Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-08-01
15:06:06 dfisher this is all with beta3
15:06:32 dfisher nova service-list sees my compute node, but cell_v2 discover_hosts can't find it
15:07:02 gibi sdague: sure, there is an overhead due to the extra indirection until the reader gets familiar with the utils but after that it is a lot easier to read the deduplicated code (not to mention chaging it)
15:07:19 jaypipes gibi, dansmith, mriedem, cdent: for the record, I was wrong, dansmith and gibi were right. sorry that I jumped the gun.
15:07:30 mriedem dfisher: what does nova-manage cell_v2 list_cells show?
15:07:34 mriedem do you have a cell0 and cell1?
15:07:55 jaypipes gibi, dansmith, mriedem, cdent: and I did not mean to shut anyone down. I just thought I'd figured out the issue
15:07:57 dfisher mriedem: yes. I have both cell0 and cell1
15:08:10 cdent jaypipes: I’ll have a aggregate map WIP up shortly
15:08:20 openstackgerrit Matt Riedemann proposed openstack/nova master: Deprecate cells v1 https://review.openstack.org/489632
15:08:20 jaypipes cdent: cool, thanks.
15:08:23 gibi jaypipes: you made me think for a while, but I like these mental challenges so no worries
15:08:24 sdague dfisher: what versions of devstack code do you have?
15:08:29 dansmith dfisher: and cell1 points to the same transport and db url as the config for the conductor and the compute?
15:08:32 mriedem dfisher: and does the transport_url for cell1 match the transport_url in nova.conf on the compute node?
15:08:48 dfisher sdague: git pull'd as of yesterday
15:08:59 mriedem dfisher: you can see the transport_url for cell1 by using --verbose with the list_cells command
15:09:11 dfisher dansmith/mriedem: hang on, checking...
15:09:22 mriedem dansmith: sdague: here is the cellsv1 deprecation https://review.openstack.org/489632
15:09:34 dansmith dfisher: this is what I asked you yesterday fwiw :)
15:09:48 dfisher I know. I'm thick :(
15:09:50 jaypipes gibi: no, I feel bad because I didn't see dansmith's comment above "oh you're right, so I should just stop looking then eh?". I didn't mean to shut him down.
15:10:20 sdague mriedem: ++
15:11:58 dfisher dansmith: i apologize but I didn't understand you meant the transport_url
15:12:04 dfisher dansmith: you said, "dfisher: make sure your conductor's config is pointed at the cell1 db, matching the cell1 cell_mapping record"
15:12:11 dansmith dfisher: yesterday or now?
15:12:12 dfisher dansmith: and I thought you meant the database connection string :(
15:12:14 dfisher yesterday
15:12:23 dfisher ok, so with the transport_url set right, i'm in business
15:12:26 dansmith dfisher: well, it's a little of both
15:12:29 dfisher thank you all.
15:12:30 dansmith dfisher: cool
15:12:41 mriedem little bit country? little bit rock&roll?
15:13:42 mriedem dfisher: we could put something in https://docs.openstack.org/nova/latest/user/cells.html#faqs for your ase
15:13:43 mriedem *case
15:14:59 mriedem by we i think i mean you
15:18:16 openstackgerrit Chris Dent proposed openstack/nova master: WIP Update RT aggregate map less frequently https://review.openstack.org/489633
15:19:09 sdague dfisher: ok, it would be good to figure out where that went wrong
15:19:25 sdague because devstack should be setting it up correctly on both controller and compute
15:19:37 sdague and if something is falling through, would be good to fix
15:19:40 mriedem sdague: i assume dfisher is doign multinode
15:19:45 mriedem w/o devstack-gate
15:19:47 sdague mriedem: sure
15:20:13 sdague but the only thing in d-g is the discover_hosts
15:20:16 openstackgerrit Jay Pipes proposed openstack/nova master: Test resize with placement api https://review.openstack.org/487958
15:20:17 openstackgerrit Jay Pipes proposed openstack/nova master: remove source provider allocs in confirm_resize() https://review.openstack.org/488510
15:20:17 openstackgerrit Jay Pipes proposed openstack/nova master: placement: remove existing allocs when set allocs https://review.openstack.org/489273
15:20:19 cdent jaypipes: that patch above does what I think we’re after ^ but needs some reactions and tweaks. gonna get lunch and then will check back
15:20:28 mriedem sdague: d-g also sets up the local.conf for the subnodes doesn't it?
