| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2021-02-24 | |||
| 13:56:22 | stephenfin | dansmith: no worries :) | |
| 13:56:29 | stephenfin | ack | |
| 13:56:57 | stephenfin | bauzas: fwiw ^ (though I rank secure-boot slightly higher rn, in case it's between the two) | |
| 13:57:50 | bauzas | ack, reviews reviews reviews | |
| 13:58:13 | bauzas | in the middle I should take opportunity of revisiting my own RPC API change | |
| 13:58:17 | stephenfin | all the reviews! | |
| 14:04:44 | stephenfin | prometheanfire: https://review.opendev.org/c/openstack/nova/+/741351 is approved so mock 4.x should be unblocked from our side now, fwiw | |
| 14:04:52 | stephenfin | *is merged | |
| 14:14:11 | sean-k-mooney | gibi: would you have time to look at this small plamcent change https://review.opendev.org/c/openstack/placement/+/777174 | |
| 14:14:42 | sean-k-mooney | gibi: it looks like we skiped a step as part of the rescent os-traits release. | |
| 14:14:52 | gibi | looking | |
| 15:03:46 | jkulik | hm ... is there a mechanism to clean up lingering build-requests? We seem to have a couple still around ... | |
| 15:14:17 | sean-k-mooney | jkulik: i belive they shoudl be cleaned up when the instance exits teh build status | |
| 15:14:43 | sean-k-mooney | so if they are lingering you proably have vms stuck in build or you have deleted them and we did not clean up properly | |
| 15:14:55 | sean-k-mooney | jkulik: do you know which is the case? | |
| 15:15:11 | jkulik | I cannot find them in any cell DB. I assume they never made it there. | |
| 15:17:55 | jkulik | so probably network problems at the time. if there's no automatic mechanism, we have to clean them manually before upgrading to rocky | |
| 15:18:13 | jkulik | they're missing the "trusted_certs" attribute and thus make server listing fail with an HTTP 500 | |
| 15:33:26 | gibi | lyarwood: do you remember why the rpc_response_timeout=1 is needed in https://review.opendev.org/c/openstack/nova/+/768351/2/nova/tests/functional/regressions/test_bug_1909120.py#31 ? That test case sometimes fails to me locally with http://paste.openstack.org/show/802963/ and I try to figure out why | |
| 15:34:52 | sean-k-mooney | gibi: hum maybe it was jus tto have it time out quickly | |
| 15:34:53 | gibi | wondering if what happens is that the timeout is enforced even if we have castAsCall used, and the actual call takes longer than 1 sec | |
| 15:35:48 | sean-k-mooney | looking at it quickly im not sure. settign it to like 5s or 10 likely is ok | |
| 15:36:59 | sean-k-mooney | we should not really be relying on that for timing anyway | |
| 15:37:22 | gibi | I don't see where it in the test helps if an rpc times out early | |
| 15:37:28 | gibi | hence my question | |
| 15:38:14 | sean-k-mooney | the closes is proabnly line 59 | |
| 15:38:31 | sean-k-mooney | but we shoudl ahve a wait before that if we were relying on it | |
| 15:39:23 | lyarwood | gibi: yeah so that's just to fail the call to detach a volume on the down compute, otherwise we'd wait for the default timeout to be hit. | |
| 15:39:52 | lyarwood | gibi: and yes it's normally a cast but thanks to castascall here we timeout | |
| 15:40:30 | gibi | lyarwood: hm so after the fix https://review.opendev.org/c/openstack/nova/+/768352 the change in the timeout is not used | |
| 15:41:08 | gibi | lyarwood: so what I see is that the _create_server fails in that test case sometimes for me locally as the scheduler calls the conductor to build the instance but the conductor takes more than 1 sec to reply | |
| 15:41:25 | lyarwood | ah yeah I see | |
| 15:41:31 | gibi | if I undertand correclty after the fix the shrinked timeout is not needed any more | |
| 15:41:36 | lyarwood | correct | |
| 15:41:52 | gibi | cool, then I will remove that to stabilize the func test on a small compute | |
| 15:42:03 | gibi | I mean on a slow node | |
| 15:42:13 | lyarwood | yup, apologies for missing that, I hadn't seen it fail anywhere like that while working on it | |
| 15:43:15 | openstackgerrit | Merged openstack/nova master: apidb: Add manage.py script https://review.opendev.org/c/openstack/nova/+/771419 | |
| 15:44:11 | gibi | lyarwood: no worries. I tend to collect a list of unstable func and unit test I see locally and time to time go thought | |
| 15:45:24 | stephenfin | lyarwood: Think you could take a quick look at this stable patch? https://review.opendev.org/c/openstack/nova/+/751240 | |
| 15:45:44 | stephenfin | Context is that it'll help a student working on openstacksdk patches for adding placement support | |
| 15:46:07 | lyarwood | stephenfin: yup looking | |
| 15:51:01 | sean-k-mooney | stephenfin: gibi lyarwood quick question https://github.com/openstack/nova/blob/master/nova/db/sqlalchemy/models.py#L1483 | |
| 15:51:22 | sean-k-mooney | the dev_type filed in the db is restrited to a lenth of 8 | |
| 15:51:27 | sean-k-mooney | type-vdpa is 9 | |
| 15:51:42 | gibi | /o\ | |
| 15:51:44 | stephenfin | drop type- | |
| 15:51:47 | sean-k-mooney | so shoudl i extend the column or just sotre vdpa | |
| 15:51:56 | stephenfin | it was silly in the first place | |
