| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-08-21 | |||
| 15:55:20 | openstackgerrit | Stephen Finucane proposed openstack/nova master: console: introduce framework for RFB authentication https://review.openstack.org/345397 | |
| 15:55:20 | openstackgerrit | Stephen Finucane proposed openstack/nova master: console: introduce the VeNCrypt RFB authentication scheme https://review.openstack.org/345398 | |
| 15:55:21 | openstackgerrit | Stephen Finucane proposed openstack/nova master: console: provide an RFB security proxy implementation https://review.openstack.org/345399 | |
| 15:57:08 | mriedem | disabling RamFilter by default hits it's first victim https://bugs.launchpad.net/nova/+bug/1712057 | |
| 15:57:09 | openstack | Launchpad bug 1712057 in OpenStack Compute (nova) "When the specified destination host deploys the virtual machine, the allocation ratio is not valid" [Undecided,Invalid] - Assigned to wangyicheng (wang-yicheng) | |
| 15:58:23 | mriedem | dansmith: did anyone mention this https://review.openstack.org/#/c/495811/ ? | |
| 15:58:26 | cfriesen_ | mriedem: whee! do we want to fill in the limits filter somewhere else? | |
| 15:58:38 | cfriesen_ | make that just limits, not filter | |
| 15:58:51 | mriedem | cfriesen_: the intent is to remove all of that stuff eventually, so i don't think so | |
| 15:58:59 | mriedem | the limits are used in the claims code in the compute | |
| 15:59:16 | mriedem | i suspect the reporter of that bug has out of tree code that's relying on the limits | |
| 16:03:16 | cdent | mriedem: what do you mean by “mention this”? gibi posted about it this morning, the author of the new live migration tests is his colleague. gibi’s going on vacation, so “left it to us” | |
| 16:03:43 | mriedem | mention as in, this could be an rc2 blocker | |
| 16:03:47 | mriedem | since it wasn't tagged | |
| 16:03:51 | mriedem | or in the etherpad | |
| 16:05:00 | cdent | I think it’s new since the weekend | |
| 16:09:48 | openstackgerrit | Stephen Finucane proposed openstack/nova master: trivial: Remove dead function, variable https://review.openstack.org/491512 | |
| 16:15:06 | dansmith | mriedem: mention what? | |
| 16:15:10 | dansmith | mriedem: oh no | |
| 16:17:07 | openstackgerrit | Stephen Finucane proposed openstack/nova master: conf: Convert 'live_migration_inbound_addr' to HostAddressOpt https://review.openstack.org/456573 | |
| 16:23:18 | lvdombrkr | hello guys, im trieng deploy tripleO , and after trying deploy overcloud get error: http://paste.openstack.org/raw/618917/ , in nova scheduler logs i see that tha node is filtred out http://paste.openstack.org/raw/618930/ | |
| 16:23:29 | lvdombrkr | but i dont uderstand why | |
| 16:23:36 | lvdombrkr | any ideas? | |
| 16:24:55 | stephenfin | lvdombrkr: Looks like you'd be better asking #openstack or #tripleo | |
| 16:28:37 | mriedem | mnestratov|2: fyi https://bugs.launchpad.net/nova/+bug/1711547 | |
| 16:28:38 | openstack | Launchpad bug 1711547 in OpenStack Compute (nova) "Nova service restart disconnects vzstorage volumes" [Undecided,New] | |
| 16:29:16 | mnestratov|2 | thanks | |
| 16:32:40 | mnestratov|2 | we will take care of it | |
| 16:32:47 | openstackgerrit | Stephen Finucane proposed openstack/nova master: doc: Address review comments for main index https://review.openstack.org/492645 | |
| 16:33:49 | mriedem | here is that fun ip filtering known issue rearing it's head https://bugs.launchpad.net/nova/+bug/1711303 | |
| 16:33:49 | openstack | Launchpad bug 1711303 in OpenStack Compute (nova) "Filter instances by IP when listing is not efficient enough" [Undecided,Confirmed] | |
