Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-02-15
21:58:02 mriedem it should just be the =None thing for the kwargs
21:58:14 mriedem and then probably a unit test to show that none of them are required except cell_uuid
22:00:16 coreycb mriedem, i think we can drop the 'if not' checks but should still pick up transport_url and db_connection from nova.conf if not passed as arguments
22:01:07 mriedem coreycb: why?
22:01:19 mriedem so that you don't have to use the command line?
22:01:33 coreycb mriedem, yes which is how create_cell works
22:01:51 mriedem dansmith: does object save() noop if the field values didn't actually change?
22:06:12 melwitt mriedem: I think all save() cares about is whether the fields on the object are dirty, if they are, it will save them
22:07:29 dansmith mriedem: depends on the implementation
22:07:40 dansmith mriedem: we used to try to make sure to not do anything if there were no changes,
22:07:52 dansmith but some people thought a save() should always have consistent refresh semantics
22:08:06 openstackgerrit Corey Bryant proposed openstack/nova master: Enable defaults for cell_v2 update_cell command https://review.openstack.org/434533
22:08:09 dansmith so in the end I think we have no particular standard you can assume
22:09:09 mriedem coreycb: i left a comment in ps1 about updating the man page
22:09:19 coreycb mriedem, ok
22:09:44 mriedem otherwise it looks good
22:17:57 openstackgerrit Corey Bryant proposed openstack/nova master: Enable defaults for cell_v2 update_cell command https://review.openstack.org/434533
22:18:14 coreycb mriedem, ok all set
22:21:03 mriedem lgtm
22:21:13 mriedem dansmith: melwitt: you guys should probably sign off
22:21:17 mriedem on ^ i mean
22:21:29 dansmith coreycb: yeah, sorry for missing that, I wrote that patch in a hurry for someone
22:22:09 coreycb dansmith, np at least that was an easy fix
22:22:21 dansmith coreycb: and now you feel like you were *partially* right, right? :P
22:22:42 coreycb dansmith, lol yes. redemption! :)
22:22:52 dansmith coreycb: but regardless, glad to see someone else poking at this stuff
22:56:10 mriedem coreycb: your unit test failed
22:56:28 mriedem http://logs.openstack.org/33/434533/3/check/gate-nova-python27-ubuntu-xenial/9e0c2cc/console.html#_2017-02-15_22_26_03_881971
22:56:49 mriedem b/c of the config option thing
22:56:55 mriedem testtools.matchers._impl.MismatchError: 'fake:///db' != u'sqlite://'
22:57:20 mriedem i can fix it quick
23:00:49 openstackgerrit Alex Schultz proposed openstack/nova master: Add pxe support for libvirt https://review.openstack.org/434549
23:02:44 openstackgerrit Matt Riedemann proposed openstack/nova master: Enable defaults for cell_v2 update_cell command https://review.openstack.org/434533
23:09:35 dansmith melwitt: I'm having unclean thoughts about ways to fool the code after an rpc call into always being to the main database for our functional tests
23:11:21 melwitt dansmith: why? what problem running into?
23:11:47 dansmith melwitt: more cells-aware api calls and hitting the recursive lock thing
23:13:33 melwitt dansmith: that reminds me, I had meant to mention this to you but while I was working on the counting quotas stuff, I found I needed to add something to the CellDatabases fixture like this https://review.openstack.org/#/c/416521/13/nova/tests/fixtures.py@398
23:13:47 melwitt I dunno if that will help at all with what you're dealing with
23:14:10 dansmith melwitt: heh, I have that same change locally ;)
23:14:14 dansmith melwitt: problem is, that doesn't make it over fakerpc, because it's not serialized
23:14:20 melwitt ah, dangit
23:14:56 dansmith melwitt: which is fine for real, but not here.. so my unclean thoughts were around hacking the serialization via mock and jamming a flag in there we could use to just say "this should always be main_context_manager"
23:14:57 dansmith or something
23:15:18 dansmith I guess just stashing the connection url would be better or something
23:15:27 dansmith but still, need a serialization hack
23:15:57 melwitt dansmith: what do you mean by "after an rpc call"? like after the rpc, you're left with the wrong db url?
