| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-07-31 | |||
| 22:10:53 | dansmith | because it expects it to be gone, since the periodic on the second node should delete it | |
| 22:16:09 | dansmith | seems to fail more often than not when I run it from tox | |
| 22:16:38 | dansmith | but passes every time when I run it with subunit.run | |
| 22:16:46 | jaypipes | dansmith: have you pushed up another gibi patch? do I need to rebase? | |
| 22:16:58 | dansmith | jaypipes: yes, but it's not stable and I'm not sure why | |
| 22:20:00 | jaypipes | dansmith: that whole time.sleep(1) and manipulating the fake.set_nodes() globals is probably the culprit... | |
| 22:20:43 | melwitt | dansmith: I think subunit.run doesn't run tests in parallel but tox does (via testr underneath) | |
| 22:20:55 | jaypipes | dansmith: you could try adding a time.sleep(1) after the second self.start_service() call... | |
| 22:24:08 | mriedem | i'm not sure why the first time.sleep(1) is needed after the first service starts | |
| 22:24:52 | dansmith | melwitt: I'm running one test with tox, so should be the same | |
| 22:25:03 | melwitt | oh, one test | |
| 22:26:30 | dansmith | jaypipes: doesn't help | |
| 22:29:07 | dansmith | oh, | |
| 22:29:19 | dansmith | I was thinking he was forcing to host1 on initial boot each time | |
| 22:29:21 | dansmith | but he's not | |
| 22:29:37 | dansmith | so maybe it's just based on which it initially lands on and then moves to | |
| 22:30:02 | dfisher | nova-compute doesn't use etcd3, does it? (pike b3) | |
| 22:30:06 | dansmith | because I'm running periodics in a set order, but the actual stuff will be reversed | |
| 22:39:28 | dansmith | yeah I think that makes it repeatable | |
| 22:43:02 | mriedem | dansmith: yeah it's random | |
| 22:43:16 | mriedem | so toggle the update_rt call based on which host i guess? | |
| 22:43:20 | mriedem | dfisher: nope | |
| 22:43:29 | dansmith | mriedem: no, we need to run it both ways and make sure it behaves the same | |
| 22:43:38 | mriedem | oh | |
| 22:43:39 | dansmith | I shall have patchification soonly | |
| 22:43:46 | mriedem | cool | |
| 22:43:57 | dfisher | mriedem: i'm seeing nova service-list show my compute node but openstack hypervisor list not show it. | |
| 22:44:13 | mriedem | dfisher: is it mapped to a cell? | |
| 22:44:22 | mriedem | nova-manage cell_v2 discover_hosts | |
| 22:44:30 | mriedem | --verbose | |
| 22:44:30 | dfisher | no, it's not mapped. | |
| 22:44:38 | dfisher | how would I map it | |
| 22:44:42 | dfisher | ? | |
| 22:46:01 | mriedem | dfisher: ^ | |
| 22:46:04 | mriedem | discover_hosts | |
| 22:46:29 | dfisher | yeah, that's not finding it :( | |
| 22:46:48 | dfisher | http://paste.openstack.org/show/617066/ | |
| 22:48:25 | dansmith | dfisher: then your compute isn't checking into the cell1 db | |
| 22:48:52 | dansmith | dfisher: make sure your conductor's config is pointed at the cell1 db, matching the cell1 cell_mapping record | |
| 22:49:08 | dfisher | ok. will poke. thanks dansmith! | |
| 22:52:26 | openstackgerrit | Dan Smith proposed openstack/nova master: Test resize with placement api https://review.openstack.org/487958 | |
| 22:52:32 | dansmith | mriedem: ^ see if you hate that | |
| 22:52:40 | dansmith | should be repeatable, and keep jaypipes honest :) | |
| 22:57:17 | mriedem | dansmith: questions inline about the az stuff | |
| 22:59:25 | dansmith | mriedem: comments inline | |
| 23:00:51 | mriedem | oh right forced_host | |
| 23:00:58 | mriedem | which there used to be a docs page for that, | |
| 23:01:02 | mriedem | but with the migration it looks like it's gone | |
| 23:01:16 | mriedem | and our api-ref doesn't mention this wrinkle of course | |
| 23:03:19 | dansmith | I admit, I had to copy some deets out of an ask.o.o article :P | |
| 23:03:42 | mriedem | there used to be a nice docs page called "Select hosts where instances are launched" | |
