| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-10-23 | |||
| 12:35:37 | sean-k-mooney | recyclehero: also people have proposed overcommiting PCPUs in the past but we have rejected it so its obvious but not to eveyone :) | |
| 12:36:13 | recyclehero | i have 1000 cpus, deal with it nova! :D | |
| 12:36:44 | sean-k-mooney | :) | |
| 12:37:08 | recyclehero | I didnt understnd the flavor part? how is it related with me not having allocation_ratio*PCPU=VCPUs? | |
| 12:37:36 | sean-k-mooney | i was wondering was this related to schduling but your actully asking about horizon right | |
| 12:37:51 | sean-k-mooney | horizon pull its infor from the hypervios api | |
| 12:38:28 | recyclehero | aha so if I even be successful with overcommiting via placement I dont get to see it in horizon | |
| 12:39:59 | sean-k-mooney | https://docs.openstack.org/api-ref/compute/?expanded=list-hypervisors-details-detail,show-hypervisor-statistics-detail#list-hypervisors-details | |
| 12:40:21 | sean-k-mooney | so it s either hitting the details endpoint or the stats endpoint | |
| 12:40:35 | sean-k-mooney | in both cases it just reports the vcpus i belive | |
| 12:41:11 | sean-k-mooney | that value is not multiplied by the allocation ratio | |
| 12:41:27 | recyclehero | so its actually PCPUs right? | |
| 12:41:40 | sean-k-mooney | well no PCPUs are a different thing | |
| 12:41:45 | recyclehero | The number of vcpu in this hypervisor. This does not take allocation ratios used for overcommit into account so there may be disparity between this and the used count. | |
| 12:42:33 | recyclehero | "VCPU = PCPU - reserverd" or am I completely wrong | |
| 12:42:39 | sean-k-mooney | it either the number of cpus on the host or it the number of cpu in the host defined by the vcpu_pin_set or cpu_shared_set | |
| 12:42:58 | sean-k-mooney | recyclehero: its not calulated that way | |
| 12:43:39 | sean-k-mooney | if you use cpu_dedicated_set and dont defien cpu_shared_set then vcpu=0 | |
| 12:44:33 | sean-k-mooney | if you use cpu_shared_set or vcpu_pin_set and dont use cpu_dedicated_set then pcpus are 0 | |
| 12:45:59 | recyclehero | sean-k-mooney: aha, but actually I dont care what I see in horzion. I have 8 cores which I am seeing in horizon. I need 128 cores ready for allocation. | |
| 12:46:01 | recyclehero | https://docs.openstack.org/placement/latest/user/index.html | |
| 12:46:03 | sean-k-mooney | recyclehero: stephenfin has a short blog on the topic https://that.guru/blog/cpu-resources-redux/ | |
| 12:46:42 | sean-k-mooney | recyclehero: by the way we normally don suggest setting allocation raitos over 4 | |
| 12:46:59 | sean-k-mooney | im planning to push a patch to change the inital allcoation ration to 4 next week | |
| 12:47:14 | sean-k-mooney | ill be bringin it up at the ptg | |
| 12:47:21 | sean-k-mooney | but back to your issue | |
| 12:47:28 | sean-k-mooney | can you check what is set in placment | |
| 12:47:33 | recyclehero | ok 4 will even do. but I will be rather be typing virt command with allocation ratio 1 | |
| 12:47:56 | recyclehero | how? api cli? | |
| 12:48:06 | sean-k-mooney | you can do it either way | |
| 12:48:16 | sean-k-mooney | there is an osc-placment plugin for the openstack client | |
| 12:48:23 | sean-k-mooney | so you can pip install that | |
| 12:48:56 | sean-k-mooney | openstack --os-placement-api-version 1.18 resource provider list | |
| 12:49:17 | sean-k-mooney | then openstack --os-placement-api-version 1.18 resource provider inventory show <uuid of host> VCPU | |
| 12:49:17 | recyclehero | openstack resource provider inventory show c8ac01e1-0b52-41f0-bd52-484070ea3aa9 VCPU | |
