| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-04-06 | |||
| 13:44:48 | johnthetubaguy | mriedem: but anyways, seems worth a peak | |
| 13:45:06 | mriedem | so i was worried about that, | |
| 13:45:14 | mriedem | because if you don't configure the scheduler things on the compute node, | |
| 13:45:29 | mriedem | that means you get the default, so nova compute is doing all of this extra rpc traffic to send something your scheduler doesn't care about | |
| 13:45:49 | johnthetubaguy | yeah, that does suck | |
| 13:46:33 | johnthetubaguy | we had this debate a few times, we should really look at being more specific on the idea of what configurations should be set where | |
| 13:47:14 | mriedem | i didn't realize until yesterday that this was scheduler config used on the comptues | |
| 13:47:18 | mriedem | but yeah it's pretty nasty | |
| 13:47:28 | mriedem | i guess the silver lining is, | |
| 13:47:32 | mriedem | when caching scheduler is gone, | |
| 13:47:37 | mriedem | so is this config option | |
| 13:47:57 | mriedem | because the only options you'd have then are filter, chance and fake, and the latter two are test-only | |
| 13:48:34 | bauzas | mriedem: remember we have 3 drivers in tree | |
| 13:48:37 | bauzas | not only 2 | |
| 13:48:47 | mriedem | we have 4 drivers in tree | |
| 13:48:54 | mriedem | caching, filter, chance and fake | |
| 13:49:01 | mriedem | the scheduler.driver option uses the choices kwarg | |
| 13:49:11 | mriedem | so unless you fork that, it's a ValueError to specify something else | |
| 13:49:25 | johnthetubaguy | remembers macsz saying something about stevedoor^ | |
| 13:50:25 | mriedem | https://github.com/openstack/nova/blob/master/nova/conf/scheduler.py#L58 | |
| 13:50:42 | mriedem | https://github.com/openstack/nova/blob/master/nova/conf/scheduler.py#L79 | |
| 13:50:50 | mriedem | i don't see how ^ is right since we use choices | |
| 13:50:56 | mriedem | "A custom scheduler driver. In this case, you will be responsible for creating and maintaining the entry point in your 'setup.cfg' file" | |
| 13:51:05 | mriedem | choices=("filter_scheduler", "caching_scheduler", "chance_scheduler", "fake_scheduler"), | |
| 13:51:11 | bauzas | mriedem: FWIW, the FakeScheduler is not existing in tree | |
| 13:51:30 | bauzas | afaik, it's only in the tests tree | |
| 13:51:45 | bauzas | you can't deploy it | |
| 13:51:47 | johnthetubaguy | yeah, its meant to be stevedoor I think | |
| 13:52:10 | dansmith | mriedem: yeah that scheduler config is nasty, and it's an upcall :( | |
| 13:52:11 | bauzas | johnthetubaguy: we already load with stevedore :) | |
| 13:52:19 | dansmith | mriedem: that was part of the source of my frustration with this crap last week | |
| 13:52:24 | johnthetubaguy | oh, maybe choices was a step too far then | |
| 13:52:47 | bauzas | dansmith: oh, you weren't knowing that computes fanout the scheduler ? | |
| 13:52:50 | mriedem | the choices definitely breaks any stevedore loading of custom drivers | |
| 13:53:11 | dansmith | bauzas: I knew we had this but it's not something I think about much | |
| 13:53:39 | bauzas | dansmith: maybe we could use placement rather, because it's for providing aggregate and instance values | |
| 13:54:07 | dansmith | bauzas: we can't use placement for the affinity bits yet, which is what this is for, AFAIK | |
| 13:54:07 | mriedem | https://github.com/openstack/nova/commit/fe3d6dba3d1db8f97dab4baffcc11dda56368096 | |
| 13:54:09 | bauzas | dansmith: I created those RPCAPI methods because I wasn't accepting the scheduler filters calling DB | |
| 13:54:40 | bauzas | mriedem: anyway, operators can run custom filters that call their own scheduler :) | |
| 13:54:42 | mriedem | anyway i'm pretty sure https://github.com/openstack/nova/blob/master/nova/conf/scheduler.py#L79 doesn't work | |
| 13:54:50 | mriedem | bauzas: that's fine | |
| 13:55:00 | mriedem | but if you can't load up your own scheduler "driver" we should remove this line https://github.com/openstack/nova/blob/master/nova/conf/scheduler.py#L79 | |
| 13:55:20 | bauzas | mriedem: I'm fine with stopping to have a custom driver | |
| 13:55:22 | mriedem | https://github.com/openstack/nova/commit/6832a417c93046750ec7eff51eb59be802fd70dc | |
| 13:55:41 | dane-fichter | mriedem: peter mentioned that you wanted us to look into https://bugs.launchpad.net/nova/+bug/1516703 ? | |
| 13:55:41 | openstack | Launchpad bug 1516703 in OpenStack Compute (nova) "crypto.py generates certs with SHA-1 digest" [Medium,New] | |
| 13:55:43 | bauzas | rather, in case operators want to use their own scheduler, they just need to use a filter | |
| 13:55:43 | johnthetubaguy | mriedem: yeah, leaving the custom bit in there is misleading/inconsistent | |
