| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-04-18 | |||
| 13:50:44 | jaypipes | naichuans: and what happens if there are allocations on a pGPU that you end up removing from the enabled_vgpu_types CONF? | |
| 13:50:54 | openstackgerrit | Lee Yarwood proposed openstack/nova master: Add regression test for bug #1764883 https://review.openstack.org/562072 | |
| 13:50:54 | openstackgerrit | Lee Yarwood proposed openstack/nova master: WIP compute: Ignore active evacuations during _init_instance https://review.openstack.org/562284 | |
| 13:50:54 | openstack | bug 1764883 in OpenStack Compute (nova) "Evacuation fails if the source host returns while the migration is still in progress" [Undecided,New] https://launchpad.net/bugs/1764883 | |
| 13:51:13 | bauzas | jaypipes: I'm fine with your point too | |
| 13:51:43 | efried | bhagyashris: Yes. I don't know where libvirt-specific functional tests live, but that would be the idea. | |
| 13:51:51 | jaypipes | naichuans: at a bare minimum, I'd like to see some handling of a failed delete of a resource provider due to, say, an allocation existing against that provider. | |
| 13:52:40 | efried | bhagyashris: But I'm saying, also a functional test that hits the specific scenario you demonstrated - where the sharing provider was ignored before the fix (the DISK_GB shows up in the compute node RP) but is observed after the fix (shows up in the sharing provider). | |
| 13:52:59 | naichuans | jaypipes: will check it and add the error handling | |
| 13:53:10 | Kevin_Zheng | efried: gibi replied in the patch, he thought it is better to add it for InstanceUpdatePayload but not in status update | |
| 13:56:19 | bhagyashris | efried: ok, any idea or is there any reference that the test written for libvirt | |
| 13:56:58 | efried | bhagyashris: I have no idea. I know nothing about libvirt :) | |
| 13:57:06 | efried | bhagyashris: I thought *you* were the expert there :P | |
| 13:57:20 | efried | Maybe kashyap knows... | |
| 13:57:23 | stephenfin | Shilpa: No idea, sadly. Could you pull it down and deploy it locally. See what happens | |
| 13:57:40 | stephenfin | Shilpa: I imagine the issue is with nova and how it's parsing the URL. Can't say for sure though | |
| 13:57:53 | jaypipes | naichuans: ty! | |
| 13:58:46 | naichuans | np :) | |
| 13:59:44 | openstackgerrit | Matt Riedemann proposed openstack/nova-specs master: Spec for volume multiattach enhancements https://review.openstack.org/552078 | |
| 14:00:01 | bhagyashris | efried: ohh :) i will try my best | |
| 14:00:54 | Shilpa | stephenfin: i have tested tempst locally for patch set 2, for noVNC v1.0.0, test_novnc failed with 'Not a valid html document in the response.' error and for stable/v0.6, test_novnc failed with 'AssertionError: True is not false : Token must be invalid because the connection closed.' | |
| 14:08:43 | gibi | Kevin_Zheng: thanks. I will look at https://review.openstack.org/#/c/560748 shortly | |
| 14:09:47 | stephenfin | jaypipes, sahid: I have small questions on the first and last of these three patches. Fancy taking a look so I can go/no-go? https://review.openstack.org/#/q/topic:bp/overhead-pin-set+status:open | |
| 14:10:38 | efried | kashyap: bhagyashris was looking for guidance as to where one might find/add libvirt-specific functional tests. I allowed as how I know nothing about libvirt.... but you might. | |
| 14:10:43 | jaypipes | stephenfin: sure thing | |
| 14:13:29 | jaypipes | stephenfin: answered. | |
| 14:13:35 | stephenfin | ta | |
| 14:22:52 | openstackgerrit | Stephen Finucane proposed openstack/nova master: Add CPUWeigher https://review.openstack.org/379525 | |
| 14:30:09 | kashyap | efried: Sorry, being pinged on 4 places, just noticed your response. | |
| 14:30:40 | kashyap | efried: This is the place where I see libvirt functional tests: nova/nova/tests/functional/libvirt | |
| 14:31:00 | efried | well that would make sense :) | |
| 14:31:01 | efried | bhagyashris: ^ | |
| 14:36:02 | sahid | stephenfin: stephenfin, jaypipes I replied to you about rasing an exception | |
| 14:36:24 | sahid | it's not something we should do actually | |
| 14:41:18 | jaypipes | sahid: yep, you made a good point. | |
| 14:41:58 | bhagyashris | efried: thank you :) | |
| 14:42:21 | bhagyashris | kashyap: Thank you :) | |
| 14:44:33 | arvindn05 | bauzas: are you available for the review of https://review.openstack.org/#/c/560718/ - Handle rebuild of instance with new image | |
| 14:44:33 | arvindn05 | ? | |
| 14:45:13 | bauzas | arvindn05: for 15 mins because of an appointment but yes | |
| 14:45:19 | arvindn05 | cool... | |
| 14:46:04 | sahid | jaypipes: cool | |
| 14:46:08 | arvindn05 | did you see alex_xu latest comment? he suggests rejecting images with new traits | |
| 14:46:37 | sahid | stephenfin: i will respin the fix put the option under [compute] section. I'm not sure i understand why of your comments about '::' | |
| 14:47:01 | stephenfin | sahid: Sweet, thanks :) Just replying to https://review.openstack.org/#/c/510897 now | |
| 14:47:24 | arvindn05 | IMO, i think its too much of an edge case to put in logic to handle it... | |
| 14:47:42 | stephenfin | sahid: If you put '::' at the end of the line, the next line is formatted as monospace text, i.e. code. I think that's what you want there, rather than the block comment | |
