Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-04-27
14:13:57 mriedem so
14:13:58 mriedem PUT /os-services/disable { "host": "host1", "binary": "nova-compute" }
14:14:00 mriedem becomes
14:14:04 mriedem PUT /os-services/{service_id} { "status": "disabled" }
14:14:12 mriedem where service_id is a uuid
14:14:28 mriedem having said all that, i think we're fine with what you're proposing :)
14:14:39 mriedem we == me and the turd in my pocket
14:15:02 dansmith ack, yep
14:42:16 openstackgerrit Dan Smith proposed openstack/nova master: re-Allow adding computes with no ComputeNodes to aggregates https://review.openstack.org/460616
14:54:22 mriedem hmm, isn't it possible to override a mock?
14:54:35 mriedem like if the test class does a global mock of something but a test case wants to override it
14:54:49 mriedem i thought just decorating the test case would override the mock from the test class...
14:54:52 mriedem but apparently not
14:58:35 artom mriedem, maybe explicit assignment will work?
14:58:58 artom nova.stuff.moar = mock.Mock()
14:59:21 zioproto mriedem: in the table block_device_mapping what is the column no_device for ? for one attachment I have value 0 and for the other NULL
15:03:20 openstackgerrit Béla Vancsics proposed openstack/nova master: [WIP] Transform aggregate.update_metadata notification https://review.openstack.org/460625
15:03:37 mriedem zioproto: good question
15:03:47 mriedem zioproto: let me consult my tome of bdm fields
15:08:12 mriedem not doc'ed in the api
15:08:24 mriedem schema says it could be a dict in the api?
15:08:38 mriedem but it's treated like a boolean,
15:08:43 mriedem it's definitely some legacy bdmv1 thing
15:09:15 mriedem https://github.com/openstack/nova/blob/master/nova/api/openstack/compute/schemas/block_device_mapping_v1.py#L34
15:10:10 mriedem https://github.com/openstack/nova/blob/master/nova/compute/api.py#L2969
15:14:15 mriedem boolean in the object model https://github.com/openstack/nova/blob/master/nova/objects/block_device.py#L85
15:14:47 mriedem boolean in the data model https://github.com/openstack/nova/blob/master/nova/db/sqlalchemy/models.py#L622
15:16:31 mriedem zioproto: i'm not really sure what the hell it means,
15:16:40 mriedem but it looks like maybe your customer did a swap volume operation on that instance?
15:16:48 mriedem because that sets 'no_device': None
15:16:49 mriedem in the bdm
15:18:06 zioproto also
15:18:06 zioproto possible
15:18:09 zioproto I dont understand
15:18:15 zioproto why when I attach this volume
15:18:20 zioproto the device is /dev/hda
15:18:26 zioproto this is the attachment
15:18:34 zioproto where this /dev/hda is generated from ?
15:19:05 mriedem zioproto: should be from nova/virt/libvirt/blockinfo.py i think
15:19:59 mriedem see find_disk_dev_for_disk_bus
15:23:33 mriedem mdbooth: god i'd love to remove all of this bdm v1 code
15:23:56 mdbooth mriedem: Hehe
15:24:03 zioproto I try with command line to force /dev/sdb
15:24:08 zioproto but it ignores me
15:24:16 mriedem zioproto: the libvirt driver ignores the device name you pass in
15:24:22 mriedem when attaching a volume, or booting from volume
15:24:30 mdbooth mriedem: How would we deprecate it?
15:24:42 lyarwood with fire?
15:24:47 zioproto but there must be a bug that assigned this /dev/hda name
15:24:52 mriedem mdbooth: i think we'd start by deprecating the legacy bdm v1 fields in the api
15:24:55 zioproto cant be both /dev/sda and /dev/hda
15:25:17 mdbooth lyarwood: Well the mechanism would be to take off and nuke it from orbit, of course. I meant the prior communication.
15:25:42 mriedem mdbooth: but that's like microversion 2.50 at this point, and unless we raise minimum microversions in the api that never goes away,
15:25:50 mriedem and even if we did, it would probably take a long time before we could remove it
15:26:11 mriedem zioproto: so the guest xml says it's /dev/sda?
