Earlier  
Posted Nick Remark
#openstack-nova - 2020-03-18
13:21:15 sean-k-mooney i think its just "pip install osc-placement"
13:21:53 ygk_12345 sean-k-mooney should I do it in the utility node in openstack ansible ?
13:21:55 sean-k-mooney well "python3 -m pip install --user osc-placement" woudl be more modern way of doing it
13:21:58 lyarwood bauzas: random scheduler question for you, if we fail to schedule an instance and it moves to ERROR can we ever retry launching that instance or can we only delete it?
13:22:20 lyarwood bauzas: working through an issue where failure to schedule an instance with attached volumes leaves these volumes in a reserved state
13:22:31 sean-k-mooney ygk_12345: this really isnt a supprot channel and definetly not for openstack ansible so i dont know.
13:22:46 lyarwood bauzas: but if we are able to schedule and/or go straight to a specific host and then fail the volumes become available
13:23:17 lyarwood I assume we want things to stay reserved *if* we can relaunch the instances somehow
13:23:38 ygk_12345 sean-k-mooney oh ok
13:35:06 luyao stephenfin: Hi, I saw you rechecked my vpmem-live-migration patch. FYI, I submitted a new version, there are a bug in the previous one causing the tempest test failed under mixed-version compute services, I think I addressed it for now, you can look at it if you have time.
13:37:12 bauzas lyarwood: I'm back
13:37:35 kashyap lyarwood: I can't see how your patch is the reason for the 'tempest-integrated-compute' failure
13:37:42 kashyap (Not that you're claiming it :))
13:39:19 lyarwood kashyap: which one sorry?
13:39:56 luyao artom: I'm afraid I can't response to you promptly in IRC , so I'll appreciate that if you can leave comments under that patch.
13:39:58 kashyap lyarwood: The blockCopy one; sorry - https://zuul.opendev.org/t/openstack/build/1796646647854d128ba128b45d5c7a14
13:40:12 artom luyao, yeah, that's fine
13:42:14 bauzas lyarwood: to answer your question, you could rebuild I think but not just 'create' again
13:42:42 artom dansmith, when you're about, you don't have to look at the entire patch, but just the https://review.opendev.org/#/c/687856/15/nova/conductor/tasks/live_migrate.py@57 bit, about using compute service versions to judge support for a thing
13:43:54 lyarwood bauzas: so we can't just `nova start $instance`
13:43:59 lyarwood bauzas: we have to rebuild?
13:44:23 lyarwood bauzas: either way, I'd assume that given that's an option the volumes should stay reserved right?
13:49:21 bauzas lyarwood: 'start' recreates the guest XML but doesn't ask for cinder and neutron
13:49:27 bauzas (AFAIR)
13:50:12 lyarwood bauzas: it should update the attachment but it wouldn't recreate it
13:51:15 bauzas yup, this
13:51:55 bauzas either way ,if the instance was set to ERROR on scheduling stage, then the related cell is cell°
13:51:55 bauzas either way ,if the instance was set to ERROR on scheduling stage, then the related cell is cell°
13:51:59 bauzas cell0 sorry
13:52:09 bauzas so you're messed up
13:52:21 bauzas that's why a rebuild or resize is the only option
13:56:41 sean-k-mooney the nova start api action calls power on the virt driver which in the libvirt driver calls hard reboot because reasons
13:57:24 sean-k-mooney so yes it recreates the xml. it might rettrive attachemnt info form cinder but it will not modify the sate of the volume in cinder
13:57:54 lyarwood sean-k-mooney / bauzas ; ack thanks
13:58:08 lyarwood so I'm thinking that either way the volume needs to remain reserved
13:58:19 lyarwood if we ERROR out during scheduling (current behaviour)
13:58:36 lyarwood or if we ERROR out in the virt driver during spawn (currently moves back to available)
13:58:37 sean-k-mooney well if you stop an instance then start it we should not hit the scuduler
13:58:43 sean-k-mooney we dont for a hard reboot
13:58:59 lyarwood right
13:59:16 sean-k-mooney a rebuild need to hit the schduler to assert the new image does not contian metadata that invaldiate the current host
13:59:26 sean-k-mooney but start and reboot dont cause nothing changes
13:59:38 lyarwood right
13:59:45 sean-k-mooney lyarwood: whats the issue you are looking at by the way
14:01:00 lyarwood sean-k-mooney: as above, when we ERROR out during scheduling attached volumes remain in a reserved state (as we don't delete their attachments in cinder), if we fail during spawn the compute layer moves the volume back to available (by deleting the attachments in cinder).
