| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-08-23 | |||
| 14:13:36 | jroll | ugh | |
| 14:13:42 | ykarel | mriedem, but to avoid that nova bug, jroll suggested to use 0 | |
| 14:13:53 | mriedem | the api says otherwise | |
| 14:14:07 | mriedem | for ironic, the flavor values will likely just be informational, | |
| 14:14:14 | mriedem | and then the extra spec resources: are meant to override i thought | |
| 14:14:30 | jroll | right | |
| 14:14:39 | jroll | which isn't working in these tripleo jobs | |
| 14:14:54 | jroll | https://github.com/openstack/ironic/blob/master/devstack/lib/ironic#L2083 | |
| 14:15:35 | jroll | ah, we don't enable cpu/ram/disk filters | |
| 14:17:43 | ykarel | hmm in tripleo we have those enabled | |
| 14:18:13 | ykarel | enabled_filters=RetryFilter,TripleOCapabilitiesFilter,ComputeCapabilitiesFilter,AvailabilityZoneFilter,RamFilter,DiskFilter,ComputeFilter,ImagePropertiesFilter,ServerGroupAntiAffinityFilter,ServerGroupAffinityFilter | |
| 14:18:13 | ykarel | enabled_filters=RetryFilter,AvailabilityZoneFilter,ComputeFilter,ComputeCapabilitiesFilter,ImagePropertiesFilter,ServerGroupAntiAffinityFilter,ServerGroupAffinityFilter | |
| 14:18:21 | jroll | right | |
| 14:18:30 | jroll | are you also using VMs here? | |
| 14:18:47 | ykarel | no, just ironic | |
| 14:18:51 | ykarel | bms | |
| 14:19:05 | jroll | ah, I'd kill those filters then, they aren't necessary | |
| 14:19:13 | jaypipes | enabled_filters should not include RamFilter, DiskFilter or ComputeFilter. | |
| 14:19:16 | jroll | and please do file a bug about the override thing, if you would | |
| 14:19:39 | jaypipes | perhaps this is the thing I was referring to when I mentioned triple-o was still relying on non-resource-class world. | |
| 14:19:41 | jroll | jaypipes: oh, we don't need those filters anywhere now, right? | |
| 14:19:50 | jaypipes | jroll: correct. | |
| 14:19:51 | jroll | even for VMs? | |
| 14:19:53 | jroll | nice | |
| 14:19:57 | jaypipes | yes, even for VMs. | |
| 14:20:10 | jaypipes | actually, lemme verify... | |
| 14:20:37 | jaypipes | there was a CoreFilter too I think, and I can't remember whether it was that or ComputeFilter which filters out "disabled" nodes. | |
| 14:21:25 | jaypipes | jroll, ykarel: you don't need CoreFilter, DiskFilter and RamFilter. ComputeFilter is still needed | |
| 14:21:28 | sean-k-mooney | ComuteFilter does the disabled host i think | |
| 14:21:32 | jaypipes | it filters out "disabled" nodes | |
| 14:21:32 | mriedem | correction! | |
| 14:21:37 | mriedem | you need those filters for caching scheduler | |
| 14:21:41 | mriedem | which doesn't use placement | |
| 14:21:43 | jaypipes | oh ffs, yes. | |
| 14:21:45 | mriedem | otherwise what jay said | |
| 14:22:03 | jroll | gdi mriedem | |
| 14:23:07 | mriedem | hey, if mgagne_ gives the ok at the ptg that we can drop the caching scheduler, then we can | |
| 14:23:14 | sean-k-mooney | mriedem: we are deleting the cachingschduler this cycle right? | |
| 14:23:26 | mriedem | sean-k-mooney: idk, we were going to drop nova-net in rocky ... | |
| 14:23:26 | sean-k-mooney | ah ptg topic | |
| 14:23:52 | sean-k-mooney | ya... stein for that? | |
| 14:24:41 | sean-k-mooney | the caching schduiler could be kicked out of tree like the ec2 api if people really wanted it for some reason. that is still a plug point right? | |
| 14:24:57 | sean-k-mooney | e.g. you can provide your own schduler driver | |
| 14:25:08 | mriedem | yes i think so | |
| 14:25:11 | mriedem | via stevedore extensions | |
| 14:25:14 | melwitt | johnthetubaguy: I pinged you a long time ago about https://review.openstack.org/533168 that you reviewed in the past, if you can get a chance to look at it | |
| 14:25:18 | openstackgerrit | Stephen Finucane proposed openstack/nova master: tests: Create functional libvirt test base class https://review.openstack.org/407055 | |
| 14:25:19 | openstackgerrit | Stephen Finucane proposed openstack/nova master: tests: Move mocking to setUp https://review.openstack.org/595802 | |
| 14:25:42 | sean-k-mooney | mriedem: ya so we could split it out if there was concerns into its own repo but ingeneral not ship it | |
| 14:25:50 | ykarel | jaypipes, jroll ack, so those filters are not needed anymore, then any reason to keep them still? | |
| 14:25:59 | johnthetubaguy | melwitt: that does have my name all over it, will try take a look | |
| 14:26:15 | jroll | ykarel: the only reason to keep them is if you use the caching scheduler | |
| 14:26:41 | jaypipes | please tell me triple-o isn't setting up the caching scheduler. | |
| 14:26:45 | mriedem | it's not | |
| 14:26:52 | sean-k-mooney | mriedem: if we did that we could move the filters that are only needed with the cahcein scheduler also | |
