| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-06-29 | |||
| 14:24:02 | sean-k-mooney | sfinucan: cool thanks | |
| 14:25:42 | openstackgerrit | Stephen Finucane proposed openstack/nova master: libvirt: create functional test base class https://review.openstack.org/407055 | |
| 14:25:42 | openstackgerrit | Stephen Finucane proposed openstack/nova master: tests: Validate huge pages https://review.openstack.org/399653 | |
| 14:29:06 | openstackgerrit | Ed Leafe proposed openstack/nova master: Extract custom resource classes from flavors https://review.openstack.org/473627 | |
| 14:30:51 | openstackgerrit | Merged openstack/os-vif master: Updated from global requirements https://review.openstack.org/467173 | |
| 14:31:21 | edleafe | mriedem: ^^ | |
| 14:32:46 | mriedem | oooo just in time | |
| 14:35:23 | mriedem | wow that was a nasty diff | |
| 14:35:51 | edleafe | mriedem: yeah, the section I changed was moved to another file | |
| 14:35:56 | edleafe | mriedem: always fun! | |
| 14:39:26 | mriedem | edleafe: dead code in the rebase | |
| 14:40:12 | edleafe | mriedem: yeah? | |
| 14:40:17 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Extract custom resource classes from flavors https://review.openstack.org/473627 | |
| 14:40:38 | mriedem | https://review.openstack.org/#/c/473627/10..11/nova/scheduler/utils.py | |
| 14:40:47 | edleafe | ugh | |
| 14:40:52 | edleafe | thanks for fixing that | |
| 14:41:05 | edleafe | odd that pep8 didn't catch that | |
| 14:41:37 | mriedem | +2 | |
| 14:41:45 | mriedem | now it's time | |
| 14:42:02 | edleafe | sweet dreams | |
| 14:42:13 | mriedem | o7 | |
| 14:47:32 | sean-k-mooney | jaypipes: bauzas if ye get a chance can ye take a look at https://review.openstack.org/#/q/topic:enable-sriov-nic-features | |
| 14:48:36 | sean-k-mooney | jaypipes: bauzas https://review.openstack.org/#/c/449257/28 in partcalar has been stuck for about a month without any really feedback but that feature is more or less complete if we can get those last few pathces over the line | |
| 14:49:10 | bauzas | sean-k-mooney: sure, no promise but I'll try to look :) | |
| 14:49:31 | bauzas | sean-k-mooney: I was horribly off for a few days before so got a shit ton of things to do now :p | |
| 14:50:14 | openstackgerrit | Merged openstack/nova master: api-ref: Fix missing parameters in API Versions https://review.openstack.org/475226 | |
| 14:50:37 | sean-k-mooney | bauzas: cool understood. life as a core gets hecktic espcially if your unwell for a few days. hope your feeling better | |
| 14:50:56 | openstackgerrit | Merged openstack/nova master: Stop using mox stubs in test_keypairs.py https://review.openstack.org/477384 | |
| 14:57:05 | openstackgerrit | Merged openstack/nova master: Fix redundant BDM lookups during rebuild https://review.openstack.org/476258 | |
| 15:03:48 | openstackgerrit | Merged openstack/nova master: Cleanup the plethora of libvirt live migration options https://review.openstack.org/471834 | |
| 15:04:33 | openstackgerrit | Merged openstack/nova master: Fix the note at the end of allocate_for_instance https://review.openstack.org/475578 | |
| 15:09:33 | mmehan | Need a help from Numa experts, I have a unique sitatuation | |
| 15:10:02 | mmehan | the VM i want to spin is using two nics from different numa and has direct port | |
| 15:10:23 | mmehan | the VM when spinned gets half cpus from Numa0 and other half from Numa1 | |
| 15:10:51 | sfinucan | mmehan: It already gets cores from two NUMA nodes, or you want it to get them that way? | |
| 15:10:54 | sean-k-mooney | mmehan: yes this is the default | |
| 15:11:12 | mmehan | what the VM vendor want is even if the nic ports are across numa they want to give cpus and memory from one numa | |
| 15:11:33 | sfinucan | mmehan: We can't do that...yet | |
| 15:12:21 | sfinucan | mmehan: You're going to need this https://review.openstack.org/#/c/361140/ | |
| 15:12:33 | sfinucan | but it's not happening in Pike, so you'll need to wait for Queens at least | |
| 15:12:36 | jaypipes | sean-k-mooney: yup, will do today. today is review day while I sit on yet another 5 hour sales call. | |
| 15:12:51 | sean-k-mooney | mmehan: currently it is expcitly not supported because that intorduce alot of qpi overhead and reduce the performace of the plathform as a whole. the spec sfinucan linked make it configurable | |
| 15:13:05 | sfinucan | ^ what he said :) | |
| 15:14:00 | mmehan | is yes i understand QPI overhead, but thats needed to keep platform uniform as of now | |
| 15:14:28 | mmehan | thanks for spec, i will go over it | |
| 15:15:12 | sfinucan | mmehan: I'm pretty sure there are patches up there that you could test against too | |
| 15:17:06 | dansmith | jaypipes: edleafe are you both good with this now such that I should review? https://review.openstack.org/#/c/475448/ | |
| 15:17:13 | dansmith | bauzas: are you actively reviewing that one? | |
| 15:17:27 | jaypipes | dansmith: I am, yes. obviously I can't vote +2 on it :) | |
| 15:17:36 | dansmith | jaypipes: yep | |
| 15:17:44 | jaypipes | thanks much to ed for fixing my style. | |
| 15:18:26 | mmehan | sfinucan: I will try to find the patches, if you have any insight please share | |
| 15:18:27 | cfriesen | mmehan: I think this is the main patch: https://review.openstack.org/#/c/390520/ | |