14:01:20 nightmare_unreal hello where can I see list of bugs for openstack CLI client or nova related bugs
14:01:43 sean-k-mooney lyarwood: ah ok
14:01:48 lyarwood sean-k-mooney: so I'm thinking that for the latter we should keep the volume reserved to avoid something or someone else coming along and using it
14:01:58 lyarwood if it isn't multiattach etc etc
14:02:51 sean-k-mooney lyarwood: the later being during spawn when we delete the attachemtn we leave it in a reserved sate
14:03:20 lyarwood sean-k-mooney: when we delete the attachments the volume moves back to an available state
14:03:37 lyarwood sean-k-mooney: I'd like to change this to keep it reserved by not deleting the attachments
14:04:23 sean-k-mooney in the case wehre start fails or boot fails or both
14:04:42 sean-k-mooney for boot it will end up in cell 0 and have host so no attachemts
14:04:57 sean-k-mooney *no host
14:05:17 lyarwood for spawn, so we have a host and a cell already
14:06:29 lyarwood brb
14:24:31 sean-k-mooney nightmare_unreal: nova bugs are here https://bugs.launchpad.net/nova
14:24:45 sean-k-mooney nightmare_unreal: the openstack client uses storyborad
14:25:52 sean-k-mooney so here https://storyboard.openstack.org/#!/project/openstack/python-openstackclient but this is all the releated projects https://storyboard.openstack.org/#!/project_group/80
14:26:15 nightmare_unreal thanks :)
14:34:45 dansmith artom: your sigh about migration func testing is lack of review I assume?
14:34:55 dansmith I'm sure I'm at least partially to blame there
14:35:51 artom dansmith, no, mostly towards myself
14:35:57 dansmith oh okay
14:35:58 artom For not staying on top of it
14:36:33 artom I wanted to do a push, well, starting now
14:36:51 artom I still have stephenfin's feedback from the previous millenium to address (about the 2/3 NUMA nodes split)
14:37:16 artom But if we manage to land it before the VPMEM live migration patch, it would make things easier
14:37:25 dansmith ack
14:37:48 dansmith I'm pretty un-jazzed about that patch
14:38:24 openstackgerrit Balazs Gibizer proposed openstack/nova master: Support unshelve with qos ports https://review.opendev.org/704759
14:38:24 openstackgerrit Balazs Gibizer proposed openstack/nova master: Enable unshelve with qos ports https://review.opendev.org/705475
14:38:37 artom dansmith, live migration with VPMEM?
14:38:44 dansmith yes
14:39:10 artom In terms of current implementation or the concept in general?
14:39:34 artom I'm just neutral-jazzed - I got involved because my pawprints are all over that code after NUMA live migration
14:39:56 dansmith well, I'm rather negative on the vpmem stuff in general, but that ship has sailed..But yeah, the current patch more acutely
14:40:09 openstackgerrit Balazs Gibizer proposed openstack/nova master: Merge qos related renos for Ussuri https://review.opendev.org/706766
14:40:17 dansmith it smacks of "let me just move around all these critical things so they work the way I need them"
14:40:44 artom Yeah...
14:40:49 lyarwood sean-k-mooney: https://bugs.launchpad.net/neutron/+bug/1867936 seen this before?
14:40:49 openstack Launchpad bug 1867936 in neutron "test_update_delete_extra_route failing due to timeout when creating subnets" [Undecided,New]
14:41:28 artom dansmith, you've essentially put words on a vague feeling I had
14:41:43 sean-k-mooney lyarwood: only in the failure of one of your patches
14:41:48 artom Which is why I would really be more comfortable if we at least had NUMA LM func tests in-tree before merging
14:42:07 gibi dansmith: hi! added the global service version check to https://review.opendev.org/#/c/705475
14:42:08 artom Those at least exercise those code paths
14:42:18 dansmith artom: well, I certainly don't think that's unreasonable, as long as we can get that moving
14:42:33 artom dansmith, ack, lemme get on that
14:42:56 lyarwood sean-k-mooney: yeah I've seen it a few times, was just wondering if someones already raised a more specific bug report for this
14:43:04 dansmith artom: using your patch to validate what we have in tree so we can more directly assess the changes this one makes would be helpful
14:43:07 dansmith gibi: ack
14:43:39 artom dansmith, the bottom two of https://review.opendev.org/#/c/705545/ would be ready, I'd say
14:43:41 sean-k-mooney lyarwood: there have been some issue with the dhcp agent by the way. e.g. it not seting up the dhcp config in a timely manner
14:43:52 dansmith gibi: I feel like we might be at the point with the cyborg set where someone else needs to take over it.. I'm not fully caught up yet this morning but I think feedback has been pending for a while now
14:43:58 sean-k-mooney this might just be a side effect of that althogh i though that neuton had fixed that
14:44:07 artom dansmith, if you have any -1s to get in on those, do it now ;)
14:44:28 lyarwood sean-k-mooney: kk
14:44:29 dansmith artom: at least the bottom one is -W, should it not be?

Earlier   Later