| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-06-19 | |||
| 14:15:01 | sq4ind | Thanks | |
| 14:15:53 | bauzas | efried: something like this ? http://paste.openstack.org/show/723830/ | |
| 14:16:14 | bauzas | efried: if so, looks a bit weird | |
| 14:16:30 | efried | bauzas: sec... | |
| 14:17:11 | bauzas | efried: oops, of course I meant http://paste.openstack.org/show/723831/ | |
| 14:18:27 | efried | bauzas: Does self.flags not let you set an opt that's not registered? | |
| 14:18:40 | bauzas | efried: unfortunately, no | |
| 14:18:44 | efried | boo. | |
| 14:18:49 | dansmith | or, fortunately. | |
| 14:18:55 | bauzas | heh | |
| 14:19:35 | bauzas | efried: self.flags is like conf.override | |
| 14:19:42 | bauzas | you need to have an existing group | |
| 14:19:57 | bauzas | actually, sec | |
| 14:20:46 | bauzas | efried: I think I can test it | |
| 14:22:04 | efried | bauzas: okay; the other option is to split apart _configure_gpu_mapping, one bit to register the groups and one to grab the values and do the validation. Then you can call the former from your test before you set self.flags; then call the latter. | |
| 14:22:39 | bauzas | efried: I think I can mock register_opts | |
| 14:24:18 | openstackgerrit | Lee Yarwood proposed openstack/nova master: Add regression test for bug #1764883 https://review.openstack.org/562072 | |
| 14:24:19 | openstackgerrit | Lee Yarwood proposed openstack/nova master: compute: Ensure pre-migrating instances are destroyed during init_host https://review.openstack.org/562284 | |
| 14:24:19 | openstack | bug 1764883 in OpenStack Compute (nova) "Evacuation fails if the source host returns while the migration is still in progress" [Undecided,In progress] https://launchpad.net/bugs/1764883 - Assigned to Lee Yarwood (lyarwood) | |
| 14:30:58 | openstackgerrit | Lee Yarwood proposed openstack/nova master: compute: Ensure pre-migrating instances are destroyed during init_host https://review.openstack.org/562284 | |
| 14:40:16 | mriedem | this should be an easy +W for someone to fix busted compat tests https://review.openstack.org/#/c/576256/ | |
| 14:42:51 | mriedem | zcorneli: i'm going to rebase your change on top of ^ ok? | |
| 14:50:05 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Implement file backed memory for instances in libvirt https://review.openstack.org/567876 | |
| 14:50:06 | openstackgerrit | Matt Riedemann proposed openstack/nova master: DNM: Libvirt file backed memory https://review.openstack.org/575766 | |
| 14:51:49 | do3meli | can someone please tell me whats the recommended way to evacuate all instances from a failed compute node (and let the scheduler decide where to place them)? we have opened https://bugs.launchpad.net/nova/+bug/1763039 and are currently bit confused on what to use. | |
| 14:51:50 | openstack | Launchpad bug 1763039 in OpenStack Compute (nova) "evacuate instance documentation not mentioning host-evacuate" [Undecided,Opinion] | |
| 14:51:57 | zcorneli | mriedem: Yep, go ahead (was hopefully going to get to that this morning) | |
| 14:52:23 | zcorneli | mriedem: dansmith: For Dan's patch to fix up ts, should we be fixing up the similar tests on other objects? | |
| 14:52:23 | mriedem | done, reviewing the spec and then the code in a bit | |
| 14:52:39 | mriedem | zcorneli: yeah | |
| 14:52:57 | mriedem | or at least reporting bugs that they are busted so low-hanging-fruit contributors can clean them up | |
| 14:53:14 | mriedem | like, the bug smash in china is this week, that would be a good opportunity | |
| 14:58:19 | openstackgerrit | zhangyangyang proposed openstack/nova master: Remove Chancefilter https://review.openstack.org/576384 | |
