Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-04-10
20:11:56 dansmith it does no clienty things
20:12:15 dansmith it does direct-to-the-database things, which is why it is very tightly coupled to the server code
20:12:21 kfox1111 ah. ok.
20:12:33 kfox1111 one other quiestion...
20:12:40 kfox1111 in the docs, I found this:
20:12:43 kfox1111 nova-manage cell_v2 discover_hosts
20:12:54 kfox1111 but not much on how you decide which cell gets which hosts.
20:13:00 mriedem sec
20:13:17 mriedem https://docs.openstack.org/developer/nova/man/nova-manage.html#nova-cells-v2
20:13:24 mriedem you might be looking at old docs, ^ is much more detailed
20:13:34 mriedem also https://docs.openstack.org/developer/nova/cells.html
20:13:39 mriedem is the super detailed version
20:13:40 kfox1111 ah. nice. thank you. :)
20:13:51 mriedem to explain what cells v1 and cells v2 is
20:13:55 mriedem and the commands for cells v2 setup
20:14:02 mriedem and scenarios for installing or upgrading to cells v2
20:14:31 mriedem by default, every nova deployment starting in ocata is a single-cell cells v2 deployment
20:14:47 mriedem we don't support multiple cells v2 cells in ocata
20:14:53 mriedem working on that for pike
20:15:09 mriedem if dansmith would stop answering questions in channel and fix his series :P
20:15:12 kfox1111 ah. ok. thats good to know too. :)
20:15:23 kfox1111 thanks for the info.
20:15:34 mriedem dansmith is also presenting on all this at the forum if you'll be there
20:15:37 dansmith mriedem: um excuse me, but there are four in the gate right now
20:15:39 mriedem plus a forum session
20:16:09 mriedem not this motherscratcher https://review.openstack.org/#/c/442736/
20:16:14 kfox1111 oh. one more. is there a way to list out the cells and what hosts are in what cells?
20:16:26 kfox1111 just to ensure all the compute nodes are registered that I'm expecting for debuggging.
20:16:33 dansmith kfox1111: seriously.. look at the docs
20:17:05 mriedem there is a nova-manage cells_v2 list_cells command
20:17:16 mriedem don't think we have one to list the host mappings per cell yet
20:17:17 kfox1111 dansmith: it may be obvious to you, but I'm not seeing it.
20:17:29 kfox1111 ah. thats why I didn't see it. thanks.
20:17:30 mriedem https://docs.openstack.org/developer/nova/man/nova-manage.html#nova-cells-v2
20:17:35 mriedem nova-manage cell_v2 list_cells
20:17:39 dansmith mriedem: discover_hosts is idempotent and --verbose will list the hosts for each
20:17:58 kfox1111 so, I just gotta run discover_hosts if I'm unsure if all are bound.
20:18:12 mriedem dansmith: that's not mentioned in the man page for discover_hosts
20:18:13 mriedem but ok
20:18:20 bauzas call me dumb, but I can't find where in the RT code instance.claim() is defined as a contextmanager?! The Claim object itself is not having __enter__ and __exit__
20:18:31 mriedem we could expand list_cells to show hosts in a given cell
20:18:34 dansmith kfox1111: you can only have one cell with hosts right now, so .. either they're mapped there or they're not
20:18:45 mriedem bauzas: the parent noopclaim class
20:18:52 bauzas oh shit, nvm
20:18:59 bauzas mriedem: yup, just missed that
20:19:04 kfox1111 dansmith: yeah. but I'm just trying to overify if thats the case or not.
20:19:15 kfox1111 there are two states there. :)
20:19:20 bauzas grmblbl, it's late
20:19:20 dansmith kfox1111: have you created other cell databases?
20:19:36 dansmith kfox1111: if so, then you messed up anyway and if not, then no hosts are registered in a nonexistent database
20:19:47 mriedem kfox1111: dansmith: to be clear, we can expand list_cells for more info, we talked about that in ocata i think, i think it'd be useful
20:19:47 kfox1111 dansmith: yeah. I have all the code for creation in. it works most of the time. but if a race condition happans and the cell is created before the compute,
20:19:57 kfox1111 I wana ensure that all the nodes madie it into the cell.
