Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-03-16
15:24:31 superdan johnthetubaguy: I can do it, I usually do
15:24:58 cdent fried_rice: I can see how it useful, so happy to do it, since I need to go in there anyway for something else. But because of the amount of searching I do to create the weekly report, I tend not to be very topic oriented (despite using the urls all the time), so it's not something that immediately leaps to mind for me
15:25:01 superdan mriedem: ^
15:25:27 fried_rice cdent: Gotcha.
15:25:36 mriedem i don't think we've merged any db migrations in rocky yet, but checking
15:26:07 mriedem nope we're good
15:26:18 superdan mriedem: will do it after this call
15:28:40 johnthetubaguy superdan: sweet thanks, just spotted a change doing a migration, and it made me stop and think
15:28:52 superdan johnthetubaguy: yeah, thanks
15:29:56 openstackgerrit Jay Pipes proposed openstack/nova-specs master: mirror nova host aggregates to placement API https://review.openstack.org/545057
15:30:24 leakypipes superdan, mriedem, bauzas, gibi, alex_xu_, figleaf: ^^ reworked mirroring spec.
15:32:51 kashyap mriedem: Was looking up the scroll this morning; about increasing timeouts due to QEMU / libvirt updates
15:32:57 kashyap https://review.openstack.org/#/c/547676/
15:33:18 kashyap Even clarkb (not on this channel now) also reported at PTG having to increase tests' timeouts
15:33:54 kashyap I think it must probably be the QEMU / libvirt slowdown after the Meltdown / Spectre fixes. /me goes to check w/ the libvirt folks
15:36:39 openstackgerrit Surya Seetharaman proposed openstack/nova stable/pike: Unmap compute nodes when deleting host mappings in delete cell operation https://review.openstack.org/553829
15:37:46 openstackgerrit Surya Seetharaman proposed openstack/nova stable/pike: Unmap compute nodes when deleting host mappings in delete cell operation https://review.openstack.org/553829
15:41:27 mriedem kashyap: yes mnaser already confirmed that
15:41:51 kashyap mriedem: I struggle to believe the libvirtd dameon _itself_ taking 1 _whole_ minute!
15:42:00 kashyap mriedem: I'm checking w/ the Red Hat perf team to see if they have any numbers on this
15:42:02 mriedem libvirt/qemu in the ocata UCA got updates and has to great a guest on first start to scan cpu features, which is not something newer libvirt in the pike UCA has to do
15:42:13 mriedem *create
15:42:22 mriedem https://launchpad.net/~ubuntu-cloud-archive/+archive/ubuntu/ocata-staging/+sourcepub/8774271/+listing-archive-extra
15:42:24 kashyap I know, they ship with new CPU models, all ending with "*-IBRS"
15:42:45 openstackgerrit Surya Seetharaman proposed openstack/nova stable/pike: Unmap compute nodes when deleting host mappings in delete cell operation https://review.openstack.org/553829
15:42:59 kashyap (Yep, very familiar w/ those updates for a couple of months already. As I had to apply to a bunch of Fedora / RHEL machines)
15:45:02 kashyap This additional slow down even bolsters the case for the couple of Cloud Providers to enable nested virt in the Gate.
15:45:23 kashyap Upstream KVM maintainer said he's going to enable it for Intel by default in the upstream kernel.
15:46:19 kashyap ClarkB, myself and a bunch of others participated in a discussion with the Cloud Providers at PTG, some cloud providers (OVH) seemed amenable to that.
15:46:24 kashyap Let's see...
15:47:42 fried_rice figleaf: Reviewed. Nothing blocking :)
15:48:17 fried_rice figleaf: Are you still -1 on https://review.openstack.org/#/c/553476/ for the name of that article, having read my excellent excuses?
15:48:54 mriedem volume test issues in the gate should be flushed once https://review.openstack.org/#/c/553794/ merges
15:49:21 jgwentworth woot, thanks
15:49:25 openstackgerrit Dan Smith proposed openstack/nova master: Add placeholder migrations for Queens backports https://review.openstack.org/553831
15:49:27 superdan mriedem: johnthetubaguy ^
15:49:56 mriedem superdan: aren't there some tests that you need to modify for the placeholder range?
15:50:07 superdan mriedem: um, no, jeez
15:50:08 superdan (yes)
15:51:03 cfriesen kashyap: for what it's worth, I haven't had good experiences with stability of nested KVM with CentOS 7.3.
15:51:21 figleaf fried_rice: yes, 'upt' is much too cryptic
15:51:23 kashyap cfriesen: Got a link to the bugs you've filed? :-)
15:51:41 kashyap cfriesen: More seriously, some Cloud Providers in the room themselves said it was stable in their environments
15:51:52 fried_rice figleaf: Having noticed your # noqa hack, I guess I can make it work with a longer name, eh?
15:51:58 cfriesen kashyap: heh...nothing useful to report, just random intermittent crashes of the first-level guest.
