Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-08-21
19:37:20 melwitt like a custom class placement doesn't yet know about?
19:37:23 efried yeah
19:37:40 melwitt ahhh that would be easier
19:38:33 melwitt I was thinking inside the box of the standard classes
19:38:33 efried (careful if you're trying to drive this stuff through the report client - some paths will automatically ensure RCs for you)
19:38:48 efried well, you could make those fail too by setting total=0 :)
19:38:52 melwitt this is for osc-placement so should be safe there
19:39:10 efried or total=foo for that matter
19:39:42 melwitt ok, well, thanks for making this sound really stupidly easy :P
19:39:48 efried hell, getting an inventory PUT to *fail* is easy!
19:39:55 efried getting it to *succeed* is the hard part
19:39:58 melwitt well hang on
19:40:06 melwitt needs to fail for only one RP though
19:40:23 openstackgerrit Adam Spiers proposed openstack/nova master: Apply SEV-specific guest config when SEV is required https://review.opendev.org/644565
19:40:23 efried are you putting multiple inventories at once?
19:40:28 efried like via a single REST call?
19:40:39 melwitt this is a new --aggregate command which will loop over RPs in an aggregate and do the same inventory set on all of them
19:40:43 melwitt and I need to fail one in the middle
19:40:57 efried right, so you need to fail one (entire) REST call in a loop of REST calls
19:40:59 melwitt for test coverage for a "log error and continue" behavior
19:41:00 melwitt yeah
19:41:12 efried so yeah, you can just make a bogus inventory value in that one call
19:41:25 efried which is kinda what you're talking about doing I guess.
19:41:29 openstackgerrit Adam Spiers proposed openstack/nova master: Enable booting of libvirt guests with AMD SEV memory encryption https://review.opendev.org/666616
19:41:58 melwitt yeah but that's not what the command wants to do, it does the same 'set' for each
19:42:40 melwitt ideally I'd want to muck with one of the RPs ahead of time such that it would fail an identical request while the others succeed
19:43:00 melwitt so I think I will have to do the allocation thing
19:45:26 efried melwitt: Yeah, you can add a CUSTOM_FOO to (only) that RP and create an allocation against it. Then since you're PUTting the same inventory (which presumably lacks CUSTOM_FOO) against all the providers, it'll fail for that one since it would amount to removing the CUSTOM_FOO inventory.
19:45:30 efried Is that what mriedem suggested?
19:45:46 melwitt yes
19:45:51 efried smart man
19:45:56 melwitt yup
19:55:42 melwitt mriedem: how are you finding the good link to the post_test_hook run under ara-report?
19:56:26 mriedem you know to get to the ara-report right? https://0bde141365cc25860c5b-0b668b63edb9a2dd0711fb78b2335400.ssl.cf1.rackcdn.com/672840/11/check/nova-next/daad30a/ara-report/
19:56:32 melwitt yeah
19:56:45 mriedem i find the post.yaml with 1 task after the devstack-tempest run
19:56:47 melwitt oh, I was clicking the > arrow
19:56:58 melwitt and just saw a ton of hashes
19:56:59 mriedem expand that, then click on the 'changed' box
19:57:05 melwitt thanks
19:57:09 mriedem that gives you a popup,
19:57:18 mriedem then hit the chain link icon in the top right
19:58:47 melwitt got it, thank you
20:05:46 mriedem efried: reading your reply to alex's thread, why do you say that the numa and pci fields in the instance migratoin context are hypervisor specific? as far as i know they aren't, since they are used by libvirt and i think the hyperv driver as well during moves
20:06:11 efried mriedem: then they're specific to libvirt and hyperv.
20:06:24 efried that's luck more than anything else
20:06:31 mriedem you mean because they wouldn't work for powervm or what?
20:06:43 efried yeah, that's the example that springs to my mind obviously.
20:07:34 efried also pretty n/a for ironic if I'm not mistaken.
