Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-04-26
15:40:23 mriedem jbernard: oh
15:40:24 mriedem {plugin-repo}
15:40:27 mriedem rather than ${plugin-repo}
15:40:53 artom mriedem, err, whoops. Not sure why that happened
15:41:09 mriedem artom: let me just -1 you right here real quick...
15:41:11 openstackgerrit Sylvain Bauza proposed openstack/nova master: WIP: Don't delete allocation if instance being scheduled https://review.openstack.org/460177
15:41:19 artom mriedem, fair
15:41:42 mriedem artom: although if the compute manager or whatever is checking that flag just does a .get('foo', False) that would be fine too
15:41:44 jbernard mriedem: i thought the $ was not needed in yaml
15:41:49 morgan mordred: ++
15:41:51 mriedem jbernard: i'm in bash
15:41:51 bauzas mriedem: POC above for the discussion we had https://review.openstack.org/460177
15:42:08 mriedem jbernard: i mean this is d-g which is bash
15:42:33 artom mriedem, that would be smarter, yes - though I think the precedent is to *not* do that
15:42:56 jbernard mriedem: ooooh, in the wrapper
15:43:05 mriedem artom: maybe not, it's less explicit i guess, idk
15:43:07 artom Maybe we need a has_capability in the generic virt driver, which then check the capabilities dict in the smart way
15:43:19 artom As opposed to overriding everything everttime
15:43:21 openstackgerrit Mikhail Feoktistov proposed openstack/nova master: Add is_vz_container function https://review.openstack.org/445947
15:43:44 mriedem encoding a default in the compute manager does duplicate the defaults, which kind of sucks
15:43:49 mriedem but i don't think those things ever change
15:44:21 mriedem apparently efried and the powervm driver aren't even cool enough yet to override capabilities
15:44:47 artom Default in the manager? I meant manager -> driver.has_capability('foo') -> return self.capabilities.get('foo', <default>)
15:45:00 efried mriedem That was changed per YOUR review comment, nah?
15:45:00 mriedem and no that's not an invitation to tell me to review whichever powervm patch adds capabilities overrides
15:45:30 mriedem artom: meh
15:45:35 mriedem artom: we're over thinking this i think
15:45:59 artom *shrug* it'd be a net improvement, no matter how small ;)
15:46:11 artom But it can be done separately
15:46:22 mriedem i agree with doing it separately
15:50:24 mriedem johnthetubaguy: we should be good to go here https://review.openstack.org/#/c/456896/
15:50:45 mriedem johnthetubaguy: we're going to need to find a 3rd sucker, i mean core, to help review the new cinder api changes if i'm going to be contributing to the series
15:52:16 johnthetubaguy mriedem: cool, I will take a peak at that now
15:52:17 cdent bauzas: left a comment on your WIP that I wanted to be sure you didn't miss since your -W
15:53:59 stackempty hola .. does anyone know if something like this is already available https://specs.openstack.org/openstack/nova-specs/specs/newton/approved/flavor-root-disk-none.html
15:54:17 mriedem stackempty: you've just stepped into something
15:55:02 mriedem stackempty: see http://lists.openstack.org/pipermail/openstack-dev/2017-February/113063.html
15:55:08 mriedem the "Enable Cinder as an ephemeral storage backend for Nova" part
15:56:16 mriedem stackempty: if you're going to be at the forum https://www.openstack.org/summit/boston-2017/summit-schedule/events/18738/using-cinder-for-nova-ephemeral-storage-backend?BackURL=https%3A%2F%2Fwww.openstack.org%2Fsummit%2Fboston-2017%2Fsummit-schedule%2Fglobal-search%3Ft%3DRiedemann%23eventid%3D18738
15:56:25 sfinucan jaypipes, johnthetubaguy: I'm going to add a 'reserved_host_disk_gb' option if that makes sense?
15:56:34 sfinucan to resolves the TODOs in https://review.openstack.org/#/c/457782/2/nova/scheduler/client/report.py
15:57:05 mriedem "It would be better to claim disk in MB, but that is hard now."
15:57:39 mriedem i don't see that changing anytime soon
15:57:45 mriedem claiming disk in mb vs gb i mean
15:57:59 mriedem stackempty: so the answer is not yet,
15:58:09 mriedem stackempty: that feature from that spec never landed and was abandoned,
15:58:23 mriedem stackempty: there have been alternative proposals since then to solve the same use case,
15:58:36 mriedem which is i don't have local disk on the computes, i need/want to use cinder for all storage
15:58:39 mriedem including ephemeral
15:58:44 stackempty Thanks a lot @mriedem .