20:20:08 kfox1111 just running discover_hosts again should fix it.
20:20:19 kfox1111 but was looking for more info to look for the condition.
20:20:25 kfox1111 but that doesn't seem like its there yet. thats fine.
20:20:25 mriedem reporting clis will be nice for people rolling into this that are new to it
20:20:38 mriedem "it" == cells v2
20:20:46 kfox1111 just rerunning discover_hosts will be an ok workournd for now.
20:21:10 kfox1111 yeah. I'm comming from many years of nova deployments. the cell v2 stuff is all very new to me.
20:21:21 dansmith discovering and listing hosts per cell is nearly the same operation
20:21:34 dansmith there's not much point in listing alone
20:21:37 kfox1111 yeah. but one makes changes, the other does not.
20:21:45 kfox1111 yes, there is.
20:21:59 mriedem the point is i'm afraid of unexpected changes
20:22:01 kfox1111 one tells you why something migh have failed in the past.
20:22:06 mriedem so just read-only type commands i'm happy with
20:22:07 kfox1111 the other fixes an issue.
20:22:33 dansmith kfox1111: discover_hosts differentiates between hosts that are existing and added,
20:22:34 dansmith but that's fine
20:22:38 kfox1111 those are potentially different operations to an operator trying to understand and fix their system.
20:22:49 dansmith for people with thousands of hosts, the list is not super pointful, IMHO
20:23:05 kfox1111 I'm only in the hundereds.
20:23:07 kfox1111 ;)
20:25:47 kfox1111 anyway. as an op, its not very intuitive to go looking for "how do I find out what hosts are bound to what cell" and expect "bind hosts to cell" to be the right answer. just fyi. might be a good faq item at least.
20:26:51 dansmith if you want to list hosts in a cell,
20:27:01 dansmith you would use the regular host list thing pointing at a cell config file
20:27:11 dansmith which we can automate from the top level for sure,
20:27:43 dansmith but it just seems not that useful to me to get 10k hosts in a list to find if host N is in cell M
20:27:46 dansmith but clearly it is
20:27:59 openstackgerrit Matt Riedemann proposed openstack/nova master: api-ref: move createBackup to server-actions https://review.openstack.org/452971
20:27:59 openstackgerrit Matt Riedemann proposed openstack/nova master: Document and provide useful error message for volume-backed backup https://review.openstack.org/452970
20:28:00 openstackgerrit Matt Riedemann proposed openstack/nova master: 2.43: Remove Location header from createImage and createBackup responses https://review.openstack.org/455443
20:28:50 mriedem dansmith: you mean nova-manage host list right?
20:28:53 mriedem pointing at the cell config?
20:28:58 dansmith yar
20:29:00 mriedem ok
20:29:08 mriedem maybe i'll start a FAQs section to the cells page
20:29:11 mriedem i thought about that at one point
20:29:15 mriedem same for placement
20:29:32 mriedem because there was a couple of weeks around ocata rc when i felt like we were answering the same questions over and over
20:30:26 mriedem oh snap today is supposed to be mitaka eol
20:36:17 bauzas mriedem: dansmith: cdent: edleafe: possibly a misunderstanding of me, but I thought we finally ended up saying that we would use the RequestSpec object as the conveyor for the RPC bus ?
20:36:46 cdent bauzas: mriedem wants it over rpc but not in the requestspec object
20:36:48 bauzas mriedem: dansmith: cdent: edleafe: but etherpad tells me we agreed it would be a separate RPC arg
20:37:13 bauzas okay, tbh, I'm not opiniated at all
20:37:38 bauzas cdent: anyway, seems an implementation detail to me (well, to a certain point since it requires RPC version bumps)
20:38:48 dansmith bauzas: the etherpad is right
20:39:33 bauzas roger.
20:41:32 mriedem -2 to stuffing the allocation list into the requestspec

Earlier   Later