openstack/nova master: Use tox 3.1.1 fixes https://review.openstack.org/582392 | |
| 20:37:02 | sean-k-mooney | its done in the hyperviors rather then in the nova dirive i think | |
| 20:37:34 | sean-k-mooney | the issue with the libvirt dirver is that neither libvirt or qemu are hyperviors | |
| 20:38:39 | artom | Ah, kvm is the hypervisor | |
| 20:38:46 | sean-k-mooney | qemu is a emulator and libvirt is just an abstraction layer/api. but are totaly dumb and just do what you tell them so nova had to build most of the logic that would normally live in a hypvervisor | |
| 20:38:54 | sean-k-mooney | artom: kvm is not a hypervior either | |
| 20:39:07 | artom | sean-k-mooney, heh, OK, define hypervisor then :) | |
| 20:39:07 | sean-k-mooney | its an aclleration driver/kernel module | |
| 20:40:06 | artom | ... allowing the kernel to act as the hypervisor (according to wiki, at any rate_ | |
| 20:40:32 | artom | No point in getting bogged down in semantics, I un-WIPed a couple of the patches, working on a third... | |
| 20:41:04 | sean-k-mooney | a hypervior is a set of services that mediates acess to hardware such that multiple kernel can share the hardware | |
| 20:41:15 | sean-k-mooney | :) true | |
| 20:42:00 | edmondsw | artom you're asking about getting rid of to_legacy_dict just for powervm or in general? I don't see it called for any hypervisor | |
| 20:42:29 | artom | edmondsw, yeah, I thought it as just powervm, but it looked to be pretty generalized | |
| 20:42:39 | artom | Ie, all drivers had _legacy_dict stuff in migrate_data.py | |
| 20:42:56 | edmondsw | yep... and all could go away at this point, right? | |
| 20:42:56 | artom | And I'm proposing to remove all of them: https://review.openstack.org/#/c/636210/ | |
| 20:43:00 | edmondsw | sounds good | |
| 20:43:23 | sean-k-mooney | artom: well as the name suggest its likely a legacy fuction that was kept for handeling mixed compute node versions | |
| 20:44:44 | artom | Changing stuff in nova is like a plastic surgeon operating on Joan Rivers's face: you have no idea who else has touched stuff before, so really all you hope for is to do the minimal amount of damage while keeping Joan alive. | |
| 20:44:46 | artom | (RIP Joan) | |
| 20:45:15 | artom | sean-k-mooney, so to answer your point, I'm not touch anything, named "legacy" or not, unless I'm damn sure it can be removed. | |
| 20:47:10 | sean-k-mooney | artom: well at least for the base class it was added by dansmith in https://github.com/openstack/nova/commit/038dfd672f5b2be5ebe30d85bd00d09bae2993fc | |
| 20:47:38 | artom | Heh, yeah, saw that. And it occurs nowhere else that I could see. | |
| 20:47:44 | artom | So I figured we're OK to get rid of them all. | |
| 20:51:12 | sean-k-mooney | artom: so based on the conversation with dan earliar are you going to revert the change to https://review.openstack.org/#/c/634827/10/nova/objects/migrate_data.py and ath the if x in y... check | |
| 20:51:30 | sean-k-mooney | e.g. revert the defults | |
| 20:51:39 | artom | sean-k-mooney, yeah | |
| 20:51:46 | artom | And add a proper check | |
| 20:51:56 | openstackgerrit | Merged openstack/nova master: Fup for the bandwidth resource provider series https://review.openstack.org/633776 | |
| 20:53:58 | sean-k-mooney | cool am im going to go to sleep earlish tonight i think. ill redeploy the series in my test vms tomorow. | |
| 20:54:42 | sean-k-mooney | i think it more or less should work at this point as the last version i tested seamed pretty close | |
| 20:55:02 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: New objects to transmit NUMA config from dest to source https://review.openstack.org/634827 | |
| 20:55:03 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] LM: Make dest send NUMAMigrateData to the source https://review.openstack.org/634828 | |
| 20:55:03 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] LM: update NUMA-related XML on the source https://review.openstack.org/635229 | |
| 20:55:14 | artom | sean-k-mooney, noice. There ^^ | |
| 20:55:57 | artom | I did the change, will continue adding unit tests where I can without too much trouble, and then I'll start on the automated testing | |
| 20:57:32 | sean-k-mooney | oh cool i am going work on some os-vif stuff for another 30 mins or so. i shold have time to spin up a devstack install and manually live migrate a test vm in the background | |
| 21:22:42 | openstackgerrit | Matt Riedemann proposed openstack/nova master: WIP: Filter duplicates from compute API get_migrations_sorted() https://review.openstack.org/636224 | |
| 21:53:21 | openstackgerrit | Merged openstack/nova stable/queens: Handle IndexError in _populate_neutron_binding_profile https://review.openstack.org/635897 | |