15:20:28 jaypipes cdent: cheers
15:20:36 mriedem otherwise you'd have n-api in enabled_services
15:20:37 cdent bbs
15:20:48 sdague mriedem: yeh, but I didn't think we had cells specific logic there
15:21:01 sdague I'd like devstack multinode without d-g to work for folks
15:21:29 mriedem sdague: there is conditional logic in devstack to know if it's a subnode or not based on n-api running
15:21:30 mriedem is my point
15:22:29 sdague mriedem: yeh
15:23:04 sdague dfisher: anyway, if you have suggestions on how to make this more clear and straightforward to setup on the devstack side, let me know
15:23:26 dfisher sdague: just mention that the transport_url on the compute node needs to match the cell1 transport_url
15:23:28 dfisher that's all you need.
15:24:26 dfisher and, again, thank you sdague, mriedem, and dansmith for putting up with dumb questions from the peanut gallery
15:24:43 gibi jaypipes, dansmith: I have one more question at https://review.openstack.org/#/c/487958/21/nova/tests/functional/test_servers.py@1370
15:25:20 sdague dfisher: any idea where that should be mentioned that it would be more obvious?
15:25:54 dfisher 1.Configure and start your compute hosts. Before step 7, make sure you have compute hosts in the database by running nova service-list --binary nova-compute.
15:25:54 dfisher sdague:
15:25:59 dfisher gah. that should be #6
15:27:00 dfisher maybe something like: 'Configure your compute node's transport_url to point to the cell1 transport_url (visible with nova-manage cell_v2 list_cells —verbose) and start the nova-compute service'
15:27:39 jaypipes gibi: that's concerning to me. I can't explain why that would be. :(
15:29:54 gibi jaypipes: sorry I pointed to the wrong assert it fails the assert 'self.assertEqual(0, len(allocations))'
15:29:55 mriedem dfisher: step 5 says, "The transport url is taken from the [DEFAULT]/transport_url setting in the Nova configuration file if not specified on the command line." - which would assume that if you created cell1 with a transport_url from the config file, you'd be using that same config file on the compute service if you have a single MQ
15:30:19 mriedem dfisher: devstack is a bit different now in that by default there are 2 conductors
15:30:50 mriedem what you suggest isn't a bad reminder though
15:32:03 mriedem dfisher: did you want to push a change to adjust that part of the docs?
15:32:10 bauzas oh man, my computer crashed
15:32:27 mriedem bauzas: it didn't crash, it's on french government mandated PTO
15:32:32 bauzas (says the one who uses AndChat)
15:32:35 mriedem hence the smoke
15:32:39 jaypipes gibi: I believe there should be a *separate* assertion above line 1370 that shows we have the correct allocation after CONFIRM_RESIZE but after the periodics run, that allocation is messed up. I think that's what dansmith was going for there.
15:32:45 dfisher I can. Not sure how, however. That and me no word good.
15:33:02 bauzas mriedem: it's a trap
15:33:12 mriedem dfisher: have you ever pushed a change to review.openstack.org?
15:33:16 dfisher yep.
15:33:26 dansmith isn't dfisher from oracle?
15:33:30 dfisher that's me
15:33:33 dansmith he might burst into flame if he pushes a patch
15:33:36 dansmith so we should probably do it for him
15:33:37 dfisher :(
15:33:57 mriedem dfisher: http://git.openstack.org/cgit/openstack/nova/tree/doc/source/user/cells.rst#n391 is the line you want
15:34:06 dfisher on it.
15:34:16 gibi jaypipes: make sense
15:34:36 jaypipes gibi: I'll add that assertion.
15:34:42 gibi jaypipes: thanks
15:35:46 dfisher dansmith: i'll let you know if you if any parts of me (fleshy or ethereal) start to smoke as I work on this
15:36:17 dansmith dfisher: I'll stand by with a fire extinguisher
15:36:30 dfisher even though I'm an Oracle employee? :)
15:36:39 mriedem sdague: redirect works https://docs.openstack.org/nova/latest/cells.html
15:36:40 dansmith dfisher: you're still a human...
15:36:45 dansmith dfisher: um, right?
15:36:58 gibi jaypipes: would it make sense if I start building some more funcitonal tests top of the your bug fix to see further ahead?
15:37:05 dansmith heh

Earlier   Later