| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-08-20 | |||
| 13:54:07 | alex_xu | sean-k-mooney: efried mriedem https://etherpad.openstack.org/p/vpmems-non-virt-driver-specific | |
| 13:54:25 | alex_xu | ^ I'm wating for Luyao ensure her code, but that is the basic idea | |
| 13:55:06 | efried | alex_xu: Whereas it would be nice to be able to do without data model changes, I think this makes sense and makes a number of things much simpler. | |
| 13:55:35 | alex_xu | efried: \o/ | |
| 13:56:35 | mnaser | https://docs.openstack.org/nova/latest/admin/configuration/schedulers.html#scheduling-considerations | |
| 13:57:04 | mnaser | is the note a bit contradicting, if i understand correctly, setting aggregate allocation ratios doesnt work in the scheduler anymore? | |
| 13:57:19 | mnaser | so AggregateCoreFilter is technically useless? | |
| 13:57:24 | mriedem | https://review.opendev.org/#/c/673496/ | |
| 13:57:38 | mnaser | ok | |
| 13:57:39 | mnaser | makes sense now | |
| 13:58:09 | mriedem | maybe "The AggregateCoreFilter, AggregateDiskFilter and AggregateRamFilter filters allow overriding per-compute allocation ratios by setting an allocation ratio value using host aggregate metadata. This provides a convenient way to manage a group of compute hosts with similar allocation ratios while leaving the configuration settings alone." should just be removed | |
| 13:58:29 | mnaser | yeah it seems to tell you "this is how to solve this problem. note, it doesn't work" | |
| 13:58:31 | mriedem | because it's essentially saying, "before ocata, ^ was true, but since ocata it's not" | |
| 13:58:46 | mnaser | so for a stein cloud, i can just update the placement service aggregate then? | |
| 13:59:11 | mriedem | you have to set the allocation ratio on each provider in the placement aggregate | |
| 13:59:24 | mriedem | mel has a convenience cli patch for doing this, sec | |
| 13:59:34 | mriedem | https://review.opendev.org/#/c/640898/ | |
| 13:59:52 | mriedem | so since pike your compute host aggregates and provider aggregates should be automatically mirrored, | |
| 14:00:10 | mriedem | the thing that isn't is the compute host aggregate allocation ratio metadata, since placement resource provider aggregates don't have metadata | |
| 14:00:11 | mnaser | oh neat, and technically i shouldn't worry about nova re-overriding it because it only adds it when the rp is first registered, right? | |
| 14:00:37 | mriedem | if you change the allocation ratio config on a nova-compute it would be updated in placement | |
| 14:00:50 | mriedem | that's where https://docs.openstack.org/nova/latest/admin/configuration/schedulers.html#usage-scenarios comes in | |
| 14:00:53 | mnaser | oh i thought we now had initial_foo | |
| 14:01:01 | mriedem | b/c some people (cern) want to manage allocation ratios in config always, | |
| 14:01:13 | mriedem | and others (mgagne) want to manage allocation ratios in an API always | |
| 14:01:29 | mriedem | config trumps api | |
| 14:01:33 | mnaser | hmm okay, so either stick to initial_foo and leave others empty or foo | |
| 14:01:33 | mriedem | if config is set | |
| 14:01:40 | mriedem | yes | |
| 14:01:42 | mnaser | yeah makes sense, i think ill stick to config for now | |
| 14:02:01 | mriedem | you just described scenario 2 | |
| 14:06:40 | mriedem | ok i see how the note got mixed in and is confusing, will push a patch to clean that up | |
| 14:07:05 | cfriesen | stephenfin: aspiers: we don't care about postgres anymore. :) | |
| 14:07:17 | mriedem | but mnaser does now | |
| 14:07:57 | mnaser | Oh I don't either | |
| 14:08:10 | mnaser | It was just a pita to run tests locally | |
| 14:08:26 | mnaser | And someone mentioned something about needing postgres libs so some postgres tests run | |
| 14:08:47 | sean-k-mooney | alex_xu: ill come back to your etherpad in a bit just going to finish the thing i was in the middle of | |
| 14:09:03 | alex_xu | sean-k-mooney: thanks! | |
| 14:11:04 | mriedem | mnaser: oh :) | |
| 14:11:06 | openstackgerrit | Matt Riedemann proposed openstack/nova master: doc: remove confusing docs about aggregate allocation ratios https://review.opendev.org/677472 | |
| 14:11:08 | mriedem | mnaser: also ^ | |
| 14:12:25 | mnaser | if we dont care about postgres ill gladly update that change to remove it entirely -- https://review.opendev.org/#/c/676684/ | |
| 14:12:50 | mriedem | cfriesen said he, and by extension windriver/starlingx, don't care about postgres anymore | |
| 14:13:02 | mriedem | that doesn't mean we're just ripping it out, there are others that use it i think | |
| 14:13:14 | mriedem | we have a job in the experimental queue to test with pg on-demnd | |
| 14:13:21 | mriedem | i think cdent dabbles with pg as well | |