| 12:49:27 | recyclehero | allocation_ratio | 16.0 | |
| 12:49:36 | recyclehero | total | 8 | |
| 12:49:40 | sean-k-mooney | there you go | |
| 12:49:50 | noonedeadpunk | sean-k-mooney: oh, btw, if we're talking about this:) Is usage of initial_*_allocation_ratio can be considered as unified and recommended approach instead of *_allocation_ratio? | |
| 12:50:27 | sean-k-mooney | noonedeadpunk: it depends on if you want to manage it via config or api | |
| 12:50:54 | sean-k-mooney | noonedeadpunk: some operators like to commit there configuration to git so want to use *allocation_ratio | |
| 12:51:09 | sean-k-mooney | other want to manage it via the api so shoudl use initail_* | |
| 12:51:13 | sean-k-mooney | to set the defualt | |
| 12:51:29 | sean-k-mooney | that basically how you chose between them | |
| 12:51:38 | noonedeadpunk | sean-k-mooney: just of the docs it seems that the idea of initial was thet it can be stored in git but can be overriden in api? | |
| 12:51:54 | noonedeadpunk | ah, I guess I see what you mean... | |
| 12:52:14 | sean-k-mooney | noonedeadpunk: yes it will set the inital value in the config and then over write it | |
| 12:52:22 | recyclehero | noonedeadpunk: https://review.opendev.org/#/c/758029/ we have takeled with jrosser and came to this. there are 2 refrence to related documentaion part. | |
| 12:52:26 | sean-k-mooney | inital is only used when creating an inventory for the first time | |
| 12:52:29 | recyclehero | https://docs.openstack.org/nova/ussuri/admin/configuration/schedulers.html#usage-scenarios | |
| 12:52:50 | noonedeadpunk | sean-k-mooney: aha, ok, I missed that part I guess | |
| 12:52:55 | sean-k-mooney | *_allocation_ration take efect every time the resouce tracker periodic task runs | |
| 12:53:34 | noonedeadpunk | so like if initial_ is overriden during runtime, it won't be really apllied | |
| 12:53:44 | recyclehero | if resource_tracker.py dont found the host it will set its inital flag to true | |
| 12:53:47 | noonedeadpunk | well, not during runtime, but after host is discovered | |
| 12:54:56 | sean-k-mooney | ya if you calll the api to change a value on the inventory then inital will not have any effect on that | |
| 12:55:09 | sean-k-mooney | inital is just used when creating an inventory and never after that | |
| 12:55:20 | noonedeadpunk | sean-k-mooney: ok, cool, thanks! | |
| 12:55:23 | recyclehero | sean-k-mooney: will u be in half an hour? I feel you are the one who save me from this misery. | |
| 12:55:26 | recyclehero | here | |
| 12:55:41 | sean-k-mooney | ill be here yes | |
| 12:55:48 | recyclehero | great thanks | |
| 12:56:04 | noonedeadpunk | recyclehero: yeah, it was exactly about that patch as I was thinking to just replace cpu_allocation_ratio with initial_cpu_allocation_ratio but it doesn't make sense for this scenario | |
| 12:57:04 | sean-k-mooney | recyclehero: this is the spec by the way that added inital_* https://specs.openstack.org/openstack/nova-specs/specs/stein/implemented/initial-allocation-ratios.html | |
| 12:57:26 | sean-k-mooney | noonedeadpunk: idealy you would support both | |
| 12:57:46 | sean-k-mooney | although only one of the two ways is acutlly used in any one deployment | |
| 12:58:36 | Diavel | Hi guys, currently working on an app which consumes rabbitmq messages from nova-compute and I am wondering - Is there a reason why there is no notification sent on `UpdateMetadata` of an instance? | |
| 12:59:16 | noonedeadpunk | sean-k-mooney: well, deployers are able to set overrides and define initial ones. and this patch will remove regular ones. so kind of both supported but not in very straightforward way.... | |
