| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-05-01 | |||
| 11:52:50 | openstackgerrit | Tetsuro Nakamura proposed openstack/nova master: Return all resources in provider_summaries https://review.openstack.org/558045 | |
| 11:52:51 | openstackgerrit | Tetsuro Nakamura proposed openstack/nova master: Return all nested providers in tree https://review.openstack.org/559480 | |
| 11:52:52 | openstackgerrit | Tetsuro Nakamura proposed openstack/nova master: Return resource providers without inventories https://review.openstack.org/559554 | |
| 11:52:53 | openstackgerrit | Tetsuro Nakamura proposed openstack/nova master: WIP: Support nested allocation candidates https://review.openstack.org/565487 | |
| 11:59:28 | openstackgerrit | sahid proposed openstack/nova master: hardware: only take into account small pages https://review.openstack.org/532168 | |
| 12:37:40 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Convert websocketproxy to use db for token validation https://review.openstack.org/333990 | |
| 12:37:41 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Update layout docs for running console proxies https://review.openstack.org/557489 | |
| 12:37:42 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Deprecate the nova-consoleauth service https://review.openstack.org/565367 | |
| 12:37:57 | mriedem | melwitt: fixed up that bottom patch so tests should pass now ^ | |
| 12:40:44 | efried | ō/ | |
| 13:01:47 | mriedem | vdrok: TheJulia: johnthetubaguy: are we ready for this now? https://review.openstack.org/#/c/508886/ or are people still needing this, i.e. do we have gaps in custom resource classes for ironic scheduling? | |
| 13:02:38 | johnthetubaguy | I think the traits support should cover the other cases now, but I could be miss-remembering | |
| 13:02:58 | johnthetubaguy | haven't heard that anyone has successfully moved between them though | |
| 13:04:20 | johnthetubaguy | although that was the non-exact case, thinking about it | |
| 13:07:10 | TheJulia | mriedem: my experience as of recent is that people are just now waking up to resource classes, but I think it should be safe to nuke those | |
| 13:08:29 | jroll | please do | |
| 13:08:41 | jroll | we should probably also raise an exception for nodes that don't have a resource class at this point | |
| 13:10:58 | jroll | mriedem: should we be removing the baremetal_enabled_filters option altogether? | |
| 13:11:34 | jroll | reading the deprecation reason, seems like we should | |
| 13:11:35 | TheJulia | jroll: well, we set it by default upon create now, the conundrum being actual user use | |
| 13:12:08 | jroll | TheJulia: we set a default resource class on node create? | |
| 13:12:16 | TheJulia | jroll: we do | |
| 13:12:22 | jroll | O_o | |
| 13:12:44 | jroll | also, we've been logging warnings since pike, I don't think we can hold back just because people are being lazy | |
| 13:13:07 | TheJulia | which people though? | |
| 13:13:27 | jroll | operators? | |
| 13:14:01 | TheJulia | well, the users of the flavors are the ones impacted | |
| 13:15:08 | openstackgerrit | Merged openstack/nova-specs master: Return resources of entire trees in Placement https://review.openstack.org/559466 | |
| 13:15:23 | jroll | TheJulia: sure, in the end, users are always the impacted people. but, like most cases, it's a result of operators not tending to their cloud properly. it's like if an operator didn't deploy the placement api before it was required, users are impacted because the operator didn't do their job, ya know? | |
| 13:15:42 | TheJulia | yup | |
| 13:15:47 | TheJulia | I do agree | |
| 13:15:50 | jroll | also, fwiw, there's only a default resource class if the operator sets one: https://github.com/openstack/ironic/blob/d917f32060d4f7237b9b4ecd6e0e769bc41d420c/ironic/conf/default.py#L74 | |
| 13:18:27 | TheJulia | hmm, I thought that was defaulted to a value, maybe I'm thinking of other config someplace | |
| 13:28:14 | mriedem | that reminds me of a bug in pike....sec | |
| 13:28:35 | mriedem | https://bugs.launchpad.net/nova/+bug/1765334 | |
| 13:28:35 | openstack | Launchpad bug 1765334 in OpenStack Compute (nova) "Ironic resource class may not be put into inventory in Pike" [Undecided,New] | |
| 13:28:54 | mriedem | if you set the resource class on the node after it's created, not during create, the inventory for it doesn't get represented in placement | |
| 13:30:16 | artom | Should we update the dashboard links from https://wiki.openstack.org/wiki/Nova/GerritDashboards ? 'cuz they're showing up empty for me | |
| 13:30:23 | mriedem | inadvertantly fixed in queens though so i'm not sure what, if anything, we should do about that bug in pike | |
| 13:30:42 | mriedem | artom: sure, they're likely all busted due to new gerrit | |
| 13:31:11 | artom | mriedem, I was secretly hoping someone else would do it, or at least show me how ;) | |
| 13:31:20 | artom | (Not the wiki part, the gerrit URL part) | |
| 13:31:22 | jroll | mriedem: any reason not to backport https://review.openstack.org/#/c/518294/ ? | |
