Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-11-21
15:34:15 gary_perkins hmmm.. good point, will check
15:34:55 mriedem fwiw, this is a functional test for exactly what you're trying to see https://github.com/openstack/nova/blob/d74f9d62c5c320539ff5a46f28b0104c3b236633/nova/tests/functional/wsgi/test_servers.py#L385
15:36:53 mriedem mnaser was trying to switch the default on that rule in stein but it turns out the flavors used in devstack, which tempest uses in our CI, have disk=0 so they explode :) https://review.openstack.org/#/c/603910/
15:36:53 gary_perkins mriedem: thanks! and no block_device_mapping_v2 property
15:37:08 gary_perkins ha!
15:37:22 mnaser mriedem: ah sorry i never actually got around cleaning that one up
15:37:34 mriedem mnaser: it's going to be non-trivial
15:37:44 mriedem gary_perkins: which policy file did you modify?
15:38:01 mriedem /etc/nova/policy.json?
15:38:17 jangutter something that still shows how much of a noob I am: regarding blueprint and specs. Is there a specific order in which things get approved normally? spec -> blueprint? Or is it not ordered?
15:38:21 mriedem make sure whatever policy file you modified is the one this option points at https://docs.openstack.org/nova/queens/configuration/config.html#oslo_policy.policy_file
15:38:26 mnaser mriedem: because now in tempest we need to use a flavor that is non root_gb=0 first before that can ever work, right?
15:38:34 mriedem jangutter: spec and then blueprint
15:38:52 mriedem mnaser: yeah, and tempest is configured with flavors that devstack sets up in our CI
15:39:17 mriedem https://github.com/openstack-dev/devstack/blob/master/lib/tempest#L191
15:39:58 jangutter mriedem: thanks!
15:40:02 mriedem tempest is also configured with the cirros images https://github.com/openstack-dev/devstack/blob/master/lib/tempest#L326
15:40:15 mriedem so maybe we just need to get the image size from glance and use that for the flavor disk size
15:40:51 mnaser mriedem: will we still be able to get coverage for bfv with root_gb=0 then?
15:41:15 mnaser or that might involve probably adding tempest tests
15:41:17 mriedem not unless you add a specific test to tempest for that
15:41:24 mriedem we have functional tests in nova for it
15:41:31 mriedem but those don't use a real virt driver
15:42:12 mriedem yay min_disk=0 on the image http://logs.openstack.org/34/615134/2/check/tempest-full/f6d351a/controller/logs/devstacklog.txt.gz#_2018-11-02_13_20_26_698
15:42:16 mriedem that's helpful
15:43:04 gary_perkins mriedem: yes, /etc/nova/policy.json originally copied from https://docs.openstack.org/nova/queens/_downloads/nova.policy.yaml.sample
15:43:45 mriedem gary_perkins: but what is [oslo_policy]/policy_file in nova.conf in the nova-api service pointing at?
15:43:57 mriedem b/c you said policy.json but you linked to a policy.yaml sample
15:45:09 gary_perkins mriedem: oh, sorry. Just trying to find my original sample. That's obviously the wrong one
15:46:42 gary_perkins mriedem: nova.conf: policy_file = policy.json
15:47:42 mriedem hmm, and you shouldn't have to restart nova-api after modifying the policy file, it should be read fresh each time
15:49:23 gary_perkins and ofc, policy.json has: "os_compute_api:servers:create:zero_disk_flavor": "rule:admin_api"
15:49:39 gary_perkins amongst may other policies!
16:17:13 mriedem gary_perkins: i'm not sure. have you tried tweaking any other policy rules to make sure it's not something with the configuration of the policy file?
16:45:47 xek gibi, there are no versioned network.floating_ip.associate network.floating_ip.disassociate notifications?
16:46:00 artom mriedem, hey, so, I'm being told from people who were at summit that you were in the room at the NFV pain points session (https://etherpad.openstack.org/p/BER-nfv-hpc-pain-points). Now, to me, that sounds like nothing could make you happier than reviewing https://review.openstack.org/#/c/599587/, right ;)
16:46:21 gibi xek: is there an unversioned one from nova about those?
16:47:43 gibi xek: OK I see there is unversioned ones for those
16:47:53 xek gibi, I see they are emitted in nova.network.floating_ips, but I'm not sure if there is an alternative somewhere else? maybe in neutron?
16:48:22 mriedem artom: lots of things could make me happier
16:48:39 artom mriedem, what if I promised ice cream?
16:48:52 mriedem xek: gibi: i don't think we'd port that to versioned notifications since they are nova-network specific and that's deprecated
16:49:05 mriedem artom: i have ice cream like 30 feet away
16:49:11 artom Foiled again!
16:49:39 artom mriedem, in seriousness, I know y'all are overloaded, I'm just trying to get it over the line, so if you have energy/bandwidth, it'd be appreciated
16:50:59 gibi xek, mriedem: yeah, I just found that it is only emitted from the nova network code patch, this is why they are not on the todo list for versioned transformation
16:51:44 gibi xek: I don't know if neturon has a similar notification mechanism
16:52:25 xek gibi, mriedem, it makes sense, thanks for clarifying that.
