Earlier  
Posted Nick Remark
#openstack-nova - 2020-02-05
09:55:26 openstackgerrit Brin Zhang proposed openstack/nova master: Store instance action event exc_val fault details https://review.opendev.org/694428
10:36:29 openstackgerrit Qiu Fossen proposed openstack/nova-specs master: specify mac for creating instance https://review.opendev.org/700429
10:37:00 openstackgerrit Merged openstack/nova master: Minor improvements to cell commands https://review.opendev.org/698053
11:07:26 stephenfin gibi: Could you revisit https://review.opendev.org/#/c/705760/ ? ralonsoh spotted some paths I'd missed
11:07:42 stephenfin and dansmith had comments but I've addressed them, afaict
11:12:44 sandonov openstack-nova
11:35:19 openstackgerrit Brin Zhang proposed openstack/nova-specs master: FUP: Fixed the invalid index in References https://review.opendev.org/705933
11:52:17 openstackgerrit Stephen Finucane proposed openstack/nova master: trivial: Remove unused 'cache_utils' APIs https://review.opendev.org/705652
11:52:18 openstackgerrit Stephen Finucane proposed openstack/nova master: trivial: Address TODO https://review.opendev.org/705653
11:52:18 openstackgerrit Stephen Finucane proposed openstack/nova master: trivial: Bump minimum version of websockify https://review.opendev.org/705654
11:52:19 openstackgerrit Stephen Finucane proposed openstack/nova master: trivial: Merge unnecessary 'NovaProxyRequestHandlerBase' separation https://review.opendev.org/705655
11:52:19 openstackgerrit Stephen Finucane proposed openstack/nova master: trivial: Remove 'run_once' helper https://review.opendev.org/705656
11:52:20 openstackgerrit Stephen Finucane proposed openstack/nova master: tox: Integrate mypy https://review.opendev.org/676208
11:52:20 openstackgerrit Stephen Finucane proposed openstack/nova master: mypy: Add type annotations to 'nova.pci' https://review.opendev.org/676209
11:52:21 openstackgerrit Stephen Finucane proposed openstack/nova master: mypy: Add nova.cmd, nova.conf, nova.console https://review.opendev.org/705657
11:52:21 openstackgerrit Stephen Finucane proposed openstack/nova master: WIP: mypy: Add type annotations to top-level modules https://review.opendev.org/705658
12:03:44 kashyap stephenfin: Hiya; I've addressed your questions, and added additional useful bits where needed: https://review.opendev.org/#/c/693844/
12:25:17 openstackgerrit Merged openstack/nova master: nova-net: Remove use of legacy 'SecurityGroup' object https://review.opendev.org/697155
12:36:31 mriosfer Hi, We got Openstack Queens RDO. Some customers are telling us that their Windows VM CPU performs bad as a hell. We deployed the lasted virtio already to check improvements. I got a video an its true work its frustating, I/O is less than <1ms
12:40:34 sean-k-mooney what are you deploying on libvirt/kvm
12:41:42 sean-k-mooney the first thing i woudl do is alter the cpu topology to be sane for windows
12:41:56 sean-k-mooney by defualt we allcoate 1 cpu socket per favor.vcpu
12:42:34 mriosfer we're ussing quemu-kvm for instances .
12:42:41 sean-k-mooney if the guest dont have a numa toplogy set hw:cpu_sockets=1 and hw:cpu:threads=2
12:42:43 mriosfer How do you do that?
12:42:59 sean-k-mooney assuming the host has Hyper thread enabled.
12:43:04 mriosfer yes of course
12:43:19 mriosfer do you setup that extra features in the flavors?
12:43:35 sean-k-mooney yes you can do it in the flavor or image
12:43:50 sean-k-mooney in the image its hw_ instead of hw:
12:44:00 mriosfer i prefer image
12:44:31 sean-k-mooney that should help the windows shcduler if you have more then 4 vcpu in the vm
12:44:45 mriosfer yes.. some cpu got 8cpu or 12vcpu
12:44:57 mriosfer and with terminal server the performance its terrible
12:45:07 sean-k-mooney ya so windows will see that as 8 to 12 cpu sockets with out default setting
12:45:30 sean-k-mooney the next thing to look at would be the graphics device model
12:45:38 sean-k-mooney you said you were on queens
12:45:41 mriosfer yes
12:45:59 mriosfer our nodes dont have pcie addtional graphic card just the integrated one by Dell
12:46:09 sean-k-mooney ya that is ok
12:46:44 sean-k-mooney on train+ would suggest using the virtio gpu but on queens you should set hw_video_model=qxl
12:47:18 sean-k-mooney i belive qxl is the best performing graphic device that qemu can emulate prior to support for virtio
12:47:28 sean-k-mooney https://github.com/openstack/glance/blob/master/etc/metadefs/compute-libvirt-image.json#L71-L84
12:47:47 sean-k-mooney you could also use the hw_video_ram element to increase that
12:47:57 sean-k-mooney i belive it default to 64mb
12:48:07 mriosfer thath ram uses the instance ram no?
