| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-02-19 | |||
| 18:59:56 | sean-k-mooney | if so can you go up one level | |
| 19:00:01 | bauzas | ah no | |
| 19:00:07 | bauzas | i'm in the root repo | |
| 19:00:14 | sean-k-mooney | ya ok | |
| 19:00:24 | bauzas | (py37) [sbauza@sbauza nova]$ pwd | |
| 19:00:25 | bauzas | /home/sbauza/git/openstack/nova | |
| 19:00:30 | sean-k-mooney | ya | |
| 19:00:44 | sean-k-mooney | so it should not be a local import issue with your current directory | |
| 19:01:03 | bauzas | I recreated the venv so many times... | |
| 19:01:47 | sean-k-mooney | i never actully got food when i said i would so im going to go get it now. if you push the chagne somewhere ill try and run it when i get back | |
| 19:02:04 | bauzas | yeah I need to drop too | |
| 19:02:07 | bauzas | but I'm about to quite | |
| 19:02:08 | bauzas | quit | |
| 19:02:12 | bauzas | frankly | |
| 19:02:29 | bauzas | that's making me maaaad | |
| 19:02:34 | bauzas | anyway, good dinner | |
| 19:04:10 | sean-k-mooney | bauzas: i would just push hte patch let teh gate check it and clone a new copy of the repo in the morning | |
| 19:04:22 | sean-k-mooney | just too make sure that its clean | |
| 19:05:51 | sean-k-mooney | bauzas: oh you proably have a local file for opertunistic db testign configured | |
| 19:06:04 | sean-k-mooney | i forget how to configure that but there is a way | |
| 19:09:18 | sean-k-mooney | or maybe that is just a thing in sqlalcemey migrate https://opendev.org/x/sqlalchemy-migrate/src/branch/master/test_db.cfg | |
| 19:12:24 | openstackgerrit | Eric Fried proposed openstack/nova master: Follow-up: Add delete_on_termination to volume-attach API https://review.opendev.org/680655 | |
| 19:13:53 | sean-k-mooney | ya i guess its hard coded https://github.com/openstack/nova/blob/e20d529dbf656ddcacb7ab3497a6c71e89985f50/nova/tests/unit/db/test_migrations.py#L17-L33 | |
| 19:37:21 | efried | stephenfin: did those | |
| 20:27:41 | bauzas | sean-k-mooney: nope I didn't configured it | |
| 20:45:06 | openstackgerrit | Lee Yarwood proposed openstack/nova master: WIP libvirt: Provide the backing file format when creating qcow2 disks https://review.opendev.org/708745 | |
| 20:53:47 | bauzas | sean-k-mooney: mmm, cloning nova on another local repo seems to work | |
| 20:54:14 | sean-k-mooney | bauzas: ya you must have a local file or something in that copy which is messign it up | |
| 20:54:25 | bauzas | possibly | |
| 20:54:32 | openstackgerrit | Lee Yarwood proposed openstack/nova master: WIP libvirt: Provide the backing file format when creating qcow2 disks https://review.opendev.org/708745 | |
| 21:04:08 | spatel | sean-k-mooney: do you thing ksmd is good for openstack ? its using 40% cpu on my compute node. | |
| 21:04:49 | sean-k-mooney | ksmd as in kernel shared memory deamon? | |
| 21:05:12 | spatel | I did and some folks saying disable for compute node | |
| 21:05:38 | sean-k-mooney | yes i would proably disable it also | |
| 21:06:02 | spatel | I will run my load-test and see if any impact. | |
| 21:06:04 | sean-k-mooney | im just checking that its what i think it is | |
| 21:06:32 | spatel | we found why erlang doesn't like two NUMA nodes. | |
| 21:07:18 | spatel | erlang scheduler default migrate process from CPU to CPU and that is why we were getting performance hit | |
| 21:07:32 | sean-k-mooney | ya so assuming its related to https://www.kernel.org/doc/Documentation/vm/ksm.txt which i think it is we recommend disable ksm on all nfv systems | |
| 21:07:54 | spatel | I will disable KSM | |
| 21:08:41 | spatel | sean-k-mooney: where i should change numa_policy to preferred ? (default is lagacy) | |
| 21:08:54 | spatel | does flavor support that feature on stein? | |
| 21:10:38 | sean-k-mooney | spatel: that has never been suported | |
| 21:11:04 | spatel | so i have only option to set it up in /etc/nova/nova.conf? | |
| 21:11:22 | sean-k-mooney | which numa policy are you refering to specificly | |
| 21:11:42 | spatel | https://specs.openstack.org/openstack/nova-specs/specs/queens/implemented/share-pci-between-numa-nodes.html | |
| 21:12:13 | spatel | I want to spin up VM across numa (default sriov won't let you do that) | |
| 21:12:16 | sean-k-mooney | ah the pci polices | |
| 21:12:22 | spatel | Yes | |
| 21:12:43 | sean-k-mooney | so you can set that in the config and i recently added it to the flavor | |
| 21:12:53 | sean-k-mooney | i need to check if it merged in train but i think it did | |
| 21:13:02 | spatel | but does it available in stein ? | |
| 21:13:22 | spatel | I am running stein (if its not available in stein then only option is nova.conf right?) | |
| 21:13:30 | sean-k-mooney | actully no it only merged in ussuri https://specs.openstack.org/openstack/nova-specs/specs/ussuri/approved/vm-scoped-sriov-numa-affinity.html | |