15:26:19 mriedem but the bdm attachment table shows it as hda?
15:26:26 sfinucan sdague: If I want to capture logging, which fixture do I need to use?
15:26:47 mriedem zioproto: can you tell if they performed a swap volume on the instance? there should be an instance action recorded for it
15:26:58 sfinucan I saw a comment on one of mriedem's patches to not mock oslo.log, but I can't find said patch
15:27:31 mdbooth sfinucan: Interesting. I was about to say I've just mocked the log object before.
15:27:46 mriedem zioproto: actually it looks like we don't record an instance action event for swap volume
15:27:47 mriedem damn
15:28:04 zioproto I know that this volume was imported, not created with openstack
15:28:12 mriedem sfinucan: mdbooth: OutputStreamCapture i htink
15:28:14 mriedem *think
15:28:20 mriedem zioproto: imported?
15:28:22 mriedem via cinder?
15:28:23 zioproto it is some machine converted from vmware or something
15:28:30 sfinucan mriedem: ta
15:29:02 mdbooth mriedem: volume-only image?
15:29:32 openstackgerrit Matt Riedemann proposed openstack/nova master: Remove compatibility code for flavors https://review.openstack.org/460377
15:30:09 sdague sfinucan: actually, I think we buried the hook for that
15:31:00 mriedem sfinucan: i was actually curious too which fixture sdague was talking about
15:31:04 mriedem since we have a few that seem related
15:31:06 sdague nova_fixtures.StandardLogging() would give us a fixture with self.logger that we can pull on
15:31:16 sdague but we didn't assign a reference to get back to it
15:31:48 mriedem so the base test case needs to store self.logger = self.useFixture(nova_fixtures.StandardLogging()).logger?
15:31:50 mriedem right?
15:32:04 mriedem and then i can do self.assertIn('foo', self.logger.info)?
15:32:23 johnthetubaguy mriedem: dansmith: was the ping I missed about the duplicate hostnames thing in os-services? I think I was really asking about FQDN vs hostname. There wouldn't be duplicate FQDNs.
15:32:33 mriedem oh self.logger.output
15:32:50 openstackgerrit Sean Dague proposed openstack/nova master: Expose StandardLogging fixture for use https://review.openstack.org/460642
15:32:50 dansmith johnthetubaguy: okay, when do we use hostnames that aren't fqdns?
15:32:57 dansmith johnthetubaguy: not for service registration, AFAIK
15:33:04 sdague mriedem: or that ^^^
15:33:30 johnthetubaguy dansmith: oh, I thought we default to hostname rather than fqdn, that might just be me getting all mixed up
15:33:41 sdague mriedem: I thought it had gotten exposed
15:33:41 dansmith johnthetubaguy: not that I've ver seen
15:34:01 sdague but apparently not, but there is a whole mechanism for it being captured, so we should us that instead of mocks
15:34:03 zioproto mriedem: so it was two disks from vmware. We converted this disks to glance images, and then to cinder volumes.
15:34:15 johnthetubaguy dansmith: ok, thats probably just my confusion then, seems fine to assume no duplicate fqdn's
15:34:41 openstackgerrit Samantha Blanco proposed openstack/nova master: Remove flavor-rxtx policy https://review.openstack.org/444036
15:34:57 dansmith johnthetubaguy: ack
15:35:32 mriedem zioproto: ok, the device name won't be stored in the cinder volumes anyway, only the mountpoint meta in cinder, but that's set by nova when the volume is attached
15:36:05 mriedem on a complete tangent, it seems like we should be recording instance actions for swap volume, since that's a thing that happens mostly out of band from a nova api pov
15:36:11 mriedem since cinder initiates it via retype
15:38:15 mriedem dansmith: btw, https://review.openstack.org/460377 turned out to be kind of a beast
15:38:34 dansmith sweet
15:38:39 johnthetubaguy dansmith: so we don't really use the fqdn in here: https://github.com/openstack/nova/blob/master/nova/conf/netconf.py#L55 although I could be reading the docs wrongly: https://docs.python.org/2/library/socket.html#socket.gethostname
15:39:06 dansmith johnthetubaguy: but that's for things like telling other computes how to contact us

Earlier   Later