| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-03-20 | |||
| 14:31:40 | efried | So it turns out that assert_called_once and assert_not_called aren't actually methods on a mock. Without speccing, they'll succeed happily just like any nonexistent method would. There's 80ish instances of these in the nova tree. May I open a bug and fix this, or is there something I don't know? | |
| 14:31:41 | dansmith | I don't even see some debug logs that we should be seeing in scheduler | |
| 14:31:59 | jroll | the only thing I've noticed so far is that the flavor isn't an exact match to the node | |
| 14:32:01 | jroll | oh, hm | |
| 14:32:03 | dansmith | like we should see this: https://github.com/openstack/nova/blob/master/nova/scheduler/host_manager.py#L590-L591 | |
| 14:32:12 | dansmith | but unless I just suck at searching I can't find it | |
| 14:32:34 | jroll | agree | |
| 14:33:11 | mriedem | efried: are you looking at std lib or the mock library? | |
| 14:33:36 | efried | mriedem How do I know which is which? | |
| 14:33:40 | mriedem | efried: b/c they are in the lib https://github.com/testing-cabal/mock/blob/2.0.0/mock/mock.py#L890 | |
| 14:33:41 | mriedem | and we use the lib | |
| 14:33:44 | mriedem | in the py27 job | |
| 14:34:00 | mriedem | https://github.com/openstack/requirements/blob/master/upper-constraints.txt#L149 | |
| 14:34:19 | efried | What I do know is that the tests I found this out on are using 'import mock'. | |
| 14:34:23 | jroll | dansmith: I wonder how relevant this is compute_manager=ironic.nova.compute.manager.ClusteredComputeManager | |
| 14:34:27 | jroll | that should not be there | |
| 14:34:33 | dansmith | jroll: w..t..f | |
| 14:34:35 | efried | So I guess that behaves differently depending on whether the mock library is installed or not? | |
| 14:34:43 | jroll | dansmith: wait, I think that's not even in ironic tree, wat | |
| 14:34:47 | efried | ...the import, that is? | |
| 14:35:03 | dansmith | jroll: that would be majorly problematic for a lot of things we've just done | |
| 14:35:11 | jroll | dansmith: indeed | |
| 14:35:26 | jroll | dansmith: confirmed, that module doesn't exist in ironic master | |
| 14:35:37 | dansmith | ummm.... | |
| 14:35:41 | mriedem | you can't load up out of tree compute managers anymore | |
| 14:35:50 | mriedem | we dropped the ability to do that awhile ago, like newton i think | |
| 14:35:51 | jroll | mriedem: so that config is a noop yeah? | |
| 14:35:57 | mriedem | most likely | |
| 14:36:00 | jroll | ok | |
| 14:36:03 | dansmith | oh, compute manager, I was thinking host manager | |
| 14:36:17 | jroll | efried: we always use the third party lib, which is api-compatible with the 3.6 version of stdlib | |
| 14:36:39 | mriedem | http://codesearch.openstack.org/?q=ironic.nova.compute.manager.ClusteredComputeManager&i=nope&files=&repos= | |
| 14:36:41 | efried | jroll Okay, must be something in my local env that's using the stdlib instead, then. | |
| 14:36:44 | openstackgerrit | Sean Dague proposed openstack/nova master: update i18n guide for nova https://review.openstack.org/447528 | |
| 14:37:07 | dansmith | jroll: is that the manager that used to always report the same hostname to allow for "HA" ? | |
| 14:37:16 | jroll | dansmith: yeah | |
| 14:38:09 | dansmith | EmilienM: so that should be removed from tripleo, especially since it'll be removed at some point | |
| 14:39:36 | jroll | dansmith: it's lone gone already | |
| 14:39:55 | dansmith | jroll: they're still setting it though yes? | |
| 14:40:11 | jroll | yeah, it should be removed, but clearly not breaking things | |
| 14:40:21 | jroll | "clearly" ehh | |
| 14:40:25 | mriedem | https://github.com/openstack/nova/commit/d9a7c16c305cd7016195cada883867411d3a5eb9 | |
| 14:40:26 | jroll | you get what I mean | |
| 14:40:30 | dansmith | ah, yeah, I figured the CONF was still there to avoid breaking people with it set | |
| 14:40:31 | mriedem | i guess we only removed the compute manager config in ocata | |
| 14:40:42 | mriedem | we don't fail on busted config | |
| 14:40:48 | dansmith | apparently not | |
| 14:40:51 | mriedem | i think there is a flag to set if you have busted config, but we don't run with it | |
