Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-09-06
18:41:07 mriedem which should be possible if we aren't supporting older-than-stein computes in train or ussuri,
18:41:15 dansmith violates the rpcapi though
18:41:22 mriedem where i think i f'ed up was that the compat code in conductor isn't handling request spec being a dict
18:41:46 mriedem no i don't think it violates the rpc api because there was no change to the conductor rpc api to remove the request spec that compute passes back
18:41:50 mriedem it's just a pass through
18:41:55 mriedem the only interface that changed was the compute rpc api
18:42:15 mriedem so the only time compute won't send a request spec back up is if it's pre-stein
18:42:15 dansmith the object could be None there which means old version right?
18:42:36 mriedem well, if you're pinned to less than conductor 1.13 yeah...
18:42:58 dansmith right
18:43:09 mriedem which was mitaka https://review.opendev.org/#/c/277800/ but yeah
18:43:15 dansmith so you can't remove the ability to take an N.0 call unless you bump majors
18:43:26 dansmith I mean,
18:43:26 mriedem right so conductor 2.0 to make request spec required and RequestSpec object
18:43:41 dansmith I realize other shit will break if you go that old, but either we stick to these rules or we don't
18:43:47 mriedem yeah
18:44:11 mriedem i just came across that todo while reviewing gibi's series and though i'm glad we have that, and was going to remove the compat, but then realized i can't
18:44:38 mriedem so i'll be amending that todo comment i guess since i don't feel like taking on a conductor 2.0 bump
18:45:27 dansmith aye
18:51:33 openstackgerrit Merged openstack/nova master: objects: Remove ConsoleAuthToken.to_dict https://review.opendev.org/652970
19:25:07 openstackgerrit Matt Riedemann proposed openstack/nova master: Handle legacy request spec dict in ComputeTaskManager._cold_migrate https://review.opendev.org/680762
19:25:09 mriedem ok i think that handles both issues (bad comment and dict compat)
19:34:18 openstackgerrit Merged openstack/nova master: Bump min for oslo.service & .privsep to fix SIGHUP https://review.opendev.org/679974
19:40:25 openstackgerrit Matt Riedemann proposed openstack/nova master: Cleanup request spec handling in _cold_migrate https://review.opendev.org/680763
19:40:54 openstackgerrit Matt Riedemann proposed openstack/nova master: Isolate request spec handling from _cold_migrate https://review.opendev.org/680763
19:41:02 efried dansmith: I haven't looked at all yet, but it's almost 3 and I'm still not dug out, and wanted to catch you today if possible -- did a) the vpmem Resource ovo stuff and b) the forbidden aggs stuff shake out satisfactorily?
19:41:26 dansmith I haven't gotten to them yet
19:50:17 dansmith efried: do you know why the isolated aggs set was rebased this morning?
19:50:32 dansmith doesnt' seem like anything changed and it means we still don't have a ci run on it yet
19:50:58 efried dansmith: I don't. Maybe so the owner can retain commiter-hood?
19:51:25 dansmith that's not how that works
19:51:54 aspiers efried: on further thought, I think https://review.opendev.org/#/c/680065/ is fine as it stands - see the comments I just posted
19:51:56 efried dansmith: oh, it looks like the bottom patch needed an ovo hash bump -- oops, that's my bad.
19:52:13 dansmith oh okay I thought I checked all the files but maybe I missed that one
19:52:21 dansmith (for changes I mean)
19:52:31 efried https://review.opendev.org/#/c/671072/14..15/nova/tests/unit/objects/test_objects.py
19:53:24 dansmith yeah I know
19:59:36 efried aspiers: beaut, thanks for following up. +A
20:04:15 aspiers efried: yw. Don't hesitate to ask if I can (attempt to) help explain other oddities in the remaining two patches
20:22:07 mriedem artom: dansmith: since stephen and sean aren't around, i'll ask you this https://review.opendev.org/#/c/621476/62/nova/api/openstack/compute/server_topology.py@66
20:22:10 mriedem since i don't know numa
20:22:53 dansmith same pagesize on all cells, is that what you mean?