15:58:59 bauzas mriedem: FWIW, I'd love to count on bytes even :)
15:59:11 mriedem hell, let's count in nibbles
15:59:34 bauzas mriedem: remember the helper script I needed to write for MB vs. GB ?
15:59:44 bauzas and why bytes are cool :)
15:59:55 bauzas or count in bosons if you wish
16:00:12 bauzas smallest atomic amount FTW
16:00:55 sfinucan seems I also stepped into something too. Oops
16:02:06 johnthetubaguy sfinucan: I guess all the flavor stuff is in GiB, so having the configuration match that probably makes good sense.
16:02:28 mriedem stackempty: there was also https://blueprints.launchpad.net/nova/+spec/create-bdm-from-flavor-extra-specs
16:02:30 sfinucan johnthetubaguy: and from the look of the above, it's unlikely that that'll change anytimesoon
16:02:36 sfinucan right?
16:02:39 mriedem but we didn't want to go the route of stuffing bdm information in flavor extra specs
16:02:44 mriedem the worst of both worlds
16:03:22 johnthetubaguy mriedem: the above spec is the reverse I thought, its just a flavor that says "must have boot volume"
16:04:00 mriedem johnthetubaguy: the UX was bad imo,
16:04:10 mriedem i think the idea was you mark your flavors as local_disks=False,
16:04:25 mriedem and if the user tries to create a server with one of those flavors but without defining bdms, it fails
16:04:41 johnthetubaguy sfinucan: my head swap was in MiB in the flavor but it seems like its not, I must have got mixed up with RAM
16:05:00 mriedem putting the burden on the user to have to sort out which flavors they can use and if they have to have volumes, etc
16:05:19 mriedem it'd be much better UX if the operator configures the compute to use cinder as an ephemeral storage backend
16:05:25 mriedem regardless of flavor
16:05:33 bauzas oh fun, we already verify existing allocations before posting
16:05:45 bauzas mriedem: jaypipes: dansmith: cdent: ^
16:06:10 cdent yes, didn't we already know that?
16:06:12 johnthetubaguy mriedem: true, but that seems years away at this point, and the other way makes diskless hypervisors upstream
16:06:22 bauzas cdent: I wrote that as a work item
16:06:30 bauzas for the scheduler claims stuff
16:06:34 mriedem johnthetubaguy: well, https://specs.openstack.org/openstack/nova-specs/specs/newton/approved/flavor-root-disk-none.html also bakes that into the API
16:06:36 johnthetubaguy mriedem: the problem is the mix of diskless and non-diskless
16:06:40 bauzas cdent: and nobody reviewed that saying "meh" :p
16:06:56 mriedem johnthetubaguy: so then we have to support local_disks=False in flavors in the API for years probably
16:07:03 mriedem even if we ever get cinder ephemeral
16:07:08 cdent bauzas: well, I think I did when discussing the change to not use RPC, but it didn't get picked up and didn't seem a relevant detail
16:07:34 mriedem johnthetubaguy: there was also https://review.openstack.org/#/c/432503/ which i really didn't want since it mixes extra specs and bdms
16:07:46 cdent bauzas: in any case, did you seem my earlier [t 3xBS]
16:07:47 purplerbot <cdent> bauzas: left a comment on your WIP that I wanted to be sure you didn't miss since your -W [2017-04-26 15:52:17.085960] [n 3xBS]
16:07:56 johnthetubaguy mriedem: yeah, the auto create seems evil
16:08:21 mriedem johnthetubaguy: anyway, we talked about cinder ephemeral at the ptg, a spec was never written, but it's a forum session to discuss it or alternatives
16:08:30 johnthetubaguy mriedem: yeah
16:08:36 mriedem if people want this, then they need to put up the dev resources to do it
16:08:41 mriedem rather than api hacks
16:09:09 mriedem cinder ephemeral imagebackend also allows us to deprecate lvm, rbd and scaleio (if that lands)
16:09:29 mriedem i think anyway
16:09:57 johnthetubaguy mriedem: my problem is the whole way we do disks is truly terrible, so to get to something clean buts a lot of burden on the users to switch to something radically different
16:10:00 mriedem having said this, there is probably a world of API operations that don't work for boot from volume servers, like shelve/unshelve and rebuild
16:10:23 mriedem users == operators?
16:10:28 johnthetubaguy mriedem: they should all work, they might be broken
16:10:33 johnthetubaguy mriedem: I was meaning API users
16:11:06 mriedem i don't see how a cinder ephemeral imagebackend is radically different from the API pov
16:11:11 mriedem the api doesn't change

Earlier   Later