Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-02-26
15:14:50 spatel sean-k-mooney: do you have experience with cellv2 deployment?
15:14:50 sean-k-mooney spatel: o/
15:15:17 sean-k-mooney spatel: dansmith is the expert but a little
15:15:35 spatel i have 150 new compute nodes waiting ready to go and i am just trying to think i of how to scale my cloud
15:15:43 sean-k-mooney spatel: mainly due to the fact all deployments are cellv2 deplpyment of one cell
15:16:16 sean-k-mooney spatel: you are considering if you should add another cell or not
15:16:22 spatel How difficult its going to be in running production cloud migrate over Cellv2?
15:16:57 spatel currently i have 200 compute nodes in production (in-used) 150 in racks waiting to join cloud
15:17:29 sean-k-mooney what release are you running again?
15:17:31 spatel soon my cloud will be 350 ( I think my current controller can handle that but at some point i have to think big )
15:17:37 spatel queens
15:19:22 sean-k-mooney spatel: you can handel 500-1000 nodes in a singel cell without too much issue the rabbitmq message bus will be the first bottelneck in the system you will hit. cells of 200 nodes would be fine but you could easily run the 350 nodes you plan to have in the short term in one cell
15:20:04 spatel hmm! currently my rabbitmq rate is 250 mesg/s around :)
15:20:40 dansmith mriedem: stephenfin: wanna hit this again or should I just +W for you? https://review.openstack.org/#/c/638173
15:21:21 sean-k-mooney spatel: you do not have a dynmaic cloud e.g. you dont spawn/kill/migrate instnaces requently for you voip usecases if i understand correctly
15:21:42 sean-k-mooney spatel: you are also not using celomiter so you have less pressure on the message bus
15:21:49 spatel Yes, you are right
15:22:04 sean-k-mooney spatel: in a public cloud env the message bus trafffic is higher
15:22:06 spatel This is my private cloud for my customer.. very few activity so far
15:22:32 spatel I am not using celomiter :)
15:22:49 sean-k-mooney spatel: i think we swapped over to cellsv2 by default in pike
15:22:54 sean-k-mooney dansmith: is ^ right
15:23:20 sean-k-mooney spatel: dan had a talk on this a while back https://www.youtube.com/watch?v=o7jkqzORij8
15:23:21 stephenfin dansmith: Done, but I'd have been happy with you fast approving too
15:23:24 dansmith sean-k-mooney: I don't really remember, but he's on queens anyway
15:23:40 dansmith stephenfin: I figured, but I had non-grammar comments, so was just trying to be polite :)
15:23:41 cfriesen mdbooth: I made the changes you suggested for https://review.openstack.org/#/c/616692/ I like the idea of adding proper unit tests for convert_image(), but maybe we can do that in T?
15:24:16 spatel Yes i am on Queens
15:24:19 mriedem i'm going to mark https://blueprints.launchpad.net/nova/+spec/handling-down-cell as complete
15:24:25 dansmith mriedem: rock on
15:24:34 sean-k-mooney spatel: so on quees you are already running cellsv2
15:24:43 mriedem also just figured out my volume attachment tracking issue with the cross-cell resize series and have it fixed, so that's nice
15:24:54 sean-k-mooney so your coices are to expand your current cell or add a new cell for the new nodes
15:25:32 spatel sean-k-mooney: when you say i am already running on cellsv2 ( what does that means, do i need to enable or made some change to create cellsv2? )
15:26:03 cfriesen mriedem: is the neutron-grenade-multinode failure here expected? https://review.openstack.org/#/c/616692/
15:26:16 stephenfin dansmith: Appreciated :)
15:26:53 mriedem http://logs.openstack.org/92/616692/20/check/neutron-grenade-multinode/ac5746b/logs/grenade.sh.txt.gz#_2019-02-26_00_36_45_216
15:26:59 spatel Watching youtube... dansmith, looking good you was in Boston man!! :) I am in boston..
15:27:40 mriedem cfriesen: known issue http://status.openstack.org/elastic-recheck/#1463631
15:27:49 sean-k-mooney spatel: all openstack deplopyment form queens/pike timeframe default to using cellsv2 with cellsv1 cells were optional. for cells v2 that was changed so all recent opensatack deployment are deploy using cellsv2
15:28:17 sean-k-mooney spatel: so what that means is in the db you have an api db and 2 cell dbs
15:28:19 mriedem *ocata
15:28:57 sean-k-mooney cell0 is special, cell1 is where all your compute nodes are
15:29:31 mriedem we do have docs for this... https://docs.openstack.org/nova/latest/user/cellsv2-layout.html
15:29:48 spatel Reading...
15:29:48 mriedem https://docs.openstack.org/nova/latest/user/cells.html#cells-v2
15:29:55 sean-k-mooney oh good point :)
15:30:06 mriedem if those docs don't answer questions, then ask here
15:30:55 spatel mriedem: let me understand basic concept of technology and then i will come back :)
15:31:03 spatel Thanks sean-k-mooney & mriedem
15:31:10 sean-k-mooney stephen do you mind if i add sphinx-autobuild to os-vif docs requirements?
