Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-11-28
21:30:17 flwang it's big one, seems no chance to backport :(
21:30:51 efried jaypipes: Responded on https://review.openstack.org/#/c/617042/ - lmk if that doesn't make sense.
21:32:53 efried jaypipes: btw, if it's specifically the use of the @property decorator you object to, I can change it to _get_report_client() for consistency with _get_resource_tracker(). I just wanted to take the opportunity to save some horizontal space, since we're already having trouble fitting things like https://review.openstack.org/#/c/617042/6/nova/compute/manager.py@a766
21:35:32 jaypipes efried: answered. yeah, we can remove all that now and just set reportclient once.
21:35:45 efried jaypipes: Okay, cool. fup?
21:36:47 jaypipes yup. lemme re-vote (though I still think it would be cool to have those two patches separate)
21:37:32 jaypipes efried: +2
21:38:01 efried jaypipes: Thanks! Re removing that flushing of the RT, I'm slightly leery of trying to do that change myself - how will we know it didn't break things? Though perhaps we could instead just self.reportclient.clear_cache() now that that's a thing.
21:38:43 openstack Launchpad bug 1804125 in OpenStack Compute (nova) "Nova placement disregards nova aggregate metadata" [Medium,Triaged]
21:38:43 mriedem heh, finally got an official bug for the regression in ocata where aggregate allocation ratios are no longer honored https://bugs.launchpad.net/nova/+bug/1804125
21:39:08 jaypipes efried: maybe. just try it? :) I'm telling you I added that code comment back when the RT was still being converted by me to pass nodename for all the methods and track multiple compute nodes (instead of having multiple instances of the RT)
21:40:41 openstackgerrit Jack Ding proposed openstack/nova master: [WIP] Flavor extra spec and image properties validation https://review.openstack.org/620706
21:49:07 efried jaypipes: um, afaict, we only get ComputeHostNotFound from get_node_uuid (in resource_tracker.py) - and I can't see where that guy is used at all. So that whole exception path may be completely unreachable.
21:51:24 openstackgerrit Merged openstack/nova-specs master: Support volume-backed server rebuild https://review.openstack.org/532407
21:57:13 jaypipes efried: not a bit unlikely :)
22:02:43 flwang mriedem: jaypipes: is there a config option in nova.conf to set the default volume type?
22:04:12 jaypipes flwang: sorry, I have node idea :( mriedem probably is your best bet. (I didn't even think we *supported* volume types actually..)
22:04:32 openstackgerrit Eric Fried proposed openstack/nova master: WIP: Use a static resource tracker in the compute manager https://review.openstack.org/620711
22:04:42 efried jaypipes: Let's see how the gate feels about that ^
22:04:59 flwang jaypipes: nova supports it in master(stein)
22:05:51 mriedem flwang: no, cinder has a config for the default volume type
22:06:12 mriedem nova only passes the volume type through, otherwise nova creates volumes w/o any volume type and you get the default from cinder
22:07:06 jaypipes efried: lol nice commit message.
22:07:10 efried :P
22:07:35 flwang mriedem: that makes sense, thank you very much
22:07:45 efried jaypipes: btw, I'm hoping you can review that whole series at some point. You're the A-1 expert in these code paths, I think.
22:08:12 efried jaypipes: This is to solve the whole thing about CERN's traffic problems.
22:08:48 efried ...that we put off in queens with that resource_provider_association_refresh conf var
22:09:51 openstackgerrit Matt Riedemann proposed openstack/nova master: Nova the aggregate allocation ratio restriction in scheduler docs https://review.openstack.org/620713
22:11:17 jaypipes efried: ack.
22:11:27 jaypipes efried: working on em....
22:11:39 efried thx
22:14:39 mriedem i have -1ed
22:15:04 mriedem pull that get_node_uuid removal out
22:20:48 openstackgerrit Eric Fried proposed openstack/nova master: Turn off rp association refresh in nova-next https://review.openstack.org/616033
22:21:27 mriedem zzzeek: could use your input on this postgresql issue if you have a sec https://review.openstack.org/#/c/619061/1
22:21:37 mriedem i threw in a thing that works, but i'm not sure it's the right way
22:21:49 efried mriedem: so do it, just do it separately?
22:21:58 mriedem efried: yes, outside of that series
22:22:02 efried ack
22:22:26 mriedem you could throw it in topic branch remove-migration-allocation-compat to line it all up nicely
22:22:31 mriedem https://review.openstack.org/#/q/topic:remove-migration-allocation-compat+(status:open+OR+status:merged)
22:23:32 openstackgerrit Eric Fried proposed openstack/nova master: WIP: Use a static resource tracker in the compute manager https://review.openstack.org/620711
22:27:53 openstackgerrit Eric Fried proposed openstack/nova master: Remove get_node_uuid https://review.openstack.org/620715
22:28:03 efried mriedem: there ya go. Don't say I never did anything for ya.