16:52:30 gibi mlavalle: ^^ do you happen to know if neutron emits notification about floating ips?
17:05:48 mlavalle gibi: what kind of notification are we talking about? the code sends notifications like this: https://github.com/openstack/neutron/blob/master/neutron/db/l3_db.py#L1408
17:06:18 mlavalle These are internal notifications. modules subscribe to these notifications to listen to events and process them
17:08:11 mriedem mlavalle: yes same thing
17:08:15 mriedem over rpc
17:14:37 mriedem mnaser: let's see what explodes https://review.openstack.org/619319
17:15:55 mnaser mriedem: seems reasonable. i'll have to fix the functional tests soon
17:15:58 mnaser ugh the logs have been rotated out already
17:16:10 mriedem yeah one thing at a time
17:16:12 mriedem not urgent
17:44:57 openstackgerrit Matt Riedemann proposed openstack/nova master: Remove NovaException logging from scatter_gather_cells https://review.openstack.org/619110
18:17:53 dansmith I think Greg's windows box has a virus
18:31:48 openstackgerrit Elod Illes proposed openstack/nova master: Transform scheduler.select_destinations notification https://review.openstack.org/508506
18:39:46 jangutter dansmith: please take this as a late night bad joke: isn't that a redundant statement?
18:40:20 dansmith jangutter: yeah, maybe I should say "Greg's windows box has a virus and is proud enough to show off in public"
18:45:28 openstackgerrit Merged openstack/nova stable/queens: De-dupe subnet IDs when calling neutron /subnets API https://review.openstack.org/616551
19:10:17 openstackgerrit Matt Riedemann proposed openstack/nova master: libvirt: implement reshaper for vgpu https://review.openstack.org/599208
19:14:55 openstackgerrit Elod Illes proposed openstack/nova master: Transform scheduler.select_destinations notification https://review.openstack.org/508506
19:15:14 openstackgerrit Merged openstack/nova stable/rocky: Add recreate test for bug 1799892 https://review.openstack.org/619075
19:15:14 openstack bug 1799892 in OpenStack Compute (nova) rocky "Placement API crashes with 500s in Rocky upgrade with downed compute nodes" [Medium,In progress] https://launchpad.net/bugs/1799892 - Assigned to Matt Riedemann (mriedem)
19:24:00 mriedem easy bug fix https://review.openstack.org/#/c/615134/
19:46:32 mriedem anyone have any bright ideas on how to assert a semaphore context manager is called in unit tests? https://review.openstack.org/#/c/609180/7/nova/tests/unit/virt/libvirt/test_driver.py
19:51:56 sean-k-mooney mriedem: as in assert the semaphore was aquired on a partaclar code path?
19:52:00 sean-k-mooney just looking
19:57:31 mriedem http://paste.openstack.org/show/735916/
19:58:57 sean-k-mooney mriedem: do you want sema.__enter__.assert_called_once() or just assert it is called
19:59:56 mriedem does'nt matter
20:00:09 mriedem point is, i don't really want to add a bunch of mocks all over test code and never assert it's used
20:00:23 sean-k-mooney ah ok
20:02:40 sean-k-mooney im not sure that does what you want
20:03:27 sean-k-mooney you are asserting that the __enter__ metod of the sema context manager is called but will that not always be called by the with statement
20:03:48 mriedem if i remove the code change for that method, the test fails
20:03:53 mriedem which is what i'd expect
20:04:51 sean-k-mooney hum ok this is one of those things in python that i always have to trow in an interpreter to prove to myself it scorrect
20:06:16 mriedem i also don't know why we have nova.privsep.qemu.convert_image but not everything uses it, there are a few places in here that are calling qemu-img convert outside of privsep
20:06:43 mriedem dansmith: do you know anything about that? ^
20:06:53 mriedem b/c i thought we declared mission accomplished on privsep stuff
20:07:10 sean-k-mooney mriedem: are the fucntion that call it out side of privsep called from function with the prive sep decorator applied
20:08:00 mriedem umm, no
20:08:12 mriedem grep _disk_raw_to_qcow2 in the libvirt driver
20:08:31 mriedem it's using processutils from oslo
20:09:04 mriedem and we're passing through a rootwrap helper
20:10:09 dansmith mriedem: not everything like that needs to run privileged
20:10:13 sean-k-mooney ttp://git.openstack.org/cgit/openstack/nova/tree/nova/virt/libvirt/driver.py#n8361 is using oslo but i dont see a rootwarap helper
20:10:30 dansmith mriedem: might be that we need image convert sometimes when reading from an lvm or something like that, I dunno
20:11:15 mriedem ok i'll pretend this never happene
20:11:20 dansmith good plan
20:11:25 sean-k-mooney assuimg the nova user has the correct permisiosn to read and write the vm images it shoudl be able to execute that command with out elevating
20:11:34 mriedem jackding: so on that io concurrency change,
20:11:45 mriedem i would just like to see that for the functional code paths that add the context manager,
20:11:53 mriedem that at least one corresponding unit test asserts the context manager is called
20:12:01 mriedem like i noted for _disk_raw_to_qcow2

Earlier   Later