| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-03-10 | |||
| 12:16:57 | sean-k-mooney | sure will do | |
| 12:18:49 | sean-k-mooney | im not sure i fully follow what addtional semantic checks i need to add in the third patch | |
| 12:19:18 | sean-k-mooney | well i know we shoudl be deduping providers | |
| 12:19:39 | sean-k-mooney | based on by name vs by uuid vs by $COMPUTE_NODE | |
| 12:20:11 | sean-k-mooney | i just havent fully loaded enough of the code into my head to fully grapse that yet | |
| 12:20:23 | sean-k-mooney | but if i get stuck ill let ye know | |
| 12:21:15 | sean-k-mooney | on https://review.opendev.org/#/c/676522/41/nova/compute/resource_tracker.py@1730 is the exception for all compute nodes | |
| 12:21:34 | gibi | yeah, I think the core comment is about identifying the same provider 3 different ways | |
| 12:22:20 | sean-k-mooney | so one thing that confused me a bit is why to we error in that case instead of merging | |
| 12:22:43 | sean-k-mooney | e.g. if i do both by name and by uuid | |
| 12:23:02 | sean-k-mooney | and add different providers/inventories in each it should be fine right | |
| 12:23:21 | gibi | both behavior could be acceptable but I guess the spec said every RP should be mentioned only once | |
| 12:23:37 | gibi | let me look i tup | |
| 12:23:39 | sean-k-mooney | ok ya well im not going to change it | |
| 12:24:02 | sean-k-mooney | no its fine i was just wondering if there was a fundemental reason i missed | |
| 12:25:14 | gibi | "In the event that a resource provider is identified by both explicit UUID/NAME and $COMPUTE_NODE, only the UUID/NAME record will be used." | |
| 12:25:19 | gibi | https://specs.openstack.org/openstack/nova-specs/specs/ussuri/approved/provider-config-file.html | |
| 12:25:37 | sean-k-mooney | right that make sense | |
| 12:25:51 | sean-k-mooney | but it does sate what happens if you use both UUID and NAME right | |
| 12:26:36 | gibi | I did not found any reference about that case yet | |
| 12:26:54 | sean-k-mooney | anyway its not an issue right now. if we want to support that in the future we can | |
| 12:27:05 | gibi | # same value is used more than once across all provider configs for name | |
| 12:27:05 | gibi | "The consuming nova-compute service will error and fail to start if the" | |
| 12:27:08 | gibi | # or uuid. | |
| 12:27:21 | sean-k-mooney | ah ok perfect | |
| 12:27:59 | sean-k-mooney | im not sure why we chose that but i also dont partically care enough to argure that its too limiting | |
| 12:28:39 | sean-k-mooney | i suspect most people will just use the name | |
| 12:29:04 | gibi | yeah, name for computes, uuid for ironic nodes | |
| 12:29:26 | sean-k-mooney | isnint the ironic node RP name also the uuid | |
| 12:30:10 | sean-k-mooney | the name has to be unique so that what i would have done to make it as simple as possible but uuid for ironic makes sense | |
| 12:31:34 | gibi | yes, for ironic name and uuid are the same in placement | |
| 12:41:53 | openstackgerrit | Merged openstack/nova stable/rocky: libvirt: Ignore DiskNotFound during update_available_resource https://review.opendev.org/711278 | |
| 13:15:55 | mordred | smcginnis: https://review.opendev.org/#/c/710214/ | |
| 13:16:06 | mordred | smcginnis: that would be the patch that broke that test | |
| 13:22:29 | mordred | jkulik: https://review.opendev.org/#/c/710214/ caused a behavior change and broke SDK's functional tests | |
| 13:23:52 | gibi | mordred: did you depend on the HTTP 500 from nova? or what was the exact behavior change that broke the SDK? | |
| 13:24:17 | mordred | the opposite - we were getting NotFound before when trying to delete something by name | |
| 13:24:24 | mordred | and now we're getting BadRequest | |
| 13:24:49 | mordred | to be fair - there's ALSO a bug in SDK which this has shown | |
| 13:25:54 | gibi | can nova do something to help SDK out? the only thing I can imagine is to revert https://review.opendev.org/#/c/710214/ and re-propose it with a microversion bump on the API | |
| 13:26:06 | gibi | would that help? | |
| 13:26:07 | mordred | which is that we were accepting a name_or_id argument to our delete method and then it looks like we're passing taht straight on - which will never have worked | |
| 13:26:50 | mordred | gibi: I'm not sure - I keep going back and forth about whether the behavior change is one of those "someone might break but that person was already broken" times | |
| 13:27:16 | gibi | yeah, I'm hesitant too | |
| 13:27:18 | mordred | I think the more I talk out loud about it - the more I'm coming to think it's actually all a bug on our side that this just showed | |
| 13:32:47 | gibi | if SDK is need to be changed anyhow then I would not jump on the revert | |
| 13:33:04 | gibi | but I can be convinced otherwise | |
| 13:38:36 | mordred | gibi: yeah - I just pushed up https://review.opendev.org/712068 - we had a layer that was erroneously passing name_or_id. we weren't noticing the change because we trapped for ResourceNotFound and then returned False (saying we didn't delete it) - so we _were_ telling the user it wasn't working ... but we can do better there and delete by name (with a GET first) if they pass us a nane | |
| 13:46:28 | jkulik | mordred: I'm sorry, that this caused you problems. Any way I could have checked for this? | |
| 13:47:33 | mordred | jkulik: well - maybe we should consider running sdk functional tests on nova changes | |