| 13:32:03 | mriedem | jroll: idk, it's tied to the vgpu blueprint, i don't know if there would be other side effects | |
| 13:32:27 | jroll | heh, was hoping you'd know because I certainly don't :) | |
| 13:34:15 | artom | Ah, cuz it's not Jenkins that's voting now, it's Zuul | |
| 13:36:09 | mriedem | ok so baremetal_enabled_filters is only used if use_baremetal_filters is True, and that defaults to False | |
| 13:36:35 | mriedem | i've got a note that we can't remove that stuff until we know all filters work with ironic, and i had a suspicion at one point that NUMATopologyFilter wouldn't work with ironic nodes, but can't really remember why now, | |
| 13:37:03 | mriedem | i posted a devstack change to enable the NUMATopologyFilter and it passed the ironic CI, but none of the requests would have stressed that filter since we don't do PCI/NUMA testing in the normal gate | |
| 13:37:25 | mriedem | so it would be nice if someone that does both pci/numa and ironic could try that out before we removed those deprecated options | |
| 13:37:57 | artom | Does a NUMATopologyFilter even make sense for Ironic? You're getting the whole machine regardles... | |
| 13:38:34 | mriedem | if your single compute installation supports both vms and bms | |
| 13:38:39 | mriedem | then you'll need to run that | |
| 13:38:45 | mriedem | if you support numa + vm | |
| 13:39:04 | artom | mriedem, ah, in a mixed cloud case | |
| 13:39:18 | artom | mriedem, well, the filter does check for host_topology, and if its absent, passes the host | |
| 13:39:27 | artom | What does an ironic host expose for that? | |
| 13:39:35 | mriedem | nothing | |
| 13:39:49 | mriedem | so it might be ok, and i can't remember what at the time was making me think it might not work | |
| 13:39:51 | artom | Ah, no, hold up, I was wrong | |
| 13:40:08 | mriedem | i think i found something during code inspection | |
| 13:40:09 | artom | If you request a topology and the host doens't have it, it fails the host | |
| 13:40:17 | artom | Which... is fine, actually :) | |
| 13:40:23 | mriedem | right it would skip the ironic node | |
| 13:40:27 | mriedem | which is what we want | |
| 13:40:54 | artom | And if you request nothing and the host has nothing, it passes | |
| 13:41:02 | artom | Which is also fine | |
| 13:42:05 | mriedem | owalsh: does tripleo still rely on the exact core/ram/disk filters for baremetal? | |
| 13:42:41 | mriedem | artom: yeah which was shown in https://review.openstack.org/#/c/553412/ | |
| 13:43:26 | artom | mriedem, logs for that have been purged :( | |
| 13:43:57 | artom | But the ironic job passed, so yey :) | |
| 13:44:47 | mriedem | yeah i verified it ran and passed the filter in the ironic job, | |
| 13:44:55 | mriedem | but as noted, the flavors used don't have any numa topology requests | |
| 13:45:17 | artom | Hrmm, so it didn't fully test it | |
| 13:45:33 | artom | In fact, it didn't test the host topology bit at all :( | |
| 13:45:37 | artom | And that's the thing we're unsure about | |
| 13:46:14 | artom | Since numa_topology_filter.py L86 will evaluate to false immediately when there's no request_topology | |
| 13:46:53 | mriedem | jroll: uh oh https://docs.openstack.org/ironic/latest/install/configure-compute.html still tells people to set use_baremetal_filters=True | |
| 13:47:08 | mriedem | artom: right | |
| 13:47:12 | dansmith | artom: if you want to get a machine with a specific topology (like more than one node) wouldn't you want the filter? | |
| 13:47:33 | jroll | mriedem: oof | |
| 13:47:34 | artom | dansmith, for Ironic? Wouldn't that be a custom trait? | |
| 13:47:40 | mriedem | i could try and hack up a numa functional test in-tree and use the ironic driver, or a fake ironic driver | |
| 13:47:50 | artom | Like SUPER_GOLD_3_NUMA_NODES or whatever? | |
| 13:47:53 | dansmith | artom: ah, I guess with CUSTOM_IRONIC_SILVER type scheduling that's true | |
| 13:49:28 | dansmith | artom: just service levels, if you have a couple different classes of box | |
| 13:50:08 | mriedem | if you combine 5 of them you get a power ranger super bot thing | |
| 13:50:31 | artom | Nah brah, Captain Planet | |
| 13:50:37 | mriedem | lame | |
| 13:50:41 | artom | Dude | |
| 13:50:45 | artom | He's our hero | |
| 13:50:55 | artom | Going to bring pollution down to zero | |
| 13:50:58 | mriedem | only a canadian would think captain tree hugger is a cool super hero | |
| 13:51:02 | artom | How can you not dig that? | |
| 13:51:39 | mriedem | speaking of, https://adequateman.deadspin.com/why-your-childrens-television-program-sucks-paw-patrol-1758314083#_ga=2.144903474.746352317.1525182679-529658380.1521051074 | |
| 13:51:48 | artom | Yeah well, I'll be here smelling my flowers while you choke on your coal roller | |
| 13:52:16 | mriedem | coal jobs f yeah! 'merica! | |
| 13:52:27 | mriedem | ok, ok, so back to this | |
| 13:52:34 | mriedem | i guess we can drop the exact filters | |
| 13:52:36 | dansmith | yeah, so this went off the rails :) | |