| 14:13:33 | stephenfin | openstack-discuss post time? | |
| 14:13:37 | mriedem | why? | |
| 14:13:48 | mriedem | it's already been discussed, several times, | |
| 14:13:52 | mriedem | including with a TC resolution years ago | |
| 14:14:02 | mriedem | and people came to the "meh, it's not hurting, so who cares" | |
| 14:14:55 | stephenfin | mnaser, aspiers and I, for a start | |
| 14:15:13 | stephenfin | idk, feels odd to keep testing something that no one uses or cares about | |
| 14:15:14 | mriedem | what is prompting this coming up? | |
| 14:15:34 | mriedem | we test and fix pg on demand when there is an issue someone reports, because there are people using it | |
| 14:15:38 | mnaser | i assume when i ran unit tests locally and psycopg2 failed to build cause i was missing headers | |
| 14:15:45 | mriedem | just because they aren't daily upstream contributors doesn't mean people are'nt using it, or that it's broken | |
| 14:15:48 | mnaser | so i was like "well i scanned the source code and couldnt find any references, why is it in requirements?" | |
| 14:16:00 | mnaser | and someone explianed something about some sort of test path that happened if the library was found | |
| 14:16:14 | mnaser | and someone asked cfriesen if they used postgres and here we are, afaik | |
| 14:16:18 | mriedem | it's really a bindep thing for the opportunistic schema migration tests | |
| 14:16:27 | mnaser | https://review.opendev.org/#/c/676684/ means we dont have to carry it anymore | |
| 14:16:32 | mriedem | we can just as easily remove mysql from test-requirements | |
| 14:18:00 | stephenfin | I must say it'd be nice not to need to install those to build our docs. What happened to the idea to move them into 'extras', like oslo.db does? | |
| 14:18:19 | stephenfin | I didn't know there were actually people using postgres though. I thought it was just cfriesen, tbh :) | |
| 14:19:42 | mriedem | prometheanfire also uses pg | |
| 14:19:44 | mriedem | and huawei | |
| 14:21:55 | mnaser | stephenfin: i think to build docs we use doc/requirements.txt now? | |
| 14:22:02 | mnaser | so you shouldn't need to install it.. i think? | |
| 14:22:12 | stephenfin | mnaser: apidoc :( | |
| 14:22:55 | stephenfin | that introspects the code so you need nova to be installed, which brings in all the dependencies | |
| 14:22:59 | mnaser | ah gotcha | |
| 14:23:10 | mnaser | hmm | |
| 14:23:10 | mriedem | we don't publish an api module index | |
| 14:23:12 | mriedem | in our docs | |
| 14:23:22 | mnaser | but we dont have psycopg in requirements, only in test-requirements i think | |
| 14:23:28 | mnaser | no db driver exists in requirements afaik | |
| 14:23:28 | mriedem | if we're building an api doc, it's a waste of time | |
| 14:24:03 | mriedem | mnaser: correct. -1 on the lower-constraints thing, otherwise i'd be +2. | |
| 14:24:10 | mriedem | might as well address efried's comments as well | |
| 14:24:47 | stephenfin | mriedem: We do. See doc/source/reference/services.rst and doc/source/reference/services.rst | |
| 14:25:04 | stephenfin | doc/source/reference/notifications.rst | |
| 14:25:04 | stephenfin | 145:.. automodule:: nova.notifications.objects.base | |
| 14:25:04 | stephenfin | doc/source/reference/services.rst | |
| 14:25:04 | stephenfin | 31:.. automodule:: nova.service | |
| 14:25:04 | stephenfin | 41:.. automodule:: nova.manage | |
| 14:25:26 | mriedem | oh god, we could just remove that | |
| 14:25:43 | stephenfin | no argument from me (y) | |
| 14:25:56 | mriedem | that's probably a relic from when you could extend that stuff via config out of tree | |
| 14:26:00 | mriedem | we removed that eons ago | |
| 14:26:13 | stephenfin | even the notifications stuff? | |
| 14:26:45 | mriedem | ew this https://docs.openstack.org/nova/latest/reference/notifications.html#the-nova-notifications-objects-base-module | |
| 14:27:12 | mriedem | that's actually useful for contributors adding new versioned notifications | |
| 14:27:23 | mriedem | is there a way to only build part of the api doc and not all of it? | |
| 14:28:04 | stephenfin | not that I'm aware of, but that autodoc package is really tough to grok | |
| 14:29:55 | openstackgerrit | Stephen Finucane proposed openstack/nova master: WIP: Move DB, virt driver deps to 'extras' https://review.opendev.org/677475 | |
| 14:29:59 | stephenfin | mnaser, mriedem, aspiers: WIP but I think that would do the trick ^ | |
| 14:30:41 | prometheanfire | if I had to do it again... | |
| 14:31:38 | efried | alex_xu, sean-k-mooney: I think we now need to write a crisp discussion-free version of the design points as refined. I recommend either creating a new etherpad or shunting the current content down and doing it above. | |