| 14:58:39 | dansmith | zcorneli: yeah I was just trying to get up a quick fix to let you depend on so you could roll forward | |
| 14:59:26 | zcorneli | mriedem: dansmith: I'll look at knocking out bug(s) for the remainder this afternoon then | |
| 15:02:28 | dansmith | mriedem: are you going to take another pass over that patch today? | |
| 15:04:30 | mriedem | dansmith: the file-backed one? i'm doing it now | |
| 15:04:39 | dansmith | ah cool | |
| 15:04:43 | mriedem | i will have a craigslist interruption here in a minute though | |
| 15:04:44 | openstackgerrit | zhangyangyang proposed openstack/nova master: Remove Chancefilter https://review.openstack.org/576384 | |
| 15:07:29 | mriedem | hmm, are we ready to remove the ChanceScheduler? ^ | |
| 15:07:36 | mriedem | i didn't even realize it was deprecated | |
| 15:08:18 | dansmith | mriedem: I thought they removed the personal services section? | |
| 15:08:51 | mriedem | dansmith: while laura is away... | |
| 15:08:55 | mriedem | comstud got me the backdoor deal | |
| 15:09:01 | dansmith | understandable | |
| 15:18:56 | kashyap | :D The sheer rawness... | |
| 15:19:15 | openstackgerrit | Brianna Poulos proposed openstack/python-novaclient master: Fix trusted-image-certificate-id help text https://review.openstack.org/576555 | |
| 15:29:48 | dansmith | kashyap: I meant personal coding services.. obviously mriedem is doing some side coding work for hire | |
| 15:29:59 | kashyap | Haha | |
| 15:30:20 | kashyap | Nice try to deflect the "original intention"... | |
| 15:31:05 | mriedem | doesn't matter, because in true craigslist form, the person confirms they are coming at 10am today, and it's now 10:30am and no-show | |
| 15:31:09 | openstackgerrit | Zhenyu Zheng proposed openstack/nova master: nova-manage db archive_deleted_rows is not multi-cell aware https://review.openstack.org/507486 | |
| 15:33:25 | mriedem | jacqueslist | |
| 15:35:40 | openstackgerrit | huanhongda proposed openstack/nova stable/ocata: [Stable Only] Remove soft-deleted instances from quota_usages https://review.openstack.org/570653 | |
| 15:35:41 | kashyap | Heh; that looks dodgy. /me stops going more off-topic, and heads offline. Tomorrow evening is my most difficult Dutch written exam; need to study. | |
| 15:49:50 | mriedem | zcorneli: dansmith: ok comments inline on the file-backed memory change https://review.openstack.org/#/c/567876/ | |
| 15:49:53 | mriedem | mostly good, | |
| 15:50:07 | mriedem | but i did wonder about the AggregateRamFilter for host aggregates that set ram_allocation_ratio!=1.0 | |
| 15:50:23 | mriedem | the spec didn't mention that nor the docs in the change | |
| 15:50:44 | mriedem | maybe it just means, don't put file-backed memory hosts in aggregates that have ram_allocation_ratio!=1.0 | |
| 15:51:56 | zcorneli | mriedem: Looking through commentary now | |
| 15:52:42 | mriedem | i think in general we will just have some gaps that we could shore up with proper scheduling filtering, including not landing a flavor that requests huge pages on a file-backed host | |
| 15:52:48 | mriedem | but we could also do follow ups on those | |
| 16:07:41 | openstackgerrit | Sylvain Bauza proposed openstack/nova master: libvirt: Configure multiple vGPU types https://review.openstack.org/564214 | |
| 16:07:42 | openstackgerrit | Sylvain Bauza proposed openstack/nova master: libvirt: mdevs returning parent and vendor PCI info https://review.openstack.org/562304 | |
| 16:07:43 | openstackgerrit | Sylvain Bauza proposed openstack/nova master: libvirt: Get inventories per vGPU type https://review.openstack.org/564215 | |
| 16:13:00 | zcorneli | mriedem: Most of the notes look good, will fix up what I can. Not sure how to handle the aggregates ram_allocation_ratio thing, except just documenting not to do that. | |
