| *** gyee has quit IRC | 01:21 | |
| *** spatel_ has joined #openstack-ansible | 02:17 | |
| *** spatel_ is now known as spatel | 02:17 | |
| *** spatel has quit IRC | 02:19 | |
| *** evrardjp has quit IRC | 02:33 | |
| *** evrardjp has joined #openstack-ansible | 02:33 | |
| *** macz_ has joined #openstack-ansible | 03:32 | |
| *** macz_ has quit IRC | 03:37 | |
| *** cyberpear has quit IRC | 04:06 | |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible-galera_server master: Revert "Update mariadb version to 10.5.10" https://review.opendev.org/c/openstack/openstack-ansible-galera_server/+/791107 | 04:15 |
|---|---|---|
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible master: Bump ansible-base to 2.10.9 https://review.opendev.org/c/openstack/openstack-ansible/+/791293 | 04:16 |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible master: Add option to remove group from inventory https://review.opendev.org/c/openstack/openstack-ansible/+/791277 | 04:17 |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible master: Change order of swift and gnocchi installation https://review.opendev.org/c/openstack/openstack-ansible/+/791261 | 04:18 |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible-os_gnocchi master: Switch gnocchi service name to service https://review.opendev.org/c/openstack/openstack-ansible-os_gnocchi/+/791254 | 04:22 |
| *** macz_ has joined #openstack-ansible | 05:33 | |
| *** macz_ has quit IRC | 05:38 | |
| noonedeadpunk | jrosser: with depends-on revert maria bump things do not fail anymore | 06:34 |
| *** klamath_atx has joined #openstack-ansible | 06:39 | |
| jrosser | well we should merge that revert ASAP | 06:43 |
| *** andrewbonney has joined #openstack-ansible | 07:13 | |
| *** partlycloudy has quit IRC | 07:23 | |
| *** partlycloudy has joined #openstack-ansible | 07:26 | |
| *** macz_ has joined #openstack-ansible | 07:34 | |
| noonedeadpunk | I think I will submit a bug | 07:35 |
| noonedeadpunk | hm, well, now I still see issues even with depends on, ie for adjutant... https://review.opendev.org/c/openstack/openstack-ansible-os_adjutant/+/777607/ | 07:36 |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible-os_adjutant master: Install mysql client libraries https://review.opendev.org/c/openstack/openstack-ansible-os_adjutant/+/777607 | 07:36 |
| noonedeadpunk | oh, it's another depends on :) | 07:37 |
| *** macz_ has quit IRC | 07:38 | |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible master: Decrease manila tempest coverage https://review.opendev.org/c/openstack/openstack-ansible/+/791202 | 07:39 |
| *** tosky has joined #openstack-ansible | 07:47 | |
| *** sakharkar has joined #openstack-ansible | 07:57 | |
| sakharkar | Is Openstack Ansible Victoria supports all endpoints on SSL? | 07:58 |
| *** sshnaidm|afk is now known as sshnaidm|pto | 08:00 | |
| noonedeadpunk | sakharkar: yep, it does | 08:06 |
| noonedeadpunk | there's haproxy_ssl_all_vips variable, and you can set openstack_service_adminuri_proto and openstack_service_internaluri_proto to https | 08:08 |
| *** masterpe has quit IRC | 08:37 | |
| *** masterpe has joined #openstack-ansible | 08:40 | |
| openstackgerrit | Merged openstack/openstack-ansible-galera_server master: Revert "Update mariadb version to 10.5.10" https://review.opendev.org/c/openstack/openstack-ansible-galera_server/+/791107 | 08:52 |
| *** jcath has joined #openstack-ansible | 09:16 | |
| jcath | hello~~~ jrosser | 09:17 |
| *** jcath has quit IRC | 09:20 | |
| *** jcath has joined #openstack-ansible | 09:21 | |
| *** prometheanfire has quit IRC | 09:24 | |
| *** prometheanfire has joined #openstack-ansible | 09:24 | |
| *** jcath has quit IRC | 09:27 | |
| *** macz_ has joined #openstack-ansible | 09:35 | |
| sakharkar | noonedeadpunk: I tried the deployment and is failing at setup openstack.yaml while creating database for keystone, | 09:39 |
| *** macz_ has quit IRC | 09:40 | |
| sakharkar | noonedeadpunk: Deployment fail error : http://paste.openstack.org/show/805378/ | 09:44 |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible master: Bump ansible-base to 2.10.9 https://review.opendev.org/c/openstack/openstack-ansible/+/791293 | 09:44 |
| noonedeadpunk | sakharkar: I think it's not related to SSL and I believe your setup-infrastructure.yml has failed somewhere first time you ran it? | 09:45 |
| noonedeadpunk | the thing here is that your utility container was not setup properly. So I'd suggest re-running utility-install.yml (maybe with -e venv_rebuild=true in case it's broken) | 09:47 |
| sakharkar | noonedeadpunk: My user_variable.yaml looks like this http://paste.openstack.org/show/805379/ | 09:48 |