| 16:47:58 | cfriesen_ | is anyone aware of a spec around enabling L3 cache in the guests in libvirt? | |
| 16:48:13 | cfriesen_ | or even a blueprint | |
| 16:49:37 | openstackgerrit | Eric Young proposed openstack/nova-specs master: Add ScaleIO ephemeral storage backend https://review.openstack.org/495922 | |
| 17:11:52 | openstackgerrit | Sumedh Degaonkar proposed openstack/nova master: Adds 'sata' as a valid disk bus for qemu and kvm hypervisors https://review.openstack.org/459794 | |
| 17:21:07 | openstackgerrit | Chris Dent proposed openstack/nova master: WIP: [placement] manage cache headers https://review.openstack.org/495380 | |
| 18:37:55 | mriedem | https://review.openstack.org/#/c/495170/ looks legit, just needs some small things cleaned up | |
| 18:47:26 | dansmith | is anyone working on the actual fix for that? | |
| 18:47:42 | mriedem | i'm investigating | |
| 18:47:48 | mriedem | found some more fun | |
| 18:48:47 | mriedem | when you're force live migrating to a specific destination host, conductor checks that there is enough ram https://github.com/openstack/nova/blob/16.0.0.0rc1/nova/conductor/tasks/live_migrate.py#L103 | |
| 18:48:54 | mriedem | via the compute node info, not placement | |
| 18:49:17 | dansmith | ah interesting | |
| 18:49:18 | mriedem | not an immediate problem, | |
| 18:49:32 | mriedem | but will be once we don't track that anymore in the RT/claim | |
| 18:49:35 | dansmith | yeah | |
| 18:49:53 | mriedem | i'll report a bug to track that | |
| 18:49:55 | sean-k-mooney | is there an equivelent to nova reset-state in the openstack client? | |
| 18:54:43 | mriedem | dansmith: i think this existing bug will suffice to track that https://bugs.launchpad.net/nova/+bug/1427772 | |
| 18:54:44 | openstack | Launchpad bug 1427772 in OpenStack Compute (nova) "Instance that uses force-host still needs to run some filters" [Low,Confirmed] | |
| 18:55:00 | mriedem | i just left a new comment for the new state of the world wrt scheduler and compute and RT | |
| 18:55:13 | mriedem | and tagged with placement so it shows up in cdent's weekly email | |
| 18:56:11 | dansmith | I guess.. the link between the two is a little vague | |
| 18:56:56 | mriedem | it could be a separate bug | |
| 18:59:43 | mriedem | dansmith: so the bug is that when forcing the live migration to a specific host, we don't call scheduler_client.select_destinations | |
| 18:59:55 | mriedem | which is eventually the thing that does the double claim on source and dest | |
| 19:00:16 | mriedem | and we don't 'heal' the allocations since everything, at least in this scenario, is pike, so the RT isn't healing | |
| 19:00:48 | mriedem | because of https://review.openstack.org/#/c/491012/ | |
| 19:01:13 | mriedem | in both cases, conductor calls def check_can_live_migrate_destination on the dest host | |
| 19:01:33 | mriedem | so we could create the allocation for the instance and that compute there, or just do it in conductor if we're bypassing the scheduler | |
| 19:02:07 | mriedem | probably simpler to just do it in from conductor, plus then we don't have to worry about check_can_live_migrate_destination trampling over allocations that the scheduler already created | |
| 19:06:41 | dansmith | mriedem: right | |
| 19:07:31 | dansmith | mriedem: I guess I was thinking we should do select_destinations with a force_host of where we're expecting to go so that it creates the allocations properly | |
| 19:08:10 | dansmith | the whole point of the mess we made in pike was to stop managing allocations from separate places (i.e. the compute nodes) so adding another one is not ideal | |
| 19:08:41 | mriedem | yeah it wouldn't be fun doing a new thing from conductor just to mimic what the scheduler is doing | |
