| *** jamielennox is now known as jamielennox|away | 00:36 | |
| *** jamielennox|away is now known as jamielennox | 00:37 | |
| *** gouthamr has quit IRC | 01:48 | |
| *** gouthamr has joined #openstack-shade | 01:49 | |
| *** gouthamr has quit IRC | 05:29 | |
| *** ioggstream has joined #openstack-shade | 08:12 | |
| *** openstackgerrit has quit IRC | 10:33 | |
| *** cdent has joined #openstack-shade | 10:33 | |
| *** cdent has quit IRC | 11:05 | |
| *** cdent has joined #openstack-shade | 11:05 | |
| *** purplerbot has quit IRC | 11:10 | |
| *** purplerbot has joined #openstack-shade | 11:10 | |
| *** cdent has quit IRC | 11:43 | |
| *** ioggstream has quit IRC | 12:06 | |
| *** cdent has joined #openstack-shade | 12:22 | |
| *** yolanda has quit IRC | 12:46 | |
| *** yolanda has joined #openstack-shade | 13:06 | |
| *** yolanda has quit IRC | 13:29 | |
| *** yolanda has joined #openstack-shade | 13:30 | |
| *** ioggstream has joined #openstack-shade | 14:07 | |
| samueldmq | mordred: you around ? | 14:10 |
|---|---|---|
| mordred | morning samueldmq ! | 14:13 |
| samueldmq | mordred: morning | 14:13 |
| samueldmq | mordred: docs say "if what you want is to be able to write an application that talks to clouds no matter what crazy choices the deployer has made in an attempt to be more hipster than their self-entitled narcissist peers, then shade is for you" | 14:13 |
| samueldmq | are you specifically talking about interoperability there ? | 14:14 |
| *** cdent_ has joined #openstack-shade | 14:17 | |
| *** cdent has quit IRC | 14:17 | |
| *** cdent_ is now known as cdent | 14:17 | |
| mordred | samueldmq: I am | 14:19 |
| mordred | samueldmq: it's my ... more aggressive way of talking :) | 14:19 |
| samueldmq | mordred: can you give a very clear example of a choice that can cause interoperability? | 14:19 |
| samueldmq | and is solved by shade ? | 14:20 |
| mordred | yes | 14:20 |
| samueldmq | mordred: ++ | 14:20 |
| mordred | some clouds require floating ips to talk to the internet. some do not require floating ips and give a directly routed ip | 14:20 |
| mordred | some clouds use qcow2 for glance images, some use vhd, some use raw | 14:20 |
| mordred | at least one cloud deploys neutron and puts it in the catalog but does not allow its users to make any neutron api calls | 14:21 |
| samueldmq | mordred: does shade solve that ? does it convert images to the expected format, for example? | 14:21 |
| mordred | whoops, sorry - nope. shade doesn't solve that one - the image related one we solve it: | 14:22 |
| mordred | is | 14:22 |
| mordred | some clouds use glance V1 PUT api to upload images, some use the glance v2 PUT api, some require you to upload the image to swift first then use the v2 task api to import it | 14:22 |
| mordred | that one shade solves :) | 14:22 |
| mordred | in all three of those cases, the shade call is "create_image" | 14:22 |
| mordred | in the server cases, the call is "create_server(auto_ip=True)" and shade will take whatever action it needs to to ensure the server you get back has an ip | 14:23 |
| mordred | oh - here's one you'll enjoy ... :) | 14:24 |
| mordred | some clouds use keystone v2 with a parameter "tenant_name" - others use keystone v3 with the parameter "project_name" :) | 14:24 |
| * samueldmq grabs popcorn and takes notes | 14:24 | |
| mordred | in shade, it's always project_name | 14:24 |
| *** gouthamr has joined #openstack-shade | 14:25 | |
| ioggstream | Hi@all | 14:26 |
| mordred | hi ioggstream ! | 14:26 |
| Shrews | some of that is openstack's fault for allowing/creating such choices | 14:26 |
| mordred | samueldmq: I've got some of these in http://inaugust.com/talks/real-slim-shade.html#/ | 14:26 |
| mordred | Shrews: ++ | 14:26 |
| samueldmq | mordred: nice! thanks | 14:27 |
| samueldmq | mordred: using different services for a given purpose may compromise interop as well, correct? | 14:28 |
| samueldmq | mordred: eg nova-network vs neutron | 14:28 |
| mordred | yup | 14:28 |
| mordred | we try our best to hide that as well | 14:28 |
| mordred | we certainly don't think an end-user should need to know | 14:29 |
| samueldmq | mordred: gotcha | 14:32 |
| *** openstackgerrit has joined #openstack-shade | 14:42 | |
| openstackgerrit | Monty Taylor proposed openstack-infra/shade master: Only do fnmatch compilation and logging once per loop https://review.openstack.org/444331 | 14:42 |
| mordred | Shrews: followup patch to address ianw's concern | 14:42 |
