17:00:41 #startmeeting mercadorproject 17:00:42 Meeting started Fri Jul 24 17:00:41 2015 UTC and is due to finish in 60 minutes. The chair is geoffarn_. Information about MeetBot at http://wiki.debian.org/MeetBot. 17:00:43 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 17:00:45 The meeting name has been set to 'mercadorproject' 17:01:33 Anyone here? If not, we can keep it short. 17:01:40 * shaleh raises hand 17:02:06 \o 17:02:07 here 17:02:12 * geoffarn_ wonders why my IRC client has switched IDs 17:03:17 So gyee and I met yesterday at HP (with others), 17:03:32 and we're pulling together design docs 17:03:54 Hope to get them up today or tomorrow 17:03:59 great 17:04:58 The basic question has been, if we build the publisher and subscriber to configure the two Keystones, will everything just work the way it's supposed to? 17:05:13 I.e. do we need to tweak Keystone? 17:05:46 For the first POC, the answer seems to be that it'll work OK 17:06:18 I thought that something was going to be needed to handle remote tokens? 17:06:50 based on chat we had the other day, i.e., subscriber getting token from publisher... 17:06:52 okrieg: for the user to auth and work yes, for pub/sub likely not 17:06:55 Not a problem, as it turns out 17:07:26 shaleh: that seems pretty fundamental 17:07:46 what good is pub/sub if we can't use it :-) 17:08:01 In the long term, we'd like to move a lot of the stuff into Keystone 17:08:31 The pub/sub handle the business logic and policy 17:08:45 okrieg: agreed and it will be done for the POC 17:09:01 okrieg: geoffarn_ said no tweaking needed for pub/sub 17:09:37 we are going to try making a custom token provider and see how far that gets us 17:09:39 However right now we can use pub/sub to relay the tokens from pub to sub, so that the user on the sub side only sees valid tokens with endpoints in the pub cloud 17:10:18 think I need a picture 17:10:29 okrieg: :-) on their way 17:10:58 okrieg we don't need to tweak Keystone. The pub and sub services can configure both keystones 17:11:07 gyee has some slides made, he is about to go on a holiday. When he gets back the wiki will be updates 17:11:47 Right - and I'm updating the sequence diagrams and API to match gyee's slides 17:12:05 we can temporarily move some keystone complexity into an agent, and then merge capability 17:12:25 THe fundamental thing is that no Keystone every has to validate a token that it hasn't generated 17:12:35 every -> ever 17:12:46 geoffarn_: ++ 17:12:48 hmmm, so user talkes to native keystone in each region, pub/sub configures the keystone on each side to get identity from sub, okay, I guess I grock it. Is that correct, the client always goes to the native keystone for pub region? 17:13:11 "generated" isn't the same as "provided" 17:13:12 geoffarn_: makes sense 17:14:03 okrieg: yes. Just like in k2k federation today 17:14:20 okrieg: the plan is to remove the onerous numerous intermediary tokens 17:15:53 not sure I understand where the intermediary tokens come from. 17:15:58 We're meeting f2f again Tuesday to review/plan the POC implementation approach. Key requirement is to be demo-ready by mid-September 17:16:47 okrieg: have you used a k2k federation as the user of a service? 17:18:04 shaleh: we are playing with k2k in the lab, for the mix and match federation work, should probably get deeper on it. 17:20:27 I'm back - my corp VPN just crashed 17:20:37 okrieg: basically you ask "home" keystone for a token, then you need to convert it to a SAML2 request, then you get an unscoped token from "foreign" keystone, then you finally ask for a scoped taken from foreign 17:20:59 okrieg: this is all down by the consumer of the service, in code or commands 17:21:04 s/down/done/ 17:21:47 Obviously this provides a lousy (non-transparent) UX, which is one thing we're trying to fix 17:22:18 ++ 17:23:03 yes, that's exactly what we are doing right now with the mix/match federation 17:23:17 woudl be super awsome if we could use the pub/sub infrastructure to get rid of that 17:23:51 okrieg: we are thinking a customer token provider with a mapping could allow the request to be done server side 17:23:51 The intent is that pub/sub sets things up for you 17:23:59 s/customer/custom/ 17:24:29 awesome! 17:24:49 okrieg: plans and reality though, we shall see 17:25:37 The mercador use case is fairly straightforward, because we don't require or expect mixing and matching 17:25:49 we can design so we can shim in 17:25:59 endpoint groups are homogeneous with respect to auth 17:26:11 not true for Mix & Match 17:26:54 geoffarnold: once we can show it working they may be able to dupe and transform for their needs 17:27:05 agreed 17:27:20 may be use, maybe extend 17:27:42 raildo, what's next for reseller? 17:28:33 geoffarnold: we finish the implementation to use is_domain in the token response... so right now we can put is_domain in the policy.json 17:28:51 fow now, we are just fixing some code reviews in the patches 17:29:04 ok 17:29:37 At some point (prob at the same time as APIv4) we need to fix domain namespace 17:29:41 please review :) #link https://review.openstack.org/#/q/status:open+project:openstack/keystone+branch:master+topic:bp/reseller,n,z 17:30:30 Are you assuming that reseller will handle domain naming for its customers? 17:31:09 hum.. I think so 17:31:41 Same as Mercador. We should align user stories 17:32:00 ok 17:32:38 geoffarnold: I think that we need to think more about the next steps for reseller in M release 17:32:50 and how to improve it to Mercador 17:32:59 I really want to understand how we can create a real "virtual region", like a "virtual machine" - potentially nested arbitrarily deep 17:33:21 raildo +1 17:35:48 I've got a building maintenance guy at the door who wants to turn off my electricity for 15 minutes. I think we'll wrap this up here. 17:36:16 Bye... 17:36:30 #endmeeting 17:36:40 have a good weekend 17:37:20 Ah, I'm now "geoffarnold" rather than "geoffarn_" 17:39:54 #endmeeting