| 13:47:57 | mordred | other than that - it was really just uncovering a latent bug on our side - so I think it's a good breakage :) | |
| 13:48:07 | brinzhang | gibi, lyarwood: I find some patch always failed by nova-live-migration task, sometimes we recheck it, that it can be passed, but sut still fails | |
| 13:48:27 | brinzhang | gibi, lyarwood:such as https://review.opendev.org/#/c/706897/ and https://review.opendev.org/#/c/700072/ | |
| 13:49:28 | brinzhang | From their zuul log, I am not find some useful message caused by live migration failed, or I missing something | |
| 13:50:44 | brinzhang | I think we should find why it always failed, IMO, that always a hidden danger | |
| 13:54:05 | brinzhang | lyarwood: I saw you want to migrate nova-live-migration job https://review.opendev.org/#/c/711604/3, that can fix this issue? I will check together :) | |
| 13:57:17 | openstackgerrit | Monty Taylor proposed openstack/nova master: Run sdk functional tests on nova changes https://review.opendev.org/712092 | |
| 13:57:26 | sean-k-mooney | brinzhang: i dont think that lyarwood new job will fix it | |
| 13:57:34 | sean-k-mooney | brinzhang: but it will make it less complex | |
| 13:57:45 | sean-k-mooney | so that might help us determin why its failing | |
| 13:58:01 | mordred | gibi, jkulik, smcginnis: ^^ sake of argument patch to run sdk functional tests over here | |
| 13:58:24 | brinzhang | sean-k-mooney: cool, that good | |
| 13:59:28 | brinzhang | sean-k-mooney: maybe lyarwood wants to migrate this job to zuulv3 | |
| 14:00:28 | brinzhang | I remembered he looks for that migrate v3 documents the day before yesterday | |
| 14:01:32 | sean-k-mooney | brinzhang: yes that is what he is doing | |
| 14:02:31 | sean-k-mooney | brinzhang: so i was suggesting spliting the job into two 1 for ceph image backend and one for cinder/lvm and qcow block migration | |
| 14:02:51 | sean-k-mooney | that will avoid the reconfiguration of the services in the post script and create 2 shorter jobs | |
| 14:03:14 | sean-k-mooney | brinzhang: lyarwood was also considering if the evacuation tests should be seperate. or not | |
| 14:03:30 | sean-k-mooney | i think they should stay in the same job | |
| 14:03:58 | sean-k-mooney | we should be able to merge the nova-livemiration-ceph job with devstack-plugin-ceph-tempest-py3 | |
| 14:04:07 | sean-k-mooney | and just run that multinode with live migration enabled | |
| 14:04:42 | brinzhang | sean-k-mooney: yeah, agree | |
| 14:05:25 | sean-k-mooney | so ideally the total job cout wont increase and devstack-plugin-ceph-tempest-py3 will just get sligly longer since it will be multi node and and nova-livemigation with cinder/qcow will get shorter | |
| 14:05:39 | lyarwood | brinzhang: it isn't going to help stable/pike | |
| 14:05:59 | brinzhang | they should stay in the same job, but when find that issue that we can separate it as two job, but finally we should merged | |
| 14:06:08 | lyarwood | there's an open bug for the vif plugging issue in master | |
| 14:06:25 | sean-k-mooney | lyarwood: there is? | |
| 14:06:39 | sean-k-mooney | that sound like something i should proably be aware of | |
| 14:07:20 | lyarwood | sean-k-mooney: it's an old one, I'll dig it up in a min | |
| 14:07:41 | lyarwood | sean-k-mooney: http://status.openstack.org/elastic-recheck/#1813789 | |
| 14:08:11 | sean-k-mooney | ah ok | |
| 14:08:48 | sean-k-mooney | oh right this si a rache with registering for the event | |
| 14:08:58 | sean-k-mooney | *this is a race | |
| 14:09:05 | gibi | mordred: I have a meeting in paralle. I will read back later | |
| 14:09:11 | lyarwood | brinzhang: for stable/pike we might want to make the LM job NV tbh unless anyone steps up to fix it | |
| 14:09:21 | brinzhang | lyarwood: Maybe vif plug has some influence, let's take a look | |
| 14:10:33 | brinzhang | lyarwood: For stable / pike, I agree with you. | |
| 14:11:26 | sean-k-mooney | we skipped this on master https://review.opendev.org/#/c/663405/ | |
| 14:14:21 | sean-k-mooney | https://review.opendev.org/#/c/667177/ i think is the final fix | |
| 14:14:28 | sean-k-mooney | we went back and fort a few times | |
| 14:14:53 | brinzhang | sean-k-mooney: this is in stable/stein, right? so from stable/stein we skip to test the migration | |
| 14:15:41 | brinzhang | sean-k-mooney: oh..sorry, that's vif-plug issue | |
| 14:16:25 | sean-k-mooney | ya but that should already be on rock too | |
| 14:16:34 | sean-k-mooney | https://review.opendev.org/#/c/670648/2 | |
| 14:17:29 | dansmith | I wonder if it's time to consider dropping the vmware driver? http://lists.openstack.org/pipermail/openstack-discuss/2020-March/013066.html | |
| 14:17:30 | openstack | Launchpad bug 1813789 in OpenStack Compute (nova) "Evacuate test intermittently fails with network-vif-plugged timeout exception" [Medium,In progress] - Assigned to Artom Lifshitz (notartom) | |
| 14:17:30 | sean-k-mooney | right so ^ is marked as a partial fix for https://bugs.launchpad.net/nova/+bug/1813789 | |
| 14:18:44 | sean-k-mooney | dansmith: well one devstack failure is proably not enough but if we have no idea how to fix it an no one responds the maybe | |
| 14:19:15 | artom | sean-k-mooney, that patch was very specific to revert resize tho | |
| 14:19:23 | sean-k-mooney | it would help if they had included the compute agent patch | |