| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-03-29 | |||
| 19:00:59 | fghaas | you're being awesome melwitt :) But I figured everyone including cfriesen was rightfully busy so I posted this to the mailing list (http://lists.openstack.org/pipermail/openstack/2018-March/046313.html) — but I'll be happy to follow up with a writeup from whatever we discuss here! | |
| 19:01:54 | melwitt | a-ha, cool | |
| 19:03:30 | cfriesen | fghaas: the Intel CPU models in libvirt/qemu generally do not include the "vmx" flag, as you have discovered. The functionality that you're looking for is exposed in kashyap's feature ("https://review.openstack.org/#/c/534384/ ") which is currently under review. | |
| 19:03:44 | kashyap | fghaas: Hi :-) | |
| 19:04:06 | melwitt | I wondered if kashyap's patch would be the answer. noyce | |
| 19:04:26 | cfriesen | fghaas: note that I've experienced instability with nested kvm, but kashyap says it's good with more recent kernel/qemu/libvirt | |
| 19:05:14 | kashyap | Yes; cfriesen -- let's not repeat "unstability" without data :-). I recall participating in an email thread w/ fghaas on KVM list about :-) | |
| 19:05:41 | kashyap | cfriesen: BTW -- Paolo (one of the KVM maintainers) said he's going the toggle the 'nested' flag to default for Intel in the upstream kernel "soon" | |
| 19:06:33 | kashyap | (It's on for AMD in the upstream kernel; but amusingly enough ... AMD is more instable w/ 'nested' than Intel!) | |
| 19:06:56 | cfriesen | kashyap: isn't it on for the AMD cpu models too? | |
| 19:07:17 | kashyap | cfriesen: See my previous comment. It *is* on for AMD. But not for Intel. At this moment. | |
| 19:07:33 | cfriesen | kashyap: I meant in libvirt/qemu, not kernel | |
| 19:09:17 | kashyap | cfriesen: What do you mean? There's no such thing as "enabling nested in libvirt / QEMU". As you know, you can configure Kernel, then tell libvirt, which inturn tells QEMU to use it. | |
| 19:10:36 | cfriesen | kashyap: I meant that the "svm" flag is included in the AMD cpu models | |
| 19:11:49 | kashyap | cfriesen: Ah, that's true. | |
| 19:12:04 | kashyap | It's late here, /me is losing ability to parse text messages | |
| 19:12:32 | openstackgerrit | Jay Pipes proposed openstack/nova master: tests for alloc candidates with nested and traits https://review.openstack.org/531899 | |
| 19:12:33 | openstackgerrit | Jay Pipes proposed openstack/nova master: placement: resource requests for nested providers https://review.openstack.org/554529 | |
| 19:13:05 | openstackgerrit | Mathieu Gagné proposed openstack/nova master: Fix rebuild of baremetal instance when vm_state is ERROR https://review.openstack.org/523559 | |
| 19:14:52 | openstackgerrit | Jay Pipes proposed openstack/nova-specs master: Standardize CPU resource tracking https://review.openstack.org/555081 | |
| 19:21:18 | fghaas | cfriesen, yes kashyap and I discussed the pros and cons at length recently, but this is really a generic question about cpu feature flags | |
| 19:21:35 | fghaas | not so much passing in the "vmx" flag specifically | |
| 19:22:26 | fghaas | but yeah that patch looks excellent :) | |
| 19:23:50 | melwitt | yeah, feature is being added this cycle https://blueprints.launchpad.net/nova/+spec/libvirt-cpu-model-extra-flags but the backports are going to be restricted to only the 'pcid' flag to address the meltdown/spectre perf mitigation. we weren't intending to backport a brand new feature in a free-form way | |
| 19:24:03 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Move get_stashed_volume_connector to compute.utils https://review.openstack.org/557810 | |
| 19:24:03 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Migrate old style volume attachments on nova-compute startup https://review.openstack.org/549130 | |