| noonedeadpunk | I'm not sure it's related atm | 10:05 |
| *** macz_ has joined #openstack-ansible | 11:36 | |
| *** macz_ has quit IRC | 11:40 | |
| *** jawad_axd has joined #openstack-ansible | 12:12 | |
| *** mgariepy has quit IRC | 12:21 | |
| *** mgariepy has joined #openstack-ansible | 12:35 | |
| *** jawad_axd has quit IRC | 12:36 | |
| *** jawad_axd has joined #openstack-ansible | 12:36 | |
| *** spatel_ has joined #openstack-ansible | 13:15 | |
| *** spatel_ is now known as spatel | 13:15 | |
| spatel | Folks, any idea, last night we hit by large DDoS and that isolate my galera cluster so now one of node saying this - ERROR 1047 (08S01): WSREP has not yet prepared node for application use | 13:17 |
| spatel | Google saying this is what happened when your node isolate.. | 13:17 |
| *** macz_ has joined #openstack-ansible | 13:37 | |
| *** d34dh0r53 has joined #openstack-ansible | 13:38 | |
| openstackgerrit | Dmitriy Rabotyagov proposed openstack/openstack-ansible-os_neutron master: Implement uWSGI for neutron-api https://review.opendev.org/c/openstack/openstack-ansible-os_neutron/+/486156 | 13:39 |
| *** macz_ has quit IRC | 13:41 | |
| openstackgerrit | Merged openstack/openstack-ansible-os_nova master: setup.cfg: Replace dashes with underscores https://review.opendev.org/c/openstack/openstack-ansible-os_nova/+/788892 | 13:48 |
| noonedeadpunk | and connection for all ports is open between them? | 13:50 |
| *** cyberpear has joined #openstack-ansible | 13:57 | |
| *** jawad_axd has quit IRC | 13:58 | |
| admin0 | spatel, as long as any one of your node is good.. galera can be recovered and fixed | 14:13 |
| admin0 | but you have to do some stuff manually also | 14:14 |
| spatel | admin0 thanks after reboot node-1 (bad one) fix issue. | 14:15 |
| spatel | I just reboot lxc container i meant | 14:15 |
| noonedeadpunk | oh, I expected you did that at first place:) | 14:25 |
| noonedeadpunk | eventually on Train I catched bug where galera didn't want to re-sync, when I decided to re-create one of the containers. It was keep saying that this new clean container is up to date and synced for some reason... | 14:26 |
| noonedeadpunk | (believe that some sort of bug) | 14:26 |
| openstackgerrit | Merged openstack/openstack-ansible stable/train: Prepare Train to EM https://review.opendev.org/c/openstack/openstack-ansible/+/790655 | 14:40 |
| noonedeadpunk | Folks ,did anybody used Cyborg for managing GPUs or other sr-iov devices? | 14:42 |
| jrosser | I only ever did sriov with nova/neutron, did pci pass through with nova and also vgpu with nova | 14:51 |
| jrosser | feels like a lot of crossover between nova and cyborg and not really obvious where it’s all heading | 14:52 |
| noonedeadpunk | oh, so you can just sr-iov gpu? as in nova doc they're talking about nics only, and saw https://indico.cern.ch/event/776411/contributions/3345183/attachments/1851624/3039917/02_-_vGPUs_with_OpenStack_-_Accelerating_Science.pdf where has been said it's not implemented (no idea when it was) | 14:53 |
| noonedeadpunk | I mean, nvidia and amd does vgpu pretty differently | 14:53 |
| noonedeadpunk | while nvidia tesla (which require licensing) does fully virtualize gpus with their driver, AMD jsut create SR_IOV devices | 14:54 |
| jrosser | I did nvidia vgpu pretty straightforward with nova | 14:54 |
| jrosser | but yes licensing (inside the VM doh!) is needed | 14:54 |
| noonedeadpunk | yeah, I actully was reading Cyborg specs for half a day and still have no idea why it's needed. as looks like nova wrapper of the days where there was no proper placement features... | 14:55 |
| jrosser | I think there are cases for generic accelerators which need firmware uploading to have a particular function | 14:56 |
| jrosser | like fpga cards, that seems to be what I understood cyborg for | 14:56 |
| jrosser | but I see they add smartnic also and then I’m confused | 14:57 |
| noonedeadpunk | I see. I was kind of thinking about AMD as well, as because of SR-IOV you can place them separatelly from servers in rack and just do pci-e over ethernet (or smth like that) | 14:57 |
| noonedeadpunk | well, at day 1 presentation they were talking about sr-iov drives like nic, gpus... and I got confused there | 14:58 |
| *** macz_ has joined #openstack-ansible | 14:58 | |
| noonedeadpunk | and well, for W they added nvidia vgpu support https://specs.openstack.org/openstack/cyborg-specs/specs/wallaby/approved/vgpu-driver-proposal.html | 14:58 |
| noonedeadpunk | so returning back to vgups, I hoped to be able to have gpu on every compute, since gpu stack will be standalone, and accessible through net | 15:00 |