22:29:15 efried speaking of which, what's the interest rate on shiny nickels?
22:29:21 mriedem it's low
22:29:32 mriedem it's always in my backpack though
22:29:37 mriedem you just need to remind me in person sometime
22:29:47 mriedem or give me your address and i'll ship it down
22:29:56 mriedem maybe in a xmas card?!
22:30:35 efried That would be awesome. What's a stamp these days, 40-odd cents? And you're not supposed to mail cash. Oh, and f xmas.
22:30:52 efried I'll put it on the etherpad for Denver
22:31:28 mriedem good idea
22:31:47 mriedem i'll write you a check for $.05
22:32:21 efried IT HAS TO BE SHINY
22:32:26 efried put glitter or something
22:33:06 mriedem we do have glitter
22:33:11 mriedem pens, glue, you name it
22:33:20 mriedem i will richard shermans this check up for you
22:33:31 mriedem ha, wrong guy
22:33:53 mriedem simmons
22:33:55 mriedem simmons
22:34:07 mriedem jaypipes: you may enjoy that slip ^
22:34:12 mriedem FOOTBAW
22:35:03 jaypipes haha
22:59:23 mriedem mordred: we should only need the endpoint type (volumev3) and interface (public) to look up a service catalog entry right? we don't need the service name for that,
22:59:35 mordred that is correct
22:59:38 mriedem unless you have multiple endpoints pointed at the same type and interface but with different names or something?
22:59:47 openstack Launchpad bug 1803627 in OpenStack Compute (nova) "Nova requires you to name your volumev3 service cinderv3" [Undecided,New]
22:59:47 mriedem trying to sort out what i can do for https://bugs.launchpad.net/nova/+bug/1803627
23:00:15 jaypipes efried: I'm quite concerned about https://review.openstack.org/#/c/615677/
23:00:17 mordred the only time service name is ever useful for anything is if a cloud has gotten itself into a bad place and has more than one endpoint with the same service type like rackspace public cloud did during their transition from legacy to openstack
23:00:40 mordred mriedem: looking
23:01:27 mordred mriedem: oh - I can help work on that patch tomorrow (it's too late today)
23:01:43 mriedem np thanks
23:01:46 efried jaypipes: Looking
23:01:48 mriedem i'm going to post something quick
23:02:20 efried mriedem, mordred: We ought to be able to fix that by exploiting some of the fancy schmancy discovery code in ksa, nah?
23:02:27 mordred I mean - efried will probably beat me to it - but getting rid of that catalog_info parameter at least is a step in teh right direction
23:02:31 mordred efried: yup!
23:02:41 sorrison mriedem: re cinder catalog in nova.conf yes the only way we can get it to work is by setting endpoint_template
23:02:54 efried You ought to be able to a) name your service any of the list of valid things, and b) search for the endpoint using any of that same list of valid things.
23:03:18 mordred yup. and you should absolutely be able to omit service_name which should default to None
23:03:29 jaypipes efried: unless I'm totally bonkers... but AFAIK, when Ironic is in the mix, that ProviderTree can contain thousands of root providers.
23:03:50 efried and yeah, long-standing TODO to get rid of all that bizarro client construction gorp
23:04:04 mordred and you should not only be able to omit service-type- you should really always omit service-type because using different service types is CRAZY - but you should be able to configure it anyway
23:04:24 mordred efried: I started looking at all that before the summit, then had thanksgiving
23:04:39 mordred I'd love to make a patch for y'all with a strawman of ripping a ton of gorp out
23:05:12 efried I'll be happy to review same, but I'm not likely to have time to code it up myself, alas.
23:05:27 efried I'm up to here
23:06:06 mordred cool. well - you did the last one - so it's my turn this time I think
23:06:20 mriedem efried: remember you had https://review.openstack.org/#/c/508345/
23:06:24 mriedem but very out of date by now
23:06:42 efried oo, indeed
23:06:50 mordred mriedem: yeah - I'll likely start by seeing if I can update that :)
23:08:11 efried jaypipes: Hum, I see your point. I did have it rigged at some point to just invalidate the tree of the failing provider. That turned out to be nontrivial and hacky, which is why I went the "invalidate everything" route. But yeah, I may have to restore that logic :(
23:16:04 openstackgerrit Matt Riedemann proposed openstack/nova master: Make [cinder]/catalog_info no longer require a service_name https://review.openstack.org/620738
23:16:05 mriedem sorrison: see how ^ floats your boat
23:17:27 sorrison mriedm: looks very buoyant, thanks!

Earlier   Later