| 19:25:14 | melwitt | (to not add unknown instability to branches that are supposed to be considered quite stable) | |
| 19:43:42 | mriedem | wolfenstein | |
| 19:43:45 | mriedem | comin' for ya | |
| 19:57:40 | openstackgerrit | Merged openstack/nova master: Add request filter functionality to scheduler https://review.openstack.org/544730 | |
| 20:07:10 | openstackgerrit | Merged openstack/nova master: Add aggregates list to Destination object https://review.openstack.org/544729 | |
| 20:07:18 | openstackgerrit | Merged openstack/nova master: Make get_allocation_candidates() honor aggregate restrictions https://review.openstack.org/547990 | |
| 20:07:29 | openstackgerrit | Merged openstack/nova master: Add an index on aggregate_metadata.value https://review.openstack.org/555851 | |
| 20:20:54 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Add nova-status check for ironic flavor migration https://review.openstack.org/527541 | |
| 20:24:46 | openstackgerrit | Merged openstack/nova stable/queens: vmware: Fixes _detach_instance_volumes method https://review.openstack.org/557727 | |
| 20:26:29 | fghaas | melwitt: bummer, I'd love to see a backport for both pcid and vmx, but I realize I'm the nutcase here (as kashyap can attest, I'm sure) :) | |
| 20:30:38 | melwitt | fghaas: from what I understand, it's not too risky but we really don't ever backport full-fledged features to stable branches. stable is reserved for bug fixes. the reason we're making an exception and backporting the necessary bits for meltdown/spectre situation is because it would make a significant difference in reducing the performance degradation | |
| 20:31:11 | fghaas | Sure, I do follow that reasoning. | |
| 20:35:21 | cfriesen | fghaas: if you want to experiment, you can locally remove the couple lines of code in the backport limiting it to "pcid" and restart nova-compute. But if it causes problems, it's officially unsupported. | |
| 20:37:00 | fghaas | Entirely understood. But great to know that this will be an easy stop to pull, figuratively speaking | |
| 20:37:57 | melwitt | true that | |
| 20:44:49 | fghaas | ok, thanks everyone (particularly melwitt, cfriesen and kashyap), that's been extremely helpful information! | |
| 20:45:56 | melwitt | np | |
| 20:48:56 | efried | melwitt: Feel like pushing https://review.openstack.org/#/c/557508/ ? "Customer" is buggin me :) | |
| 20:50:19 | efried | jaypipes: Easy reviews in the upt bp (but outside the series so you mighta missed 'em): https://review.openstack.org/#/c/553475/ and (less critical) https://review.openstack.org/#/c/553476/ | |
| 20:50:21 | melwitt | it has been tab'd | |
| 20:50:25 | efried | melwitt: Thanks! | |
| 20:50:30 | melwitt | nova meeting in 10 minutes | |
| 21:01:43 | melwitt | gdi my irc client is freaking out. sorry | |
| 21:14:38 | fghaas | whoa melwitt you beat me to the ML writeup — thanks again! | |
| 21:18:26 | openstackgerrit | Matt Riedemann proposed openstack/nova stable/pike: add check before adding cpus to cpuset_reserved https://review.openstack.org/557622 | |
| 21:27:01 | openstackgerrit | Kashyap Chamarthy proposed openstack/nova master: libvirt: Allow to specify granular CPU feature flags https://review.openstack.org/534384 | |
| 21:30:14 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Log a warning and add nova-status check for old API service versions https://review.openstack.org/557506 | |
| 21:31:19 | openstackgerrit | Michael Still proposed openstack/nova master: Move create_tap_dev to the new utility method. https://review.openstack.org/557835 | |
| 21:31:19 | openstackgerrit | Michael Still proposed openstack/nova master: Move IVS VIF plugging into the new utilities module. https://review.openstack.org/557552 | |
| 21:35:00 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Log a warning and add nova-status check for old API service versions https://review.openstack.org/557506 | |