20:22:57 mriedem yeah
20:23:21 dansmith they will all have a pagesize and they'll all be identical
20:23:30 dansmith I asked that earlier in artom's series
20:23:42 mriedem if they will all have a page size why is that conditional?
20:24:12 dansmith oh, not sure they will all always have a pagesize, but if any do, they'll be the same
20:24:35 dansmith now, that's a libvirt/x86 thing.. I can imagine power allowing multiples because some team in DB2 asked for it in 1996
20:24:54 mriedem ok, that works for me
20:24:55 dansmith all that probably depends on the hypervisor and arch
20:25:11 mriedem this is an intel (tm) api so don't worry about it
20:25:15 dansmith so I'm sure this api will be broken for other ones that aren't considered here
20:25:19 dansmith WONderful
20:25:46 mriedem powerstackers was burned out of governance b/c eric wouldn't lead them anymore
20:25:59 mriedem and the z folks have moved onto openshift on mainframes
20:26:06 dansmith shocking
20:26:37 mriedem has to be weird when the german customers ask for sles + openshift on their z mainframe
20:52:51 mriedem so this is the 2.78 server numa topology command output right now in the proposed change: http://paste.openstack.org/show/772110/
20:52:54 mriedem that....sucks
20:58:12 sean-k-mooney libvirt/qemu support having muliple page size even in the same numa node but we dont support that in openstack and likely never will
20:58:40 mriedem sean-k-mooney: ok i'll stop writing up that spec for ussuri
20:58:43 mriedem i thought you'd be on board
20:58:48 mriedem but i guess you don't like that kind of excitement
20:59:10 sean-k-mooney if you want to debug it when it does not work feel free
20:59:23 mriedem heh
20:59:31 mriedem obviously i'm joking
20:59:55 mriedem i've been trying to come up with some kind of hardware schmorgasbord name generator thing to get your goat
21:00:45 sean-k-mooney just pick 3 random letters and its proably some future NFV hardware thing
21:00:51 mriedem FUK
21:01:01 sean-k-mooney :)
21:01:09 mriedem failover uber kernel
21:01:13 mriedem it's a new DR thing
21:01:28 sean-k-mooney i thought it was a brexit reference
21:01:35 sean-k-mooney but ok :)
21:01:37 mriedem you've been boris'ed
21:01:59 mriedem btw https://www.youtube.com/watch?v=dXyO_MC9g3k if you haven't seen that, it's pretty good
21:02:35 sean-k-mooney i have not, i look forword to it :)
21:04:12 mriedem ok, time for some cli design,
21:04:15 mriedem who's in?
21:04:17 mriedem http://paste.openstack.org/show/772111/
21:05:02 mriedem that's better than before, but the list of nodes in json string is not great,
21:05:23 sean-k-mooney is that the output for the toplogy api?
21:05:25 mriedem but if you dumped each node dict as a separate property, what would you name the property? because that host_node field is only in the response for admins,
21:05:29 mriedem yeah, in the cli
21:05:31 mriedem proposed cli
21:05:55 mriedem so if you're an admin, you could name each one as node{host_node} so like node0 and node1
21:06:01 sean-k-mooney can we pretty print the json blob
21:06:11 mriedem but if you're not admin, calling them node0 or node1 could make the user think they are actually on those physical host numa nodes
21:06:57 mriedem hmm, maybe
21:14:58 sean-k-mooney its a litte more readable. http://paste.openstack.org/show/772112/ i didnt put it in the ascis table
21:15:11 sean-k-mooney but you could also do somithing like
21:15:44 mriedem that's what i already have http://paste.openstack.org/show/772111/
21:16:02 sean-k-mooney |node.[0].cpu_pinning | {"0": 0,"1": 5}
21:16:10 mriedem anyway, i've got this better than it was so i'll push it up and someone else can hack on it
21:16:22 mriedem it's not even my patch
21:17:48 sean-k-mooney the latest windows update sucks ...
21:17:57 mriedem 10...?
21:18:02 sean-k-mooney ya
21:18:07 mriedem i'm rocking 7
21:18:28 sean-k-mooney it install an update last night and now its haning if i boot it in a vm

Earlier   Later