12:48:13 sean-k-mooney no
12:48:16 sean-k-mooney its addtional host ram
12:48:24 mriosfer oh , ok
12:48:41 sean-k-mooney it can only be set on the image if the flavor allows it by setting hw_video:ram_max_mb
12:49:25 sean-k-mooney i would start with the qxl video model and then if that is not enough after the cpu changes then increae the ram to say 128 or 256
12:49:46 mriosfer sean do you know why error? https://gyazo.com/10156259f833a20decbdce821ef8e31b
12:50:06 sean-k-mooney yes
12:50:18 sean-k-mooney this is one of the model that is in the metadefs
12:50:24 sean-k-mooney so there is a drop down for it
12:50:32 sean-k-mooney use the filter box above
12:50:37 sean-k-mooney instead of a custom filed
12:51:21 sean-k-mooney our you can scoll down there should be a libvirt dropdown in the availabel metadata
12:51:47 mriosfer oh
12:51:56 mriosfer im fuck*** noob
12:52:09 sean-k-mooney the gui is "trying" to be help full
12:52:26 sean-k-mooney it actully really annoyed me when they added that "feature"]
12:52:46 mriosfer rebooting instance will apply the new changes?
12:52:52 sean-k-mooney no
12:52:55 sean-k-mooney unfortunetly not
12:53:05 sean-k-mooney you can only update the instnace with a rebuild
12:53:13 sean-k-mooney which would loose data on the root disk
12:53:18 mriosfer oh fuck
12:54:03 sean-k-mooney we make a copy of the data in the vm when we boot it so that it intentionall will not be affected by change in glance
12:54:19 sean-k-mooney if we did not you could break all your deployed vms with a typo
12:54:24 sean-k-mooney which would be bad
12:54:49 sean-k-mooney in your case once you are happy with the final image and its performace
12:55:02 mriosfer mmm maybe i can do a workarround at ceph level :)
12:55:04 sean-k-mooney you could do a db update to add the missing keys to the copy
12:55:42 sean-k-mooney then a hard reboot of the vm would fix it
12:56:20 sean-k-mooney i belive we store the instance image keys in the system_metadata table prefixed with img_
12:56:28 mriosfer now looks much better the image : https://gyazo.com/cacf129fc79e2f1b23b607704b645d71
12:56:59 sean-k-mooney yes but you will need to update the flaovr to have the hw_video_ram work
12:58:20 sean-k-mooney the reason being we did not want tenant to be abel to upload there own image and request 256GB of video ram
12:59:08 mriosfer do you also setup the Max vCPU threats and vCpu Cores at flavor level?
12:59:39 sean-k-mooney you dont need too but can
12:59:49 sean-k-mooney for the video ram it will hit this code https://github.com/openstack/nova/blob/a948a803b561606a892133940915caae610c080b/nova/virt/vmwareapi/vmops.py#L334-L338
13:00:16 mriosfer that just for vmware no?
13:00:32 sean-k-mooney oh wrong file let me check the libvirt driver
13:00:43 sean-k-mooney vmware would reject it if the flavor is not set
13:01:44 sean-k-mooney ya its the same https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L5057-L5063
13:01:57 sean-k-mooney i would just lave that out for now
13:02:21 sean-k-mooney as you would also have to update the embeded copy of the flavor or do a resize to pick up the change
13:02:56 mriosfer yes i will do for sure
13:02:58 sean-k-mooney if you find qxl and the toplogy chagne are not enough the try the ram change
13:03:14 mriosfer if it works i will send you a pack a beers!
13:03:59 sean-k-mooney ha :) well on other question is the slowness you are finding with the horizon console or do you still see that with rdp in the guest
13:04:13 mriosfer i see in the rdp guest
13:04:25 stephenfin bauzas: Assuming efried and dansmith aren't online yet, I think you ought to look at this and send it through so we can unblock neutron
13:04:26 mriosfer i dont recomend users work through horizon
13:04:29 sean-k-mooney right ok then ya this shoudl help
13:04:42 stephenfin We can use a follow-up if dansmith isn't happy with any of my answers
13:05:03 sean-k-mooney i have seen issue with novnc adding lag in the past
13:05:05 bauzas stephenfin: which change do you want me ?
13:05:12 stephenfin bauzas: oh, whoops https://review.opendev.org/#/c/705760

Earlier   Later