| 19:08:54 | mriedem | also because takashi wants to do this same thing for cold migrations | |
| 19:09:06 | mriedem | well, takashi and others... | |
| 19:09:34 | dansmith | aye | |
| 19:10:13 | mriedem | so now i'm looking at what conductor does when there isn't a forced destination host https://github.com/openstack/nova/blob/16.0.0.0rc1/nova/conductor/tasks/live_migrate.py#L153 | |
| 19:10:38 | mriedem | the request spec stuff gets pretty confusing in here | |
| 19:10:44 | mriedem | for compat things | |
| 19:11:24 | mriedem | we wouldn't call _find_destination i realize, | |
| 19:11:43 | mriedem | i'm just not sure how much we need to pass in the existing request spec with some "use this host, seriously" field set | |
| 19:13:42 | mriedem | would the scheduler also run the filters on the forced host in this case? | |
| 19:14:49 | mriedem | that might take the 'force-ness' out of it a bit if you fail to force because of a scheduler filter - but would allow us to remove this code https://github.com/openstack/nova/blob/16.0.0.0rc1/nova/conductor/tasks/live_migrate.py#L103 | |
| 19:14:59 | dansmith | mriedem: the scheduler won't know why you're running it | |
| 19:15:16 | dansmith | the point is, | |
| 19:15:32 | dansmith | the select_destinations() call should eliminate the need to manually check for memory space on the destination, | |
| 19:15:49 | mriedem | from conductor | |
| 19:15:51 | mriedem | i agree | |
| 19:15:53 | dansmith | because if it doesn't return the host you asked for then you fail with "sorry this peg doesn't fit in that hole" | |
| 19:15:55 | dansmith | yeah | |
| 19:16:03 | mriedem | yeah which is nice | |
| 19:16:07 | mriedem | removes my issue with that code not using placement | |
| 19:17:21 | mriedem | so it seems all we need to do then is have conductor set force_hosts/nodes on the request spec and call select_destinations - if that raises NoValidHost, we know what to do | |
| 19:17:47 | mriedem | i seriously feel like if i try to monkey with the request spec there is some landmine that will blow off at least 2 fingers | |
| 19:18:39 | dansmith | hah | |
| 19:18:54 | mriedem | this could also go away assuming you have the ComputeFilter enabled https://github.com/openstack/nova/blob/16.0.0.0rc1/nova/conductor/tasks/live_migrate.py#L85 | |
| 19:22:07 | mriedem | i think it's basically this that we need https://github.com/openstack/nova/blob/16.0.0.0rc1/nova/compute/api.py#L3880-L3884 | |
| 19:26:52 | mriedem | god, why doesn't this check if self.requested_destination is set when converting down to a legacy filter_properties dict? https://github.com/openstack/nova/blob/16.0.0.0rc1/nova/objects/request_spec.py#L347 | |
| 19:27:04 | mriedem | requested_destination essentially supersedes force_hosts/nodes doesn't it? | |
| 19:30:32 | dansmith | mriedem: afaik yeah | |
| 19:31:04 | mriedem | weird, ok, because it's not consulted at all, from what i can tell, when converting a request spec to an older version | |
| 19:32:10 | mriedem | oh you know what else, this bypasses our same-cell check | |
| 19:32:21 | mriedem | so you could force a host for live migration in another cell | |
| 19:33:39 | mriedem | this force live migration thing is all sorts of f'ed | |
| 19:35:02 | dansmith | yeah, so all the more reason to let it do the cell and memory check in one go of select_destinations right? | |
| 19:35:23 | mriedem | yeah, although i don't see where in the scheduler we check that the requested destination is in the same cell... | |
| 19:35:44 | mriedem | get_host_states_by_uuids ? | |
| 19:36:06 | mriedem | https://github.com/openstack/nova/blob/16.0.0.0rc1/nova/scheduler/host_manager.py#L635 | |