| 13:56:12 | mriedem | dane-fichter: yeah it's just an old bug, if it doesn't make sense we should invalidate or mark as won't fix | |
| 13:56:29 | bauzas | mriedem: like I also said, https://github.com/openstack/nova/blob/master/nova/conf/scheduler.py#L61 is wrong, because fake_scheduler doesn't exist in tree | |
| 13:56:48 | johnthetubaguy | mriedem: that might actually be a hyper-v thing from claudiub and team? | |
| 13:56:51 | mriedem | it has to be in choices to workaround oslo | |
| 13:57:05 | mriedem | johnthetubaguy: not sure what that means | |
| 13:57:09 | dane-fichter | mriedem: nova cert is being removed this cycle right? Seems like this functionality should be removed as well | |
| 13:57:14 | johnthetubaguy | mriedem: we added that for hyper-v I think | |
| 13:57:18 | mriedem | dane-fichter: there is a spec for that | |
| 13:57:25 | mriedem | johnthetubaguy: added what for hyperv? | |
| 13:57:26 | mriedem | fake_scheduler? | |
| 13:57:33 | johnthetubaguy | x.509 cert generation | |
| 13:57:39 | johnthetubaguy | to replace the ssh keypair type | |
| 13:57:41 | bauzas | dansmith: anyway, I tried to think about reschedules | |
| 13:57:48 | bauzas | dansmith: I still think we need them | |
| 13:57:57 | bauzas | because of X or Y | |
| 13:58:37 | dansmith | X or Y? | |
| 13:58:52 | mriedem | nova meeting in 2 minutes | |
| 14:00:23 | openstackgerrit | Roman Podoliaka proposed openstack/osc-placement master: WIP: implement API version 1.0 https://review.openstack.org/451877 | |
| 14:02:14 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Remove a fallacy in scheduler.driver config option help text https://review.openstack.org/454196 | |
| 14:04:17 | claudiub | mriedem: johnthetubaguy you mean the x509 certificates? | |
| 14:04:29 | johnthetubaguy | yeah, for windows auth | |
| 14:04:30 | claudiub | those would the windows equivalent of ssh keys | |
| 14:04:43 | claudiub | so, not hyper-v specific, windows guests specific | |
| 14:05:19 | dane-fichter | johnthetubaguy: So what I'm hearing is that generate_winrm_x509_cert() is still needed? | |
| 14:05:30 | johnthetubaguy | dane-fichter: I think so, yeah | |
| 14:05:38 | johnthetubaguy | dane-fichter: even after we kill nova-cert | |
| 14:05:51 | openstackgerrit | Dan Smith proposed openstack/nova master: Add functional test for two-cell scheduler behaviors https://review.openstack.org/452006 | |
| 14:05:54 | dane-fichter | johnthetubaguy: ok. good to know. | |
| 14:06:29 | johnthetubaguy | dane-fichter: although I am willing to believe its something we should deprecate in the API, us generating certs is probably the wrong thing to do for winRM, but I am not sure | |
| 14:06:47 | johnthetubaguy | dane-fichter: needs a look, I guess | |
| 14:08:04 | dane-fichter | johnthetubaguy: I'll take a look at it. | |
| 14:16:11 | ildikov | johnthetubaguy: mriedem: lyarwood: hi, a quick reminder that we plan to switch to 1600 UTC with the Cinder-Nova meeting today | |
| 14:17:03 | johnthetubaguy | ah, I forgot about tht | |
| 14:17:44 | ildikov | johnthetubaguy: sorry, I should've sent out a reminder mail earlier, I only got to propose a patch to change the slot officially, it's not merged yet | |
| 14:17:53 | ildikov | johnthetubaguy: can you make it to the earlier slot? | |
| 14:18:04 | johnthetubaguy | ildikov: that should be OK | |
| 14:18:13 | ildikov | johnthetubaguy: cool, tnx | |
| 14:18:48 | openstackgerrit | Scott Moser proposed openstack/nova master: Limit exposure of network device types to the guest. https://review.openstack.org/400883 | |
| 14:27:45 | openstackgerrit | Scott Moser proposed openstack/nova master: Limit exposure of network device types to the guest. https://review.openstack.org/400883 | |
| 14:38:13 | mriedem | sfinucan: artom: https://github.com/openstack/nova/blob/master/nova/tests/functional/test_metadata.py is what i was thinking of | |
| 14:38:46 | mriedem | we'd have to fix mikal's sins here for network data https://github.com/openstack/nova/blob/master/nova/tests/functional/test_metadata.py#L64 | |
| 14:39:02 | mriedem | since ^ is all using nova-network right now | |
| 14:44:07 | bauzas | mriedem: wait, we unset instance.host when we offload the instance from the compute ? | |
| 14:44:18 | bauzas | (after shelving) | |
| 14:45:01 | bauzas | I know it requires a reschedule, but I was thinking we were keeping the reference before the unshelve and that were just overriding the host value | |
| 14:45:13 | dansmith | bauzas: no, once offloaded, it loses host, AFAIK | |
| 14:45:23 | bauzas | dansmith: so in the compute? | |
| 14:45:28 | bauzas | because AFAIK, you shelve | |
| 14:45:40 | bauzas | and then, a couple of mins after (config) then it's offloaded | |
| 14:46:02 | bauzas | AFAIK, it's a periodic task running on the compute which offloads the instance off | |
| 14:46:06 | dansmith | right | |