| 12:59:16 | Diavel | there are for imo every other operation you can run from horizon, but there is no for metadata update | |
| 13:00:03 | sean-k-mooney | noonedeadpunk: ah i see well that makes sense i guess | |
| 13:00:34 | sean-k-mooney | noonedeadpunk: i still prefer how kolla ansible does it with config overrides rahter then provideing wrappers around things | |
| 13:01:15 | noonedeadpunk | sean-k-mooney: we encourage to use config overrides https://opendev.org/openstack/openstack-ansible-os_nova/src/branch/master/defaults/main.yml#L501 | |
| 13:01:24 | sean-k-mooney | Diavel: when you ser updateMetadata what do you mean | |
| 13:01:26 | noonedeadpunk | but I mean we should have some defaults set? | |
| 13:01:32 | sean-k-mooney | Diavel: is it the server proerties | |
| 13:01:43 | sean-k-mooney | Diavel: or image/flaovr metadata | |
| 13:02:00 | sean-k-mooney | noonedeadpunk: oh you have both | |
| 13:02:09 | noonedeadpunk | yep :p | |
| 13:02:37 | noonedeadpunk | and these vars can be overriden with config_overrides so overrides have prescedence | |
| 13:02:37 | sean-k-mooney | ya its better in the long run i think since the user become more familar with the upstream project names and there for the release note make more sense to them | |
| 13:04:10 | Diavel | sean-k-mooney: it's the server metadata, not flavour | |
| 13:04:16 | sean-k-mooney | Diavel: are you refering to this metadata https://github.com/openstack/nova/blob/master/nova/notifications/objects/instance.py#L58 | |
| 13:04:19 | sean-k-mooney | ah ok | |
| 13:04:39 | sean-k-mooney | well that is updated when you do an openstack server set --properties | |
| 13:04:46 | sean-k-mooney | which is not an instance action | |
| 13:05:26 | sean-k-mooney | it might be part of https://github.com/openstack/nova/blob/master/nova/notifications/objects/instance.py#L287 | |
| 13:05:49 | sean-k-mooney | we dont have the filed ther ebut that might be triggered form the openstack server set | |
| 13:06:18 | noonedeadpunk | btw, need to remove *_allocation_ratio from config (so that value would be None) is not super convinient from our prespective:))) being able to set it as just empty string to make it not used is more neat and I guess from python prespective is not harder in implementation? | |
| 13:06:46 | sean-k-mooney | Diavel: wa it think its the update https://github.com/openstack/nova/blob/master/nova/notifications/objects/instance.py#L590-L598 | |
| 13:07:07 | noonedeadpunk | as `if str()` and `if None` are both false.... | |
| 13:07:28 | noonedeadpunk | but whatever | |
| 13:08:35 | sean-k-mooney | Diavel: so i think you just need to addd it to the payload and extend https://github.com/openstack/nova/blob/6a5e158756edc6c01bb23c26dcfc4c3a3df703dc/nova/notifications/base.py#L170-L207 | |
| 13:08:39 | noonedeadpunk | ah, well, I guess it might be limitation of oslo.config... as it should be int, so can't be just empty | |
| 13:09:37 | sean-k-mooney | noonedeadpunk: i think you can set it to "cpu_allocation_ratio=" | |
| 13:10:06 | noonedeadpunk | oh, really? I thought we can't.... | |
| 13:10:32 | Diavel | sean-k-mooney: thanks a lot ;) will take a look at that part | |
| 13:10:54 | sean-k-mooney | noonedeadpunk: you can use 0.0 https://github.com/openstack/nova/blob/master/nova/conf/compute.py#L420-L453 | |
| 13:11:41 | sean-k-mooney | noonedeadpunk: but i think y ou can also set it to None | |
| 13:12:23 | noonedeadpunk | i thought that None in config is not None, but str('None')? | |
| 13:12:30 | noonedeadpunk | but yeah | |