| 14:47:49 | stephenfin | e.g. | |
| 14:47:57 | stephenfin | here's some descriptive text:: | |
| 14:48:02 | stephenfin | ||
| 14:48:13 | stephenfin | config_option = config_value | |
| 14:48:15 | bauzas | arvindn05: the problem is that we had a huuuuuge problem during Queens about whether we should accept to check the image when rebuilding | |
| 14:48:23 | arvindn05 | i would just go back to the original behaviour of trying to find destinations via placement if the image changes.... | |
| 14:48:25 | bauzas | arvindn05: we had a CVE for that | |
| 14:48:38 | bauzas | arvindn05: then we found some problems with the CVE new behaviour | |
| 14:48:44 | bauzas | arvindn05: then we tried to fix that | |
| 14:48:52 | sahid | oh i think i understand what you mean now | |
| 14:48:56 | bauzas | arvindn05: now we're like middleground | |
| 14:49:41 | arvindn05 | can you reference the conversation for me? i dont have that history unfortunately | |
| 14:51:08 | bauzas | arvindn05: https://etherpad.openstack.org/p/nova-ptg-rocky L317 and below | |
| 14:51:22 | bauzas | we discussed about that during the PTG | |
| 14:51:28 | bauzas | as a retrospective if you wish | |
| 14:51:44 | arvindn05 | will take a look | |
| 14:52:22 | bauzas | arvindn05: so, see, before adding *more* to rebuild about scheduler check, maybe let's discuss about the above | |
| 14:52:25 | arvindn05 | i am trying to understand what the problem is with going to placement always. Only in case that the current host is at capacity, the current host may not get picked during placement. I guess we can add preferred destinations to the placement api so we can prioritize current host if it matches | |
| 14:53:25 | bauzas | arvindn05: there is nothing really related to placement | |
| 14:53:53 | bauzas | arvindn05: it's more about the main issue said in L321 | |
| 14:54:24 | bauzas | calling placement for checking traits is just a new scheduler check | |
| 14:54:47 | bauzas | but looks like mriedem isn't around | |
| 14:55:25 | bauzas | we agreed during the PTG to defer the instance.save() *after* rebuilding and mriedem said he was volunteering thart | |
| 14:55:37 | arvindn05 | yup...if you can make it to the nova meeting tommorow....maybe we can discuss in open topics... | |
| 14:56:03 | openstackgerrit | Sylvain Bauza proposed openstack/nova master: libvirt: mdevs returning parent and vendor PCI info https://review.openstack.org/562304 | |
| 14:56:15 | bauzas | AFAIR, the nova meeting is EU-compatible this week | |
| 14:56:18 | bauzas | so, yup | |
| 14:56:35 | arvindn05 | great :) | |
| 14:57:14 | arvindn05 | i am not sure mreidem patch would effect me.. | |
| 14:57:36 | arvindn05 | i think we first need to get concensus on whether to invoke placement or not.... | |
| 14:57:46 | bauzas | arvindn05: that's correct | |
| 14:58:04 | bauzas | arvindn05: but until we fix the original bug, adding more checks to the scheduler means more problems | |
| 14:58:21 | bauzas | see my problem ? | |
| 14:58:32 | bauzas | s/my problem/my concern | |
| 14:58:47 | bauzas | jaypipes: efried: oh shit, saw the ML thread | |
| 14:59:00 | bauzas | I thought there was consensus on numbered req groups | |
| 14:59:13 | bauzas | jaypipes: efried: so I wrote my NUMA proposal accordingly | |
| 14:59:18 | efried | bauzas: Meet us in #openstack-placement | |
| 14:59:50 | arvindn05 | bauzas: yup...it does muddy the waters...but still not seeing how it would effect my changes though. how does instance.save() call effect my flow? | |
| 15:01:10 | bauzas | arvindn05: your flow is about adding an extra check to the scheduler which eventually calls placement for traits query, right? | |
| 15:02:37 | openstackgerrit | sahid proposed openstack/nova master: compute: introduce cpu_shared_set option https://review.openstack.org/561912 | |
| 15:02:38 | openstackgerrit | sahid proposed openstack/nova master: libvirt: always pass emulator threads policy https://review.openstack.org/561913 | |
| 15:02:39 | openstackgerrit | sahid proposed openstack/nova master: libvirt: place emulator threads on CONF.cpu_shared_set https://review.openstack.org/510897 | |
| 15:02:48 | arvindn05 | bauzas: kindof. The extra check calls the GET /resource_providers/{hostUUID}/traits to get traits and match them against the image | |
| 15:03:51 | arvindn05 | so we dont call placemnt allocation candiadates...but make a query to find the traits of the current host and match them against the image traits | |
| 15:04:39 | arvindn05 | bauzas: essentially its this check set(image_traits_required).issubset(set(current_host.traits)) | |
| 15:07:07 | bauzas | arvindn05: sure, but that restricts the subset of hosts right? | |
| 15:07:31 | bauzas | arvindn05: so a host that'd be accepted in the past wouldn't be now accepted, leading to possibly more NoValidHosts | |
| 15:08:52 | arvindn05 | bauzas: we are not checking agaisnt a subset of hosts though...only the current host. | |
| 15:09:24 | arvindn05 | we would get novalid hosts only if the current host the instance is residing upon does not have the traits | |
| 15:10:12 | arvindn05 | the hostuuid in the RP call basically comes from the resource spec `spec_obj.force_hosts[0], spec_obj.force_nodes[0]` | |
| 15:10:25 | arvindn05 | *request spec | |
| 15:10:58 | bauzas | arvindn05: okay here is the thign | |