15:31:18 stephenfin autobuild?
15:32:12 stephenfin sean-k-mooney: Oh, didn't know that was a thing. Doesn't it need to into global requirements if you do?
15:32:13 sean-k-mooney if i install it then vs-codes ret extention can automatically run sphix to provide a live preview of the rst file you are editing
15:32:34 sean-k-mooney am i dont think so
15:32:52 sean-k-mooney but i can add it there too if we want
15:33:03 mdbooth cfriesen: Will look, thanks.
15:33:16 stephenfin Maybe ask on #openstack-requirements. If not, I don't mind. I assume you can't just install it on your system?
15:33:40 cfriesen sean-k-mooney: the TPM code should be ready to take for a spin (without cold migration, hoping to do that today)
15:34:03 sean-k-mooney i can but im writing up some dev docs on how to set up vs-code for development of os-vif
15:34:34 cfriesen sean-k-mooney: no worries. you'll likely need a new os-traits release.
15:34:45 sean-k-mooney stephenfin: for not i can just add a note to pip install it.
15:35:25 stephenfin sean-k-mooney: Yeah, it does feel like something that would be better off installed systemwide (like tox)
15:35:52 sean-k-mooney not really
15:36:22 sean-k-mooney you could but i try to avoid poluting my system with lots of random packages.
15:36:35 sean-k-mooney its nicer to have it in the tox venv but its minor
15:37:29 cfriesen incidentally, I've uploaded a new version of the flavor/image validation patch addressing alex_xu's comments. Would appreciate review as it's on a runway.
15:43:15 mriedem apparently someone doesn't know how to count https://etherpad.openstack.org/p/nova-runways-stein
15:48:21 cfriesen mriedem: I'm talking about the top one in that list, with "jackding" associated with it.
15:50:07 mriedem yeah the show topology one was added to the current runways, i fixed it
15:51:32 cfriesen ah, okay. I couldn't figure out what the problem was. :)
16:12:05 sean-k-mooney stephenfin: actully auto build is not needed.
16:12:47 sean-k-mooney well it is if you are using sphix specifc stuff
16:15:39 openstackgerrit Balazs Gibizer proposed openstack/nova master: Fup for the bandwidth series https://review.openstack.org/639159
16:17:26 gibi mriedem, jaypipes, efried: fixed the comments from the already merged patches ^^
16:19:02 mriedem gibi: ok, did you see my comments on https://review.openstack.org/#/c/622421/ ?
16:19:12 mriedem there were a lot but it's a big patch
16:19:40 gibi mriedem: that is my next patch to look at
16:20:10 gibi mriedem: do you want me to split out the report client change from https://review.openstack.org/#/c/622421/ ?
16:22:27 mriedem gibi: up to you
16:22:34 mriedem it might help move things along faster
16:24:16 gibi mriedem: OK I will do it
16:29:06 gibi mriedem: regarding leaking the port allocation if a port is detached in neutron
16:29:30 gibi mriedem: is this a valid case that nova binds a port in neutron then the admin unbind it in the neturon API?
16:29:45 gibi mriedem: I thought that if the nova bound it the nova needs to unbound it
16:30:21 gibi mriedem: I can start looking at what to do when extrenal event comes if this is a valid use case
16:32:08 mriedem gibi: neutron does not prevent it
16:32:11 mriedem like cinder does
16:32:56 mriedem i don't know how common that could be so i'd be ok with a todo for now to deal with it later. i assume it would only ever happen because something broke and an admin was fumbling around trying to fix things.
16:34:29 gibi mriedem: OK, then I will add a TODO.
16:34:36 openstackgerrit sean mooney proposed openstack/os-vif master: modify functional base.py to allow using vscode https://review.openstack.org/638058
16:36:17 mriedem i have no idea if neutron has policy rules in place to only allow certain users to update a port's device_id
16:36:28 mriedem i know port binding attributes require admin
16:37:09 gibi mriedem: I will check with the neutron guys in my team
16:40:43 openstackgerrit Matt Riedemann proposed openstack/nova master: Improve CinderFixtureNewAttachFlow https://review.openstack.org/639382
16:40:43 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Fix the leak in the cross-cell revert resize code https://review.openstack.org/638314
16:45:52 openstackgerrit Merged openstack/os-vif master: docs: Add API docs for profile, datapath offload types https://review.openstack.org/638395
16:52:58 sean-k-mooney jaypipes: stephenfin im planning on working on working on some os-vif gate jobs after stien feature freeze. would you object to me skipping the tempest jobs in os-vif if the change is only affecting the docs directory.
16:54:17 sean-k-mooney the other tox jobs are quick but if it is a docs only change ore release not only change tempest adds little vlaue
16:55:46 stephenfin fine by me

Earlier   Later