| 15:18:38 | sfinucan | https://review.openstack.org/#/q/topic:bp/share-pci-between-numa-nodes | |
| 15:20:58 | strigazi | Hi cores, can you have a look at https://review.openstack.org/#/c/476122 ? | |
| 15:22:20 | edleafe | dansmith: was just giving it one last scan before re-applying my +1 | |
| 15:30:30 | sfinucan | cfriesen: If I'm running a RT guest, I need an RT host kernel, right? | |
| 15:34:02 | bauzas | dansmith: yup | |
| 15:34:45 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: Enable cold migration with target host(1/2) https://review.openstack.org/408955 | |
| 15:36:17 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: Enable cold migration with target host(2/2) https://review.openstack.org/408964 | |
| 15:36:23 | cfriesen | sfinucan: well, technically no but your RT response won't be as good. | |
| 15:36:50 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: Add functional tests for cold migration to same host https://review.openstack.org/414926 | |
| 15:37:06 | openstackgerrit | Takashi NATSUME proposed openstack/nova master: api-ref: Add parameters in cold migrate action https://review.openstack.org/410042 | |
| 15:37:07 | cfriesen | sfinucan: so if you actually care about RT then yes. | |
| 15:38:03 | sfinucan | cfriesen: Fair enough :) I'm trying to summarize this and will reply to the mail once I do | |
| 15:40:51 | sfinucan | cfriesen: So what I'm getting out of this is that 'emulator_threads_policy' is too resource intensive, and we should be instead allocating "overhead threads" on one or two host cores | |
| 15:41:01 | sfinucan | Is that right, or do I need to delve deeper? | |
| 15:41:14 | sfinucan | /reread the thread | |
| 15:41:18 | cfriesen | sfinucan: lol | |
| 15:42:03 | cfriesen | sfinucan: basically that's the opinion of some people--that dedicated a whole separate pCPU to the emulator threads consumes too many pCPUs | |
| 15:42:19 | cfriesen | s/dedicated/dedicating | |
| 15:43:21 | sfinucan | Grr. I said that exact thing during reviews of that spec, but didn't put it succinctly enough | |
| 15:43:47 | cfriesen | sfinucan: I'm not entirely convinced that sharing emulator threads on a single pCPU is *guaranteeed* to be safe, like if one instance starts seriously hammering on I/O or does live migration or something | |
| 15:44:17 | sfinucan | I think I suggested using 'vcpu_pin_set', but another configuration option (a mask, from the looks of things?) is much better | |
| 15:44:33 | sfinucan | cfriesen: I've just reached that mail now, heh | |
| 15:47:58 | cfriesen | sfinucan: hence my overcommit ratio suggestion, then operators could tune the number of RT instances per "emulator pCPU" to whatever they feel comfortable with..setting it to 1 gives you the behaviour of "emulator_threads_policy=isolate" | |
| 16:08:21 | openstackgerrit | Merged openstack/nova master: XenAPI: device tagging https://review.openstack.org/333781 | |
| 16:08:34 | efried | sfinucan Found bugs in https://review.openstack.org/#/c/476159/ :) | |
| 16:17:53 | openstackgerrit | Dan Smith proposed openstack/nova master: Sanitize instance in InstanceMetadata to avoid un-pickleable context https://review.openstack.org/478991 | |
| 16:25:39 | sfinucan | efried: How dare you | |
| 16:25:48 | sfinucan | I'll take a look shortly :) | |
| 16:27:09 | efried | sfinucan I didn't touch that one. | |
| 16:27:42 | efried | sfinucan BTW, I don't think your monkeypatch will play well with my explicit newlines and box sizes, except by coincidence. | |
| 16:28:06 | sfinucan | I was planning to drop the newlines and most of the box sizes, tbh | |
| 16:28:25 | sfinucan | the only box size I planned to keep was the one for the box that's currently truncated | |
| 16:29:08 | efried | sfinucan As you wish; though the squished-in bottoms make it look like something's truncated (even when it's not). | |
| 16:29:17 | efried | And I think there might be two long boxes now. | |
| 16:29:31 | sfinucan | efried: good call. Maybe I should keep all the sizes so and just strip the newlines? | |
| 16:29:37 | sfinucan | Given that they're not necessary | |
| 16:30:05 | efried | sfinucan Either way, unless you figure out how to autosize the boxes, it'll need to be trial and error to get it rendering right. | |
| 16:30:28 | sfinucan | Yup. Hopefully we can figure something out instead :( | |
| 16:44:47 | openstackgerrit | Merged openstack/nova master: Add keypairs field to InstanceCreatePayload https://review.openstack.org/419730 | |
| 16:45:34 | openstackgerrit | Merged openstack/nova master: Add key_name field to InstancePayload https://review.openstack.org/463002 | |
| 17:03:49 | dansmith | artom: why in the metadata test did you subclass FakeDeviceMetadata and FakeDeviceBus ? | |
| 17:04:33 | dansmith | those aren't registered objects, which is a problem for deserializing an Instance, but it also doesn't seem to matter if I use the real devices | |
| 17:04:37 | dansmith | er real objects | |
| 17:10:38 | jaypipes | sean-k-mooney: https://review.openstack.org/#/c/449257/ needs rebasing. can you handle? not sure where ralonso is today. | |
| 17:11:01 | sean-k-mooney | jaypipes: he went home about an hour ago | |
| 17:11:16 | jaypipes | sean-k-mooney: can you rebase or perhaps he can do in morning? | |
| 17:11:34 | sean-k-mooney | ill ask him to do it in the morning | |