| 21:53:27 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Add nova-status check for ironic flavor migration https://review.openstack.org/527541 | |
| 21:55:42 | openstackgerrit | Eric Fried proposed openstack/nova stable/queens: Use ksa session for cinder microversion check https://review.openstack.org/557837 | |
| 21:55:51 | efried | mriedem, melwitt: queens cherry-pick ^ | |
| 21:55:58 | efried | Thanks for the prompt reviews, y'all. | |
| 21:56:12 | melwitt | lol, I was about to say, that's a lightning fast backport proposal | |
| 21:57:31 | efried | melwitt: I cheated a teeny bit | |
| 21:57:44 | efried | but I have pressure | |
| 21:59:29 | openstackgerrit | Matt Riedemann proposed openstack/nova stable/pike: Only attempt a rebuild claim for an evacuation to a new host https://review.openstack.org/550555 | |
| 22:00:37 | edleafe | efried: just for you: https://www.youtube.com/watch?v=nxxSIX3fmmo | |
| 22:00:52 | mriedem | https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:stable/pike+label:Code-Review=2 is happy, but needs a non-tony person | |
| 22:01:35 | openstackgerrit | Arvind Nadendla proposed openstack/nova master: Update ImageMetaProp object to expose traits https://review.openstack.org/557795 | |
| 22:02:32 | melwitt | cool, will find someone to bug for the rest. thanks | |
| 22:08:50 | mriedem | efried: looking at the history of that microversion check, it was added in pike, and it's used if we actually created attachments in pike, but we don't | |
| 22:08:55 | mriedem | so don't think we need to fix anything in pike there | |
| 22:09:19 | efried | mriedem: Okay, cool. Thanks for tracking that down. | |
| 22:09:49 | efried | mriedem: Wait, is that just the one in the attach flow? Cause we also do the microversion check when we're just creating the client from nothing. | |
| 22:10:04 | mriedem | is_microversion_supported isn't in pike | |
| 22:10:35 | efried | mriedem: But does the cinderclient() method have the _check_microversion call? | |
| 22:10:38 | mriedem | https://review.openstack.org/#/c/469579/3/nova/volume/cinder.py doesn't do anything unless the microversion arg is passed in | |
| 22:10:50 | mriedem | and the only things that pass that in, attachment create/update/delete, | |
| 22:10:54 | mriedem | aren't going to get used in pike | |
| 22:11:05 | mriedem | b/c we don't attach volumes the new way until queens | |
| 22:11:25 | mriedem | so there is some plumbing in pike but it's not used | |
| 22:11:28 | efried | mriedem: So... there was no reason for that code to go into pike? | |
| 22:11:40 | mriedem | well, | |
| 22:11:46 | mriedem | you're talking about a series of changes over 2 years | |
| 22:11:52 | mriedem | so we got in what we could get in when we could get it in | |
| 22:12:03 | efried | Hindsight, and all that. | |
| 22:12:16 | mriedem | my hindsight is 30/40 | |
| 22:12:30 | efried | Anyway, cool beans. I guess if it is broke, someone will yell. | |
| 22:12:37 | efried | hopefully before EOL. | |
| 22:12:42 | mriedem | there is no more eol | |
| 22:12:43 | mriedem | remember | |
| 22:12:51 | efried | I haven't really been following that. | |
| 22:12:59 | efried | But great, so we have all the time in the world :) | |
| 22:13:05 | melwitt | EOL is EOL | |
| 22:13:09 | melwitt | cacao to cacao | |
| 22:13:22 | mriedem | http://lists.openstack.org/pipermail/openstack-dev/2018-March/128862.html | |
| 22:26:31 | efried | mriedem: Is series https://review.openstack.org/#/c/526753/ approved for rocky? | |
| 22:27:16 | efried | looks like not yet, but spec is up. | |
| 22:31:39 | mriedem | spec isn't approved | |
| 22:31:50 | mriedem | oh we nacked that at the ptg | |
| 22:32:24 | mriedem | if i'm thinking of the same thing | |
| 22:44:06 | efried | TheJulia: Sorry for the delay on https://review.openstack.org/#/c/554762/ | |