| 21:13:48 | sean-k-mooney | before ussuri you can only set it in the nova.conf for pci alias | |
| 21:13:58 | sean-k-mooney | meaning it does not work with neutron sriov ports | |
| 21:14:00 | spatel | nova-api service or compute? | |
| 21:14:09 | sean-k-mooney | you have to set it in both | |
| 21:14:17 | spatel | i will give it a try.. | |
| 21:14:43 | sean-k-mooney | it will not work if you are using neutron sriov port e.g. vnic_type=direct | |
| 21:15:10 | sean-k-mooney | it will only be applied to pci devices that are requested vai an alias in the flavor | |
| 21:15:29 | sean-k-mooney | neutron sriov only supports legacy before ussuri | |
| 21:18:17 | spatel | sean-k-mooney: ohhhh! | |
| 21:18:28 | spatel | i am using vnic_type=direct | |
| 21:18:58 | sean-k-mooney | yes the abiltiy to chagne that was only added in ussuri upstream | |
| 21:19:23 | sean-k-mooney | i have backported it downstream but this is not a candiate for backporting upstream | |
| 21:21:13 | sean-k-mooney | spatel: this https://review.opendev.org/#/c/674072/ is the patch you want | |
| 21:23:51 | spatel | any plan that patch get merge in future release? | |
| 21:24:20 | sean-k-mooney | its merged in master so it will be released in ussuri | |
| 21:24:41 | sean-k-mooney | but its adding a new feature so it cant be backported upstream | |
| 21:24:57 | spatel | sean-k-mooney: totally understand.. | |
| 21:25:32 | sean-k-mooney | downstream we are aboult to do a partial backport with out the support for the image property or notification as we cant backport object changes | |
| 21:25:43 | sean-k-mooney | *able | |
| 21:25:56 | sean-k-mooney | that said we avoid it when we can | |
| 21:26:59 | spatel | make sense to avoid thatif there are lots of moving part | |
| 21:27:20 | spatel | http://openstack-in-production.blogspot.com/2015/08/ept-and-ksm-for-high-throughput.html | |
| 21:27:31 | spatel | What do you think about EPT ? | |
| 21:27:45 | sean-k-mooney | you should alway always have it enabeld | |
| 21:28:15 | spatel | CERN suggesting to disable to gain performance | |
| 21:28:16 | sean-k-mooney | if you turn of extended page tables it almost as bad as disbaling kvm and falling back to qemu | |
| 21:28:58 | spatel | hmm! interesting. | |
| 21:29:12 | sean-k-mooney | our perfomance team messurad a huge performance hit to disabeling for the specter/meltdown mitigations | |
| 21:29:49 | spatel | I have that disabled on compute machine but not on VM. | |
| 21:33:25 | sean-k-mooney | spatel: read https://www.redhat.com/pt-br/blog/understanding-l1-terminal-fault-aka-foreshadow-what-you-need-know specifcally the "Beyond bare metal" section alth the full blog is good too | |
| 21:33:43 | spatel | reading... | |
| 21:42:54 | sean-k-mooney | if you turn of EPT you fallback to shadowtables which will result in significat qemu overhead in the 10s or % range based on there testing when fixing l1tf | |
| 21:43:37 | spatel | Yes, you are right. (also based on your link) | |
| 22:40:42 | efried | dansmith: results for the Enum thing are back. Func passed; one unit failed but I'm not sure if it means we need a migration; and grenade failed but I can't tell if it's related. https://review.opendev.org/#/c/708707/ | |
| 23:34:54 | openstackgerrit | Eric Fried proposed openstack/nova master: Use stock mock https://review.opendev.org/708262 | |
| #openstack-nova - 2020-02-20 | |||
| 00:09:58 | openstackgerrit | Eric Fried proposed openstack/nova master: hacking: force explicit import of python's mock https://review.opendev.org/708768 | |
| 00:18:28 | efried | johnthetubaguy: Are you interested in taking over the coding work for vTPM? | |
| 00:18:51 | efried | The design is pretty well described at this point, and the base enablement code is very close to done. | |
| 00:23:16 | efried | stephenfin, gibi, bauzas, alex_xu, johnthetubaguy (and melwitt, dansmith FYI, though I don't expect you to be there so early): In Thursday's Nova meeting, I would like to consider the two spec freeze exceptions that have been requested. | |
| 00:23:17 | efried | They're the first two on this etherpad: https://etherpad.openstack.org/p/nova-ussuri-planning | |
| 00:23:17 | efried | Also, please have a look at the specs and see if you think they're close enough to go. | |
| 00:23:17 | efried | Please attend the meeting if you can. | |
| 00:38:43 | openstackgerrit | Merged openstack/nova master: Follow-up: Add delete_on_termination to volume-attach API https://review.opendev.org/680655 | |
| 01:05:30 | sean-k-mooney | efried: dansmith while i was fixing the cyborg devstack plugin config generation i decied to fix a few other things too. | |
| 01:05:59 | sean-k-mooney | i have one more patch to write to automate the installation of the cyborg python client | |
| 01:06:20 | sean-k-mooney | which ill do tomorrow and then i can add the DNM nova multinode testing job | |