| 14:41:00 | dansmith | okay, I know I've seen it before | |
| 14:41:21 | jroll | thanks for confirmation mriedem | |
| 14:41:48 | dansmith | well, I dunno why we're not seeing some of the other messages that should be in scheduler | |
| 14:42:08 | dansmith | EmilienM: how can I see the exact git hash of nova that is being tested here? | |
| 14:42:24 | mriedem | is the IronicHostManager overriding the methods that have the logging you expect to ese? | |
| 14:42:26 | mriedem | *see | |
| 14:42:44 | dansmith | I don't think so, but let me look | |
| 14:42:50 | jroll | shouldn't be | |
| 14:42:59 | dansmith | yeah, no | |
| 14:43:58 | dansmith | I think the nova hash is 47d90a3 | |
| 14:44:15 | dansmith | in this run http://logs.openstack.org/periodic/periodic-tripleo-ci-centos-7-ovb-nonha/12d034e/logs/rpm-qa.txt.gz | |
| 14:44:34 | EmilienM | dansmith: yes that | |
| 14:45:18 | dansmith | so actually, if we get nothing back from placement, we'd not hit the messages I'm looking for, I think | |
| 14:45:19 | dansmith | so it must be that | |
| 14:45:20 | dansmith | but I have no idea why | |
| 14:45:27 | EmilienM | ok so I need to remove ironic.nova.compute.manager.ClusteredComputeManager | |
| 14:45:39 | dansmith | EmilienM: yeah, that's just a dead config opt | |
| 14:45:48 | EmilienM | replaced by what? | |
| 14:45:56 | dansmith | not replaced by anything | |
| 14:45:58 | mriedem | EmilienM: https://github.com/openstack/nova/commit/d9a7c16c305cd7016195cada883867411d3a5eb9 | |
| 14:46:28 | EmilienM | ok | |
| 14:46:35 | jroll | replaced by AWESOME | |
| 14:47:06 | EmilienM | do you think removing the option would help tripleo issue? | |
| 14:47:07 | jroll | EmilienM: the need for that was replaced by https://github.com/openstack/nova/commit/6047d790a32ef5a65d4d6b029f673ce53c3d4141 | |
| 14:47:10 | jroll | no | |
| 14:47:11 | EmilienM | or just cleanup | |
| 14:47:17 | jroll | just cleanup | |
| 14:47:18 | EmilienM | k | |
| 14:47:56 | EmilienM | "There's no configuration to enable this mode; it's always running." | |
| 14:48:11 | EmilienM | jroll: ❤ | |
| 14:48:33 | jroll | :) | |
| 14:48:57 | mriedem | seems wasteful for FilterScheduler._schedule to even call HostManager.get_filtered_hosts if we already know we don't have any hosts (resource providers from placement) | |
| 14:49:32 | openstackgerrit | Gábor Antal proposed openstack/nova master: Add json style checking for sample notifications https://review.openstack.org/443677 | |
| 14:49:35 | mriedem | but maybe the spaghetti makes sense someohw | |
| 14:49:37 | dansmith | mriedem: right and we don't | |
| 14:50:02 | dansmith | mriedem: https://github.com/openstack/nova/blob/master/nova/scheduler/filter_scheduler.py#L175-L176 | |
| 14:50:03 | openstack | Launchpad bug 1674375 in puppet-nova "compute_manager is deprecated, we should stop using it" [High,Triaged] - Assigned to Emilien Macchi (emilienm) | |
| 14:50:03 | EmilienM | jroll: /me taking care of this cleanup https://bugs.launchpad.net/puppet-nova/+bug/1674375 | |
| 14:50:18 | mriedem | dansmith: i'm talking about this https://github.com/openstack/nova/blob/master/nova/scheduler/filter_scheduler.py#L108 | |
| 14:50:21 | jroll | EmilienM: nice, thanks! s/deprecated/removed/ btw :) | |
| 14:50:23 | mriedem | hosts is [] as you just poined out | |
| 14:50:25 | mriedem | *pointed | |
| 14:50:38 | mriedem | but we still go through self.host_manager.get_filtered_hosts and all of it's checks on an empty list | |
| 14:50:43 | mriedem | which seems like a waste of cycles | |
| 14:50:53 | openstackgerrit | Silvan Kaiser proposed openstack/nova master: Removes getfattr from Quobyte Nova driver https://review.openstack.org/428646 | |
| 14:50:54 | dansmith | mriedem: ah, well, it's empty, but yeah I see what you mean | |
| 14:50:56 | mriedem | unrelated to this problem | |
| 14:50:57 | dansmith | not very impactful | |
| 14:50:59 | dansmith | yeah | |
| 14:51:10 | mriedem | but i thought we were always paranoid about scheduler perf | |
| 14:51:18 | mriedem | like, no logging | |
| 14:51:40 | dansmith | that will fail after the first filter, though, | |