23:17:16 dansmith melwitt: after an rpc call you have context.db_connection=None
23:18:03 melwitt that should fall back to main, right
23:18:23 dansmith it does, but it means we call get_context with that =None, which means we take the lock,
23:18:33 dansmith which we have to do because we need to use whatever target_cell has set global state to
23:19:35 dansmith oh and your change doesn't actually work for me,
23:19:41 dansmith because what I'm hitting is the recursion on the lock,
23:20:06 bauzas jroll: or others, any way to guess the hypervisor_hostname given a specific instance ?
23:20:10 dansmith so I have to do it before I take the lock, which is fine because the lock is just to protect global state, which isn't a problem if our context is targeted
23:20:22 bauzas I need to know the ComputeNode record related to a specific instance
23:20:55 bauzas but I don't have a specific helper method in the object module for that
23:21:26 bauzas mmmmpf
23:21:39 bauzas I can see a 'node' attribute
23:21:48 bauzas in the instance object
23:25:14 melwitt dansmith: I still don't understand the problem. it's failing to acquire the lock or it keeps recursing or?
23:25:36 dansmith fails to acquire the lock recursively yeah
23:27:55 melwitt I'm trying to remember how it tries to acquire it recursively. I know we talked about this a couple of months ago
23:28:10 dansmith well, this is a new way
23:28:22 dansmith let me continue my unclean quest and see if I get anywhere
23:28:28 dansmith and then you can crap all over it
23:29:25 melwitt heh, okay
23:32:22 bauzas calling it a day \o
23:34:39 melwitt dansmith: I guess it seems like it shouldn't be able to recurse unless a get_context_manager happened within a target_cell, or something
23:35:04 dansmith melwitt: right, but the target_cell sets ctxt.db_connection and then the rpc call removes it
23:36:23 melwitt dansmith: so you're within a target_cell, and then an rpc call occurs and returns, then something else does a db access and get_context_manager can't acquire the lock? all within one target_cell?
23:36:30 dansmith no
23:36:44 dansmith you're in a target cell, then inside the rpc call but your db_connection=None
23:36:55 dansmith no call/return
23:37:10 melwitt oh, on the "other side"
23:37:27 dansmith right
23:37:32 melwitt okay
23:38:53 melwitt yeah, something I played around with but got nowhere was I was trying to assign a default db url per service, so each would know its own default instead of doing the _last_ctxt_mgr thing
23:39:09 dansmith right, which is kinda this
23:39:13 melwitt so that things wouldn't blur across service boundaries
23:39:15 melwitt yeah
23:41:09 openstackgerrit Sujitha proposed openstack/nova master: Remove mox from nova/tests/unit/db/test_db_api.py https://review.openstack.org/434557
23:43:24 melwitt dansmith: looking at my old notes I was trying to patch the rpc dispatch to slap the db url for the service onto the incoming context so all non target_cell db access would use it
23:43:45 dansmith right, that'd be one way,
23:43:58 dansmith but you have to then know what service you're in and what default connection it should be
23:44:08 melwitt but the hard part was trying to match up the service:url and I didn't find a good non-messy way to do that
23:44:10 melwitt yeah
23:44:11 dansmith whereas if we can make it stick during a target call...
#openstack-nova - 2017-02-16
00:05:49 dansmith oh. snap
00:07:57 openstackgerrit STEW TY proposed openstack/nova master: Transform instance.resize_revert notification https://review.openstack.org/396811
00:08:43 melwitt dansmith: good news?
00:08:56 dansmith melwitt: yes, you'll be crapping on it soon
00:09:04 melwitt haha, nice
00:09:13 openstackgerrit STEW TY proposed openstack/nova master: Transform instance.resize_revert notification https://review.openstack.org/396811
00:16:12 openstackgerrit Dan Smith proposed openstack/nova master: Make start and stop admin actions cells-aware https://review.openstack.org/433362
00:16:13 openstackgerrit Dan Smith proposed openstack/nova master: Make CellDatabases fixture work over RPC https://review.openstack.org/434563
00:16:16 dansmith melwitt: ^
00:20:08 melwitt dansmith: looks pretty good to me, actually. it's self-contained and makes it clear what's happening
00:20:10 dansmith melwitt: it works for my test, but haven't checked to see if it breaks anything else just yet
00:20:39 dansmith \o/
00:29:37 dansmith heh, so many unit tests pass None for the context so I'll have to be defensive there

Earlier   Later