15:52:01 kashyap cfriesen: The cardinal rule here is to try with latest CentOS (stable 7.4)
15:52:20 kashyap cfriesen: I used to track all the bugs I reported for nested virt upstream, a year or so ago: https://kashyapc.fedorapeople.org/kernel-kvm-bugs.txt
15:52:26 kashyap All of them fixed now
15:52:41 cfriesen cool, will have to give it a try with 7.4
15:53:08 kashyap cfriesen: But, one of the blocking points is the migration of the L1 guest (with the mere presence of L2 in it; it doesn't even have to run). And upstream is working on it as we speak
15:53:11 fried_rice mriedem: What am I tracking for these volume failures to be resolved before I clog the gate with more rechecks?
15:53:18 kashyap cfriesen: s/one of the/the only/
15:53:37 cfriesen kashyap: live migration I assume?
15:53:53 fried_rice mriedem: https://review.openstack.org/#/c/553794/ ?
15:53:59 kashyap cfriesen: Yep, live migrating L1 (while L2 is in it; online or offline)
15:54:32 figleaf fried_rice: you don't have to make it *that* long!
15:55:03 fried_rice figleaf: update-provider-tree will push me over 80. Anything else doesn't really make sense. upd-prov-tree? Ew.
15:55:33 mriedem fried_rice: yes, as noted above
15:55:39 mriedem "(10:48:59 AM) mriedem: volume test issues in the gate should be flushed once https://review.openstack.org/#/c/553794/ merges"
15:55:44 fried_rice mriedem: oh, see it now, thx
15:56:46 openstackgerrit Dan Smith proposed openstack/nova master: Add placeholder migrations for Queens backports https://review.openstack.org/553831
15:57:36 fried_rice figleaf: Looks like your gabbi has some failures?
15:58:08 fried_rice figleaf: Oh, because now you have to send a dict-with-generation to set aggregates :)
15:58:43 figleaf fried_rice: saw that. Working on it
16:00:17 fried_rice figleaf: Added helpful link
16:00:30 fried_rice (not that you needed it)
16:06:45 openstackgerrit Eric Fried proposed openstack/nova master: update_provider_tree devref and docstring updates https://review.openstack.org/553476
16:06:56 fried_rice figleaf: Done ^
16:10:44 superdan fried_rice: okay, so I get that you're hitting your peak on review efficiency and finding issues,
16:10:56 superdan fried_rice: but that should fall short of finding issues in things I need to get merged
16:11:18 fried_rice superdan: Send me that list. See where it gets ya.
16:11:24 superdan heh
16:11:40 fried_rice superdan: Are you talking about the member_of thing?
16:11:44 superdan yeah :)
16:11:48 fried_rice zuul had me beat on that one.
16:12:00 fried_rice I'll have to try harder.
16:12:07 superdan that was already approved, but failed the gate, now -1d
16:12:17 superdan I'm not serious, of course, I'm glad you found stuff
16:12:32 fried_rice yeah, the rebase needed to take into account the functionality of the thing it got rebased against.
16:12:46 fried_rice TBH, I shoulda seen that on my first pass.
16:12:48 superdan is this the microversion you two were racing for?
16:13:47 superdan stackalytics should discount reviews where you're clearly taking shots at someone else to win a version race :P
16:15:18 cdent reviewers should discount stackalytics
16:15:46 superdan s/reviewers/everyone/
16:16:45 cdent superdan: I was trying to (quickly) come up with a regexp on your statement but I got lazy and instead printed the output. (meaning, yes, I agree)
16:17:44 superdan heh
16:18:37 figleaf superdan: it was losing the microversion race that made the member_of patch fail. For once, can't blame fried_rice
16:18:48 superdan figleaf: ah
16:19:41 superdan 1.21 is a better microversion anyway
16:19:49 fried_rice It can drink
16:20:03 fried_rice (wheat, unfiltered IIRC)
16:32:28 openstackgerrit Ed Leafe proposed openstack/nova master: Add 'member_of' param to GET /allocation_candidates https://review.openstack.org/552098
16:32:36 figleaf leakypipes: superdan: fried_rice: ^^ func tests fixed
16:33:04 figleaf Had to resort to the gabbi HISTORY stuff, since I couldn't assume generation=0
16:33:22 figleaf Besides, we're not supposed to assume that, right? :)
16:48:33 fried_rice figleaf: Ideally you should get it from the RP creation. I did that in a couple of places in my other patch. It's a PITA in gabbi. (Not such a problem in real code.)
16:48:51 fried_rice FWIW, I wouldn't -1 hardcoded generation numbers in a gabbi test.
16:49:03 fried_rice and would even write them myself :)
16:51:11 cdent fried_rice: it's not a pita in gabbi, just do what figleaf has done
16:52:12 superdan lyaaaaaaaaaaaaar: jgwentworth: looks like those reverts are failing some tests.. on two of the three patches at least

Earlier   Later