| 14:26:54 | mriedem | well, i assume it's not :) | |
| 14:26:56 | jaypipes | yay. I can step back from the edge. | |
| 14:27:06 | mriedem | The Edge (tm)? | |
| 14:27:10 | melwitt | johnthetubaguy: sweet, thanks | |
| 14:27:40 | jaypipes | mriedem: touche | |
| 14:27:44 | ykarel | jroll, how that's configured, caching scheduler | |
| 14:28:06 | jroll | heh | |
| 14:28:16 | sean-k-mooney | jaypipes: by default no there is a conf override thing in tripleo so someone could set teh conf values manually but im pretty sure that would make the deployment unsupported downstream | |
| 14:28:53 | jroll | ykarel: conf.scheduler.scheduler_driver | |
| 14:29:29 | sean-k-mooney | ykarel: now that you know. please never set it to anything other then the filter schduler | |
| 14:29:51 | ykarel | jroll, driver=filter_scheduler | |
| 14:29:57 | ykarel | jroll, sean-k-mooney we have ^^ | |
| 14:30:10 | jroll | cue denver: <tripleo> we rely on the caching scheduler, please don't remove it! | |
| 14:31:35 | sean-k-mooney | jroll: at this point anything we have so much stuff using placement im not even sure you could deploy a fucntional nova with the caching schduler unless you never migrate or resize anything | |
| 14:32:20 | jroll | sean-k-mooney: if anyone can hack it, it's tripleo ;D | |
| 14:33:01 | jaypipes | sean-k-mooney: "that would make the deployment unsupported downstream" wait, what!? but I thought triple-o wasn't a RH product? :P | |
| 14:33:24 | sean-k-mooney | jaypipes: i was refering to OSP | |
| 14:33:40 | sean-k-mooney | e.g. osp being the downstream | |
| 14:33:40 | jaypipes | sean-k-mooney: ok dokey. :) | |
| 14:33:56 | sean-k-mooney | tripplo has too many names | |
| 14:34:12 | jaypipes | sean-k-mooney: and you just gave it another ;) "tripplo" | |
| 14:34:18 | sean-k-mooney | its tripleo, rdo director, osp director and apex | |
| 14:34:35 | jaypipes | sean-k-mooney: oh, you mean exactly what RH did with OpenShift? | |
| 14:34:59 | jaypipes | sean-k-mooney: or at least, *the current incarnation* of OpenShift. | |
| 14:36:00 | sean-k-mooney | Openshift is (kubernetes + extentions) and i think its called Openshift container platform or somthing as down stream product. i should proable know that... | |
| 14:36:35 | jaypipes | sean-k-mooney: yes, I'm referring to the evolving meaning of openshift(.com/org) over the last 5 years... | |
| 14:41:46 | mriedem | mgagne_: i hoped that's what you'd say, based on earlire discussions re heal_allocations | |
| 14:41:46 | mgagne_ | mriedem: I don't think the caching scheduler will be needed anymore if placement does the job and I can migrate to it easily. we tested that already I think | |
| 14:41:57 | mgagne_ | mriedem: if caching scheduler can still be used in a out of tree fashion, there is a plan B if anything goes wrong. | |
| 14:43:03 | mriedem | yes, that's possible | |
| 14:45:14 | sahid | mriedem, melwitt not sure of what you can put in the release note or config doc instead of don't mix rx/tx queue size option if you don't use guests with only virtio devices :/ | |
| 14:45:37 | sahid | i commented a think we could have a patch which is limiting the impact | |
| 14:49:40 | openstack | Launchpad bug 1788634 in OpenStack Compute (nova) "Noisy DEBUG log: Defaulting the value of the field 'numa_membw_values' to None in MetricPayload" [Low,Confirmed] - Assigned to Balazs Gibizer (balazs-gibizer) | |
| 14:49:40 | gibi | edmondsw, mriedem: bug about the noise debug log https://bugs.launchpad.net/nova/+bug/1788634 | |
| 14:49:42 | sean-k-mooney | sahid: well enduses can change the interface model by uploading an imagge and setting the vif_modle in the metatdat | |
| 14:50:09 | sean-k-mooney | sahid: so the nova code needs to handel that correctly and not set the queue lenght in that case | |
| 14:50:40 | edmondsw | gibi cool, tx | |
| 14:54:59 | sahid | sean-k-mooney: yes and only virtio model works with queues, so is user is setting something else, we don't want use queues | |
| 14:55:47 | sahid | did i miss something? | |
| 14:56:18 | sean-k-mooney | sahid: right, no just makeing sure that is the current behavior. i belive we check for that in the code but if we dont its a bug | |
| 14:57:10 | sahid | a even less impacting solution is to reproduce what is done https://github.com/openstack/nova/blob/34956bea4beb8e5ba474b42ba777eb88a5eadd76/nova/virt/libvirt/vif.py#L153 | |
| 14:57:22 | sahid | for rx/tx queue block | |
| 15:00:14 | kosamara | Continuing from the meeting on https://review.openstack.org/#/c/579897 | |
| 15:01:12 | jaypipes | stephenfin: you fixing up https://review.openstack.org/#/c/595802/? | |
| 15:01:27 | stephenfin | jaypipes: as we speak | |