| 21:53:27 | openstackgerrit | Merged openstack/nova stable/queens: Add functional regression test for bug 1794996 https://review.openstack.org/623354 | |
| 21:53:28 | openstack | bug 1794996 in OpenStack Compute (nova) queens "_destroy_evacuated_instances fails and kills n-cpu startup if lazy-loading flavor on a deleted instance" [High,In progress] https://launchpad.net/bugs/1794996 - Assigned to Matt Riedemann (mriedem) | |
| 22:13:32 | mriedem | oh blarg, our NeutronFixture is not a KSA client fixture... | |
| 23:03:06 | mriedem | efried: hmm, i'm hoping this isn't a regression http://paste.openstack.org/show/744898/ | |
| 23:03:28 | mriedem | this is on rollback of a failed resize where conductor reverts the allocation move by trying to swap the allocations back from the migration consumer to the instance consumer | |
| 23:03:40 | mriedem | the GET /allocations/$consumer_id in the code is using placement 1.28 | |
| 23:03:45 | mriedem | which should have project_id in the response | |
| 23:04:26 | mriedem | i wonder if maybe allocations is {} ? | |
| 23:05:10 | efried | That seems possible. | |
| 23:05:55 | efried | reproducible? Stuff in some debug logs? | |
| 23:06:56 | mriedem | OH I KNOW | |
| 23:07:15 | mriedem | so this is the conductor MigrationTask trying to swap the allocs, but the compute service in this case already reverted them | |
| 23:07:24 | mriedem | so conductor is trying to move something that doesn't exist b/c it's already been moved | |
| 23:07:36 | mriedem | not a regression, just not robust code | |
| 23:08:58 | efried | seems like we've seen this before | |
| 23:09:09 | efried | thought we had a fix up that would ignore empty source allocations | |
| 23:10:56 | mriedem | this is a pretty specific part of nova so i'm guessing it's just not been an issue before, | |
| 23:11:21 | mriedem | especially because with normal resize, conductor rpc casts to compute and if something fails in compute, it reverts the allocations and conductor doesn't get a chance (b/c of the cast) | |
| 23:11:24 | openstackgerrit | Eric Fried proposed openstack/nova master: Use placement.inventory.inuse in report client https://review.openstack.org/568639 | |
| 23:11:27 | mriedem | with cross-cell resize it's synchronous | |
| 23:11:35 | mriedem | so compute reverts and then conductor tries to revert as well | |
| 23:17:10 | efried | mriedem: here's the change I was thinking of. Slightly different but similar race https://review.openstack.org/#/c/617040/ | |
| 23:18:35 | mriedem | yeah this isn't a race that i'm dealing with now, just new code paths that aren't robustified yet | |
| 23:18:51 | mriedem | laying my old cross-cell resize functional tests on top of my current series and flushing out some bugs | |
| #openstack-nova - 2019-02-12 | |||
| 01:09:10 | openstackgerrit | Matt Riedemann proposed openstack/nova master: WIP: Start functional testing for cross-cell resize https://review.openstack.org/636253 | |
| 01:09:31 | mriedem | woooowee | |
| 01:09:35 | mriedem | dansmith: ^ well that's progress | |
| 01:09:57 | mriedem | retrofitted my previous functional tests to the existing series, flushed a bunch of bugs, and have it tested up through VERIFY_RESIZE | |
| 02:01:21 | openstackgerrit | Merged openstack/nova master: Plumbing for allowing the all-tenants filter with down cells https://review.openstack.org/635145 | |
| 03:48:33 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: Remove _legacy_dict methods https://review.openstack.org/636210 | |
| 03:48:34 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: Add migration param to check_can_live_migrate_destination https://review.openstack.org/634605 | |
| 03:48:34 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: New objects to transmit NUMA config from dest to source https://review.openstack.org/634827 | |
| 03:48:35 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] Introduce live_migration_claim() https://review.openstack.org/635669 | |
| 03:48:35 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] Use live_migration_claim() to check dest resources https://review.openstack.org/634606 | |
| 03:48:36 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] LM: Make dest send NUMAMigrateData to the source https://review.openstack.org/634828 | |
| 03:48:36 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: [WIP] LM: update NUMA-related XML on the source https://review.openstack.org/635229 | |
| 03:48:37 | openstackgerrit | Artom Lifshitz proposed openstack/nova master: Trivial: reorder hashes according to object_hashes.txt https://review.openstack.org/636261 | |
| 03:49:10 | artom | Actually I should probably put the trivial one at the very bottom... | |