20:07:45 mriedem ironic doesn't migrate
20:07:50 mriedem but yeah
20:08:03 efried so like, *really* n/a :P
20:08:07 mriedem those fields are only populated if the instance has numa/pci stuff
20:08:11 mriedem which baremetal instances won't have
20:08:45 efried right. And in alex_xu's new world, `resources` would only get populated if the instance has local resources to track.
20:09:22 efried Any time we just need a count, we leave that bit empty and just use the placement inventory structure.
20:15:35 mriedem yeah i haven't read the whole message yet
20:15:40 mriedem nor the etherpad
20:15:45 mriedem only what i've gathered from irc conversations
20:16:16 mriedem efried: mordred: also, this is the patch to remove [neutron]/url from the neutron nova/neutron compute install guide in favor of a KSA-tastic world https://review.opendev.org/677810
20:16:26 mriedem hopefully i didn't misrepresent in the commit message
20:19:06 openstackgerrit Adam Spiers proposed openstack/nova master: Enable booting of libvirt guests with AMD SEV memory encryption https://review.opendev.org/666616
20:22:07 efried mriedem: nicely done, +1
20:27:16 mriedem \o/
20:27:30 mriedem i know you sdk/kds/sta people are particular
20:28:34 mriedem heh, i just realized that archive_deleted_rows returns 1 meaning, we archived some stuff, keep going until you get rc=0,
20:28:43 mriedem but if the command blows up in some unexpected way, we return 1 (rather than like 255)
20:28:53 efried that's eek
20:28:54 mriedem dansmith: that seems not ideal yeah? ^
20:29:22 mriedem https://github.com/openstack/nova/blob/master/nova/cmd/manage.py#L2625
20:29:37 dansmith hmm,
20:29:42 dansmith I thought we had an error case
20:29:56 dansmith oh
20:30:02 dansmith you mean if it bubbles up to main
20:30:47 mriedem right
20:30:51 mriedem TypeError or something dumb
20:31:44 dansmith yeah
20:32:01 dansmith honestly, that "return 1" there in main probably conflicts with other things too, no?
20:32:12 dansmith 255 is a common thing for "I dunno wtf happened"
20:33:33 mriedem right, nova-status returns 255, that's what got me looking at this
20:38:00 openstackgerrit Matt Riedemann proposed openstack/nova master: Document archive_deleted_rows return codes https://review.opendev.org/677819
20:38:48 efried dansmith: couple easy ones, if you have a sec: https://review.opendev.org/#/c/676707/ and successor?
20:40:00 dansmith mriedem: yeah I dunno what danger might lurk by changing that now, but surely seems like the right thing
20:40:10 mriedem https://bugs.launchpad.net/nova/+bug/1840978
20:40:11 openstack Launchpad bug 1840978 in OpenStack Compute (nova) "nova-manage commands with unexpected errors returning 1 conflict with expected cases of 1 for flow control" [Undecided,New]
20:40:14 mriedem yeah, just said that in the bug report,
20:40:19 mriedem i think it's just a simple upgrade reno yeah?
20:41:11 dansmith mriedem: well, more like the impact to tripleo, osa, etc
20:41:18 dansmith this potentially affects lots of manage commands
20:41:39 dansmith although this is really SHTF level fail cases, so nobody should really be depending on that
20:43:17 mriedem i counted simple_cell_setup, map_instances and heal_allocations as returning 1 in non-error cases
20:44:06 efried nobody should be relying on 1 being a real failure, because if they are, they're f'ed anyway.
20:44:21 dansmith mriedem: no, I mean cases in all the other commands that could raise
20:44:22 efried so I think this is a situation where it's okay to change it.
20:44:50 dansmith efried: that's my point, the only case would be somewhere that someone tests "is something done yet" by checking if something fails, and uses ret=1 to check
20:44:54 openstackgerrit Eric Fried proposed openstack/nova master: Revert "Filter UnsupportedServiceVersion warning" https://review.opendev.org/677825
20:45:17 efried right, which is invalid.
20:45:21 efried I'm agreeing with you
20:45:24 efried mark your calendar
20:46:08 dansmith I don't think it's invalid

Earlier   Later