| 15:52:11 | gibi | OK, that works for me the type prefix seems redundant | |
| 15:52:15 | sean-k-mooney | ya i really doubt it save any space in reality | |
| 15:52:28 | sean-k-mooney | cool ill do that so | |
| 15:52:45 | sean-k-mooney | and ya the prefix is not really useful | |
| 16:15:55 | prometheanfire | sean-k-mooney: yep, thanks :D | |
| 16:16:07 | prometheanfire | stephenfin: meant that for you about the mock thing :D | |
| 17:52:35 | openstackgerrit | Clark Boylan proposed openstack/nova master: DO NOT MERGE testing a thing https://review.opendev.org/c/openstack/nova/+/777444 | |
| 18:00:49 | lyarwood | gibi: https://blueprints.launchpad.net/nova/+spec/ephemeral-storage-encryption - I've moved this to Deferred, not sure if that's the correct state but I'm punting it to at least X, I'll document why upstream next week once I'm back from PTO. | |
| 18:01:36 | openstackgerrit | Clark Boylan proposed openstack/nova master: DO NOT MERGE testing a thing https://review.opendev.org/c/openstack/nova/+/777444 | |
| 18:01:37 | gibi | lyarwood: thanks for that | |
| 18:01:52 | gibi | lyarwood: right now Deferred is OK, I will do the cleanup after FF anyhow | |
| 18:02:07 | lyarwood | gibi: ack cool | |
| 18:20:18 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: Remove short call timeout from func test https://review.opendev.org/c/openstack/nova/+/777447 | |
| 18:20:56 | gibi | lyarwood: a small fix for the func test we talked about ^^ | |
| 19:06:47 | gregwork | can nova flavors be created that have multiple ephemeral disks? | |
| 19:06:59 | gregwork | i know you can nova boot --ephemeral x --ephemeral y | |
| 19:07:35 | gregwork | trying a nova flavor-create --ephemeral x --ephemeral y doesn't respect this via the CLI | |
| 19:13:12 | sean-k-mooney | gregwork: yes and no | |
| 19:13:24 | sean-k-mooney | you can have multiple ephmeral disk | |
| 19:13:36 | sean-k-mooney | i dont think you can spcify that in the flavor | |
| 19:13:52 | sean-k-mooney | i think you specify that durign instance creation | |
| 19:14:21 | sean-k-mooney | there is quite an obsucre way to do this ill see if i can find it but its not commonly used as far as i am aware | |
| 19:15:06 | sean-k-mooney | ok not that obscure | |
| 19:15:08 | sean-k-mooney | ova boot --flavor test-flavor --image cirros --nic net-name=test-network --ephemeral size=1,format=ext4 --ephemeral size=2,format=ext4 test-vm | |
| 19:15:34 | gregwork | so pick whatever flavor and just boot it with whats needed | |
| 19:15:35 | sean-k-mooney | gregwork: but basically the amount of space in the flavor can be subdevied during boot | |
| 19:15:54 | gregwork | oh | |
| 19:15:58 | sean-k-mooney | ya provided you dont excee the amount of space in the flavor its fine | |
| 19:16:14 | gregwork | do the value of "disk" | |
| 19:16:28 | sean-k-mooney | disk is root disk | |
| 19:16:45 | gregwork | yeah so im curious as to what you mean by the space in the flavor | |
| 19:17:06 | sean-k-mooney | therre are 2 storage vaules in the flavor | |
| 19:17:45 | sean-k-mooney | disk and OS-FLV-EXT-DATA:ephemeral | |
| 19:18:22 | sean-k-mooney | https://docs.openstack.org/api-ref/compute/?expanded=add-flavor-access-to-tenant-addtenantaccess-action-detail,create-flavor-detail#create-flavor | |
| 19:18:42 | sean-k-mooney | OS-FLV-EXT-DATA:ephemeral is what is used to limit --ephemeral | |
| 19:19:32 | gregwork | i was under the impression that that value is what will be attached as a separate volume to root | |
| 19:19:43 | gregwork | i did not know it could be sub divided | |
| 19:19:57 | sean-k-mooney | gregwork: some things to keep in mind are ephemeral disk are not snapshot so that also means they are not preserved on shelve | |
| 19:20:27 | sean-k-mooney | gregwork: most people dont if you dont pass any args that will be prented as a singel addtional disk to the instance by default | |
| 19:20:50 | gregwork | thats kind of cool | |
| 19:20:56 | sean-k-mooney | but if the flaovr has addtional ephemeral storage you can subdevide it | |
| 19:21:07 | sean-k-mooney | gregwork: the other thing to kep in mind is that resize is undefined | |
| 19:21:28 | sean-k-mooney | or rather resize where the epmeral size changes is undefiend (read broken as hell) | |
| 19:21:57 | sean-k-mooney | so its cool be there be dragons if you try to resize down | |
| 19:22:01 | sean-k-mooney | up might be ok | |
| 19:22:14 | sean-k-mooney | but again very rarely used feature | |
| 19:22:25 | sean-k-mooney | also very old i think it predate cinder | |
| 19:23:35 | gregwork | the situation we are trying to work around is this quick poc we are running needs fast disk and cinder is overloaded. So they are going to stuff some flash in 3 computes dedicated to this project | |
| 19:23:46 | sean-k-mooney | gregwork: if you want to do it with the openstack client you have to manually build the block device mappings dict and use the local volum type to do this | |
| 19:23:47 | gregwork | so we are going to run off the local ephemeral | |
| 19:25:20 | sean-k-mooney | as long as people are awrae fo the limiations, local scratch space is exactly what that flavor value is for | |