Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-10-10
14:42:29 sean-k-mooney which i think we already do
14:43:33 mriedem maintaining the xenapi driver isn't hard either but we've deprecated that
14:43:37 mriedem anyway, i threw it on the meeting agenda
14:44:15 sean-k-mooney well the last type i looke the xen server side fo that stilll need python 2 to work
14:44:31 sean-k-mooney infact when i looked at the start of stine it still need 2.6
14:44:37 mriedem i think you are confusing the guest image stuff which used to be pinned to like 2.4
14:44:51 sean-k-mooney maybe
14:44:59 mriedem the driver itself clearly works with py3 otherwise we'd have had to blacklist all of it's unit tests
14:45:06 sean-k-mooney i was looking a kolla-ansible thread on xen suppport
14:45:33 sean-k-mooney they were having issue with xen due to some pinned python stuff
14:49:22 openstack Launchpad bug 1826269 in kolla-ansible "kolla-ansible xen fails with package." [Undecided,In progress] - Assigned to James (james-o-benson)
14:49:22 sean-k-mooney this might have been the issue i was thinking ov https://bugs.launchpad.net/kolla-ansible/+bug/1826269
14:55:19 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_image() support to the compute/{rpcapi,api,manager} https://review.opendev.org/687138
14:55:19 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_image() driver method and libvirt implementation https://review.opendev.org/687137
14:55:20 openstackgerrit Dan Smith proposed openstack/nova master: Add image caching API for aggregates https://review.opendev.org/687140
14:55:20 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_images() to conductor https://review.opendev.org/687139
14:55:21 openstackgerrit Dan Smith proposed openstack/nova master: WIP: Add image precaching docs for aggregates https://review.opendev.org/687348
15:12:43 dansmith mriedem: so on the api thing, are you asking me to stop catching the non-nova exceptions in conductor api so that they bubble up?
15:24:03 mriedem those should probably be 500s right?
15:24:13 mriedem e.g. if we can't auth to glance or glance is down
15:24:19 mriedem we shouldn't 400 in that case
15:25:21 dansmith well, that's one reason it could happen, but I dunno what all else
15:25:22 mriedem i would think there are 2 obvious errors we could get from glance which would be (1) the image doesn't exist and (2) the user doesn't have access to the image, but i'd think that would result in a 404 from glance (not sure if it'd be a 403)
15:25:23 dansmith but I guess so
15:25:55 dansmith so unrelated to that,
15:26:06 dansmith in fixing up the not-using-microversion problem in the api patch,
15:26:26 dansmith I subclassed the aggregates test for my two new tests, and now I'm getting this: https://pastebin.com/3McztMky
15:26:36 dansmith which seems to be in some really ancient code, so I can't imagine it has changed recently
15:26:41 dansmith but it's auth-related
15:26:52 dansmith does that resonate with you?
15:27:06 mriedem i bet efried would know off the top of his head since he was just ripping that out
15:27:25 dansmith I was going to ping him until I saw the code was really old
15:27:43 dansmith but yeah
15:27:45 mriedem is that using the admin_api or api?
15:27:56 dansmith admin, but just like the test I subclassed, which runs fine
15:27:59 mriedem i wouldn't expect a keyerror though, i'd expect a 403
15:28:11 mriedem is that code up or still local?
15:29:00 efried dansmith: unless you have https://review.opendev.org/#/c/687416/ or above in your sandbox...
15:29:06 dansmith I wonder if it's because I'm doing the microversion wrong, hang on.. just realized the parent doesn't have a dot in the version
15:29:20 mriedem yeah so if the api fixture was created with version v2,
15:29:24 mriedem and you're passing a microversion header,
15:29:29 mriedem i'd think something is going to blow up there
15:29:32 dansmith yeah, that's it.. so in here: https://review.opendev.org/#/c/687140/5/nova/tests/functional/test_aggregates.py
15:29:39 mriedem yar
15:29:49 dansmith I misread it as api_version and effectively changed v2 to v2.80 and it explodes that obscure way
15:30:38 dansmith am I supposed to be able to set api_major_version='v2.80' ?
15:31:14 mriedem no
15:31:20 mriedem it's the thing that goes into the fixture init
15:31:24 mriedem either v2 or v2.1
15:31:52 mriedem https://review.opendev.org/#/c/687140/6/nova/tests/functional/integrated_helpers.py@437
15:31:55 dansmith cool, well I have confirmed that it's failing as expected (I guess ;D)
15:32:34 dansmith okay, microversion= works, got it
15:32:46 dansmith a teensy bit obscure in the failure mode
15:33:12 mriedem obviously the fixture or TestOpenStackClient or whatever could detect that you're trying to send a microversion header to a v2(.0) route and blow up in a more obvious way
15:33:17 efried what I can contribute to the conversation is, if you see x-server-management-url in a traceback, it's breaking in test code, nothing to do with real life.