| 16:13:18 | mriedem | zcorneli: interested in dansmith's thoughts on that as well | |
| 16:13:40 | dansmith | mriedem: sorry I'm working on a downstream thing atm | |
| 16:13:41 | mriedem | zcorneli: as noted inline, i think we could robustify the scheduling landmines in follow ups if needed | |
| 16:13:50 | mriedem | in other words, i don't consider it blocking | |
| 16:15:29 | zcorneli | mriedem: I agree, I think it will act appropriately if it does get into any of those gaps (we're using immediate file allocation, instead of lazy, so qemu will fail to start up if you exceed the capacity of the backing store, for example). Then someone could look at the docs and see the notes about "Don't do this thing" | |
| 16:15:32 | TheJulia | out of curiosity, is the nova-v1-cells job in a healthy state? | |
| 16:15:44 | TheJulia | err, nova-cells-v1 | |
| 16:15:45 | zcorneli | Not the greatest situation, but it's at least "correct" | |
| 16:16:30 | dansmith | mriedem: if there was some hypervisor that couldn't do memory overcommit, and you put it into an aggregate with overcommit set, that'd be the same situation here | |
| 16:16:59 | dansmith | I don't recall what the solution was for aggregate-based overcommit post placement though | |
| 16:17:25 | mriedem | there are a few specs i think...but not clear on that either | |
| 16:17:41 | mriedem | TheJulia: there was a known regression over the weekend, resolved yesterday | |
| 16:17:51 | mriedem | http://status.openstack.org/elastic-recheck/#1777478 | |
| 16:17:53 | TheJulia | mriedem: thanks | |
| 16:18:39 | mriedem | zcorneli: the MemoryPagesUnsupported conflict will result in a reschedule to another host, | |
| 16:18:57 | mriedem | so yeah you might land on a host that supports huge pages but isn't configured for file-backed memory | |
| 16:19:11 | mriedem | assuming the operator hasn't used host aggregates to make those features mutually exclusive | |
| 16:19:46 | zcorneli | mriedem: Which seems OK, because file-backed isn't requested by the user / flavor. | |
| 16:19:58 | mriedem | it's just a lot of host aggregate whack-a-mole for the operator to have to deal with, and we aren't great about exlusion filtering | |
| 16:20:05 | mriedem | zcorneli: right, but huge pages are | |
| 16:20:45 | mriedem | that's why i was thinking, in the future we could report a capability for the host and a scheduler filter could say, 'oh this host is file-backed memory and the request has huge pages, i know those don't work, so exclude that host | |
| 16:21:05 | zcorneli | mriedem: So if they end up landing on a host without file-backed, but with huge pages, they've gotten a "correct" thing. But the scheduler definitely had to do some extra work in it. | |
| 16:21:23 | mriedem | well, or they got lucky | |
| 16:21:28 | mriedem | or the operator setup aggregates to make that happen | |
| 16:21:59 | zcorneli | Yea, agreed. I think we had talked something similar to this early on too, with the migration thing (<Rocky -> Rocky+file-backed). | |
| 16:22:01 | mriedem | anyway, we could optimize in the future if needed | |
| 16:22:09 | zcorneli | Having it be a capability so the scheduler can avoid that breakage. | |
| 16:22:11 | dansmith | mriedem: well, that would be baking some libvirt-ness into the scheduler a bit | |
| 16:22:15 | mriedem | i imagine operators will add their own scheduler filters for this | |
| 16:22:23 | dansmith | mriedem: because maybe hyper-v could do that, for example | |
| 16:22:35 | mriedem | dansmith: yeah i know | |
| 16:22:50 | dansmith | exposing some can-do-huge-pages thing and then having scheduler filter out those hosts if requested would be okay thoug, | |