| *** spatel has quit IRC | 15:00 | |
| noonedeadpunk | but I think I won't know for sure if it's possible to make this work without some POC env, which be super costy... | 15:01 |
| jrosser | you can prototype with T4 for nvidia | 15:01 |
| jrosser | that does all the vgpu stuff | 15:01 |
| noonedeadpunk | well, I know that nvidia will work with their vgpu implementation | 15:02 |
| jrosser | still $$$$ but less than $$$$$$ | 15:02 |
| noonedeadpunk | I'm not sure about AMD :) | 15:02 |
| noonedeadpunk | they have pretty different concept of vgpus https://www.amd.com/en/graphics/workstation-virtual-graphics | 15:03 |
| jrosser | one of the subtle things is now the gpu gets divided, if you want multiple vm sharing a gpu | 15:03 |
| jrosser | *how | 15:03 |
| *** spatel_ has joined #openstack-ansible | 15:03 | |
| noonedeadpunk | well, yes, with SR-IOV you can't do qos, and protect from noisy neighbours | 15:03 |
| *** spatel_ is now known as spatel | 15:03 | |
| noonedeadpunk | while nvidia with full virtualisation solves that | 15:04 |
| spatel | noonedeadpunk thanks for that info.. my openstack is queens | 15:04 |
| jrosser | for nvidia the operator has to decide in nova config “divide this into quarter GPU” or whatever | 15:05 |
| jrosser | and it’s very static setup, even though the underlying nvidia stuff does not have that limitation any more | 15:05 |
| noonedeadpunk | but I think you can pass more like several cores into instance? | 15:05 |
| jrosser | would need to check that, not sure | 15:06 |
| noonedeadpunk | it should be defined with flavor isn't it? | 15:06 |
| spatel | sriov is technology it can apply anywhere, like gpu, disk IO, nic etc.. i don't think its only for NIC | 15:07 |
| noonedeadpunk | `openstack flavor set vgpu_1 --property "resources:VGPU=1"` | 15:07 |
| jrosser | really it depends on the workload, if everyone wants the same allocation it’s all very simple | 15:08 |
| noonedeadpunk | not scenario of public cloud :) | 15:08 |
| jrosser | but if you want to offer very small gpu instances for developers and a farm of the largest possible instance for rendering or something, then you have to think really hard about how you allocate those sizes to the physical gpu | 15:09 |
| noonedeadpunk | yeah, agree | 15:10 |
| noonedeadpunk | from other side with native sr-iov like amd do - everybody can use like 100% of gpu... | 15:11 |
| noonedeadpunk | and all others are struggling | 15:12 |
| jrosser | yeah, a bit unrelated but I got some U.2 nvme format pcie video encoders coming | 15:13 |
| jrosser | the go in where nvme drives would normally go and will need pci passthrough | 15:13 |
| noonedeadpunk | and still no-use for Cyborg? what a pity :) | 15:14 |
| jrosser | really interesting to see general purpose pci devices turning up in disk drive form factor | 15:14 |
| noonedeadpunk | yeah, that's true actually | 15:14 |
| noonedeadpunk | and I think I like that, because pci is super limited in amount, when you might want to have raid controller or proper networking card... | 15:15 |
| jrosser | oh well AMD server :) | 15:15 |
| noonedeadpunk | yeah, we're also about to switch to these :) | 15:16 |
| noonedeadpunk | But I think I will go nvidia way as well. At least that's know to work (and even had some experience with that) | 15:19 |
| *** spatel has quit IRC | 15:38 | |
| openstackgerrit | Merged openstack/openstack-ansible-haproxy_server master: Use integrated tests for haproxy_server https://review.opendev.org/c/openstack/openstack-ansible-haproxy_server/+/790090 | 15:51 |
| *** macz_ has quit IRC | 16:12 | |
| *** gyee has joined #openstack-ansible | 16:29 | |
| *** spatel_ has joined #openstack-ansible | 16:38 | |
| *** spatel_ is now known as spatel | 16:38 | |
| *** rh-jlabarre has quit IRC | 17:16 | |
| *** andrewbonney has quit IRC | 17:41 | |
| spatel | I know i asked this question before but again i would like to ask if someone has better answer, How do you guys do capacity planning in private or public cloud? | 17:58 |
| spatel | my management folks like to know these number for sales team and for other reason also | 17:59 |
| *** macz_ has joined #openstack-ansible | 21:03 | |
| *** macz_ has quit IRC | 21:07 | |
| *** macz_ has joined #openstack-ansible | 21:08 | |
| *** spatel has quit IRC | 21:35 | |
| *** macz_ has quit IRC | 21:45 | |
| *** macz_ has joined #openstack-ansible | 21:49 | |
| *** dpawlik has quit IRC | 22:45 | |
| *** dpawlik7 has joined #openstack-ansible | 22:53 | |
| *** tosky has quit IRC | 23:11 | |
| *** macz_ has quit IRC | 23:14 | |
Generated by irclog2html.py 2.17.2 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!