15:33:48 efried the microversion and api_major_version class members are for template substitutions afaiu
15:33:58 mriedem hells nah
15:34:01 efried the latter also used to construct the auth url
15:34:04 mriedem they are for the api fixture and resulting client
15:34:22 dansmith efried: yeah I know it's just test stuff
15:34:31 efried if you want to send a request at a specific microversion, I think you still need to use a kwarg to your API method.
15:34:49 mriedem well,
15:34:54 mriedem the api fixture client stuff handles that
15:35:16 mriedem self.api = self.useFixture(nova_fixtures.OSAPIFixture('v2.1')).api
15:35:24 mriedem self.api.microversion = '2.80'
15:35:26 mriedem something like that
15:35:27 efried oh, I guess the fixture also uses the microversion as the default if you don't specify a kwarg.
15:35:32 mriedem correct
15:36:22 efried the aforementioned series rips some of that magic out of the test-only paths so it happens more like it would in real life. But I don't think what you're doing would be affected (other than to fail differently if you f it up)
15:48:50 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_images() to conductor https://review.opendev.org/687139
15:48:51 openstackgerrit Dan Smith proposed openstack/nova master: WIP: Add image precaching docs for aggregates https://review.opendev.org/687348
15:48:51 openstackgerrit Dan Smith proposed openstack/nova master: Add image caching API for aggregates https://review.opendev.org/687140
15:49:14 dansmith mriedem: okay, conductor/api now lets non-nova exceptions up to the top
15:50:20 dansmith I didn't even really try on the release note, because I bet you're going to want all kinds of fancy links to docs and stuff, which don't exist until the next patch
15:50:30 dansmith so I figure I'll take my -1 on that :P
15:51:59 mriedem i think i said something to the effect of we can add a release note for the new route and microversoin and config option and in the docs patch update the release note to link to the docs for more details, i.e. keep the reno short and sweet
15:52:16 dansmith cool, well, then it may work after all :)
15:54:53 mriedem ok i left a comment in there, but haven't gone over everything again yet
15:55:17 dansmith s'fine, let me begrudgingly add your requested comments before you hit it up again anyway
15:59:46 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_image() support to the compute/{rpcapi,api,manager} https://review.opendev.org/687138
15:59:46 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_image() driver method and libvirt implementation https://review.opendev.org/687137
15:59:47 openstackgerrit Dan Smith proposed openstack/nova master: Add image caching API for aggregates https://review.opendev.org/687140
15:59:47 openstackgerrit Dan Smith proposed openstack/nova master: Add cache_images() to conductor https://review.opendev.org/687139
15:59:48 openstackgerrit Dan Smith proposed openstack/nova master: WIP: Add image precaching docs for aggregates https://review.opendev.org/687348
15:59:52 dansmith word salad baby
16:02:25 mriedem ok +2 on the bottom 2 changes, i'll get to conductor and the api after lunch
16:03:00 dansmith sha-wing
16:10:00 mloza Hi. I have this issue with nova conductor where instances are stuck creating. This happened after I did a re-ip of the machines. Other nova services except conductor. This is the log https://clbin.com/pErEA
16:11:05 dansmith mloza: if the ip of your rabbit changed you probably need to update your cell mappings
16:11:10 mriedem lyarwood: i've left a few thoughts on https://review.opendev.org/#/c/682594/
16:11:14 mriedem lyarwood: two things:
16:11:14 dansmith mloza: nova-manage cell_v2 update_cell
16:11:17 dansmith mloza: also, please see topic
16:11:29 mriedem 1. seems we could workaround the instance uuid lock causing MessagingTimeout by using the long_rpc_timeout on that rpc call,
16:11:46 mriedem 2. your cleanup in the api is best effort - might not catch all cases but it's worth a try, yeah?
16:12:02 lyarwood mriedem: yup sorry just working my way towards this
16:12:21 lyarwood mriedem: 1. nice idea, yeah that would be useful, 2. yes it's just best effort
16:12:29 mriedem 3. long-term we should create the bdm in the api, pass it down to compute, and then delete in api if we hit a failure
16:12:45 mriedem i.e. implement vishy's todo from like 2012
16:12:54 lyarwood mriedem: yeah just read that comment, yup that would be much much cleaner

Earlier   Later