| 23:03:46 | mriedem | and it had this information in it | |
| 23:03:51 | mriedem | but apparently it's been deleted | |
| 23:05:05 | mriedem | alright i'll check results after i get back from maya's gymnastics class, which is full of fun and surprises | |
| 23:06:33 | dansmith | ack | |
| 23:27:04 | mikal | Does anyone here understand what causes hairpins to fail to enable? | |
| 23:27:14 | mikal | I'm trying to unravel that code to be less ... processy | |
| 23:35:18 | jaypipes | mikal: sorry, no :( | |
| 23:36:00 | jaypipes | dansmith: sorry, was dinnering. what change did you make to make that test stable? | |
| 23:36:21 | dansmith | jaypipes: made sure to boot the instance on a specific node consistently | |
| 23:36:33 | dansmith | jaypipes: since the allocations are screwed up by one running the periodic before the other, | |
| 23:36:52 | dansmith | the order of boot, migration, and which node runs the periodic last affect the outcome | |
| 23:37:18 | dansmith | jaypipes: so now it runs each test twice, starting and finishing on a different node each time | |
| 23:37:25 | dansmith | since it always runs the periodic in the same order, | |
| 23:37:38 | dansmith | that will cover us against ordering issues if you pass all four tests | |
| 23:37:51 | dansmith | and mriedem is going to work on a single-node variant I think | |
| 23:43:07 | jaypipes | dansmith: k | |
| 23:43:15 | jaypipes | dansmith: so I'm good to pull and rebase? | |
| 23:44:01 | mikal | jaypipes: its ok, this is all nova-net code and might go to heaven soon anyways | |
| 23:44:14 | jaypipes | mikal: heaven? | |
| 23:44:25 | dansmith | jaypipes: cha | |
| 23:44:41 | mikal | jaypipes: would you prefer "is put out to pasture"? | |
| 23:44:42 | dansmith | jaypipes: remember things are upside down for mikal | |
| 23:44:51 | jaypipes | heh | |
| 23:44:54 | mikal | "kicks the bucket" | |
| 23:44:59 | mikal | "goes for a dirt nap" | |
| 23:46:54 | dansmith | jaypipes: oh looks like there is a pep8 error in that test patch, maybe you can fix when you rebase? | |
| 23:47:02 | jaypipes | dansmith: yuppers. | |
| 23:47:22 | dansmith | jaypipes: and note that two of those tests are self.skipTest()ed so you'll want to unskip them as soon as you can | |
| 23:47:31 | jaypipes | yup, got it. | |
| #openstack-nova - 2017-08-01 | |||
| 00:04:29 | jaypipes | dansmith: the reason it's failing is because you're forgetting to run that _run_perioidics() after calling delete on the server. | |
| 00:05:03 | jaypipes | dansmith: on the revert one. | |
| 00:05:32 | dansmith | jaypipes: the reason what is failing? | |
| 00:06:00 | dansmith | tests pass locally for me, and delete should clean up all the allocations without a periodic run, no? | |
| 00:06:20 | jaypipes | nope. | |
| 00:06:32 | dansmith | um why? | |
| 00:06:35 | jaypipes | dansmith: delete doesn't clean up placement at all. | |
| 00:06:46 | dansmith | delete doesn't call self.update() or whatever in compute manager? | |
| 00:06:58 | dansmith | that would mean you can't boot instances on a compute node until after the periodic runs to clean up | |
| 00:06:58 | jaypipes | update is only inventory. | |
| 00:07:53 | jaypipes | dansmith: welcome to my hell. | |
| 00:08:01 | dansmith | no, I'm serious | |
| 00:08:21 | dansmith | that makes zero sense to me.. surely we'd have people beating down our door if deleting an instance didn't free up space for another one | |
| 00:08:44 | dansmith | jaypipes: but again, the tests pass for me, so what are you saying is failing? | |
| 00:09:09 | jaypipes | dansmith: nm, I'm wrong. | |
| 00:09:27 | jaypipes | I think... | |
| 00:09:28 | dansmith | self._update_resource_tracker() is in _complete_deletion() | |
| 00:09:32 | dansmith | surely that updates things | |
| 00:09:44 | dansmith | yeah, calls update_usage() | |
| 00:09:44 | jaypipes | only if the uuid is in self.tracked_instances. | |