| *** cdent has quit IRC | 15:51 | |
| *** yolanda has quit IRC | 16:06 | |
| *** yolanda has joined #openstack-shade | 16:06 | |
| *** yolanda has quit IRC | 16:11 | |
| *** yolanda has joined #openstack-shade | 16:18 | |
| *** yolanda has quit IRC | 16:26 | |
| *** yolanda has joined #openstack-shade | 16:27 | |
| *** cdent has joined #openstack-shade | 17:16 | |
| *** ioggstream has quit IRC | 17:53 | |
| *** ioggstream has joined #openstack-shade | 18:05 | |
| *** ioggstream has quit IRC | 18:16 | |
| openstackgerrit | Morgan Fainberg proposed openstack-infra/shade master: Convert test_services to requests_mock https://review.openstack.org/444475 | 19:15 |
| *** gouthamr has quit IRC | 20:04 | |
| *** gouthamr has joined #openstack-shade | 20:17 | |
| *** cdent has quit IRC | 20:34 | |
| openstackgerrit | Thanh Ha proposed openstack-infra/shade master: Remove pbr cap https://review.openstack.org/444496 | 20:35 |
| notmorgan | mordred: i'm sad you supported catalog changing in shade via v2 interfaces. | 21:07 |
| notmorgan | mordred: the catalog stuff is massively hard to model | 21:08 |
| notmorgan | because v2 and v3 are so different (yes we suck here) | 21:08 |
| mordred | notmorgan: yah. I'm sad that it was so drastically different :( | 21:34 |
| notmorgan | in a really really crappy way | 21:35 |
| mordred | yah. tell me about it | 21:36 |
| notmorgan | like... i want to take v3 catalog out back and make it disappear | 21:36 |
| notmorgan | there was _zero_ reason to change it... except for the sake of changing it | 21:36 |
| * notmorgan stil would love to see keystone *only* hold a list of services and a base-dns for a region. | 21:37 | |
| notmorgan | and everything else is srv records in dns | 21:37 |
| notmorgan | the whole endpoint thing is pretty insane. | 21:37 |
| mordred | notmorgan: yup. I also want that | 21:37 |
| mordred | but then I want dns to be a first-class citizen in openstack | 21:37 |
| notmorgan | the endpoint modelling is going to be just awful in shade | 21:38 |
| notmorgan | for rest-only | 21:38 |
| notmorgan | ftr | 21:38 |
| notmorgan | i just looked in keystone, and frankly i'm surprised v2 <-> v3 catalog even works | 21:39 |
| *** ioggstream has joined #openstack-shade | 21:43 | |
| samueldmq | notmorgan: v3 -> v2 does not. v2 -> v3 works because the migration adds a legacy_endpoint_id into the v3 entity iirc | 21:44 |
| notmorgan | samueldmq: and that is totally and utterly broken | 21:46 |
| notmorgan | someone screwed up royally in designing the api then | 21:47 |
| notmorgan | the way it's stored in v3 is just dumb | 21:47 |
| notmorgan | there is zero reason you should ever have the same concept in a new version of an API have no compatibility backwards. | 21:48 |
| mordred | +100 | 21:48 |
| samueldmq | ++ agreed | 21:48 |
| samueldmq | domain was okay because there was no domian concept before. | 21:48 |
| notmorgan | i'd open this as a bug against v3 ftr | 21:48 |
| notmorgan | domain is a new concept | 21:48 |
| samueldmq | and then everything in v2 was in the default domain | 21:49 |
| mordred | yup | 21:49 |
| notmorgan | service, region, endpoint is not | 21:49 |
| notmorgan | domain is 100% correct in not having compat leak back to v2. and then having a default domain things ended up in makes a lot of sense | 21:49 |
| mordred | totally agree | 21:49 |
| notmorgan | i actually think this needs to be opened as a bug against v3 and then have a fix backported | 21:50 |
| notmorgan | as much as i loath to "fix v2" | 21:50 |
| notmorgan | in short, v3's catalog is very bad. | 21:50 |
| notmorgan | this is, and you can quote me on this, a case where we should deprecate v3's catalog interface and go back to using the v2 semantics. | 21:51 |
| samueldmq | I found https://github.com/openstack/keystone/commit/2f851340ee8969193b9dcc1913401aa9b33c5d97 | 21:53 |
| samueldmq | where the commit message clearly describes | 21:53 |
| samueldmq | "Endpoints created on the v3 API are not exposed on the v2 API." | 21:53 |
| notmorgan | yeah that is not "clearly" that is/was a poor design decision | 21:54 |
| samueldmq | notmorgan: if we open that bug, I don't think it's fixable | 21:55 |
| samueldmq | other than deprecation, as you just said | 21:55 |
| notmorgan | i do think it would be, we group endpoints in a way with automatic legacy_endpoint_ids generated when using v3 | 21:56 |
| notmorgan | also... the fact we support "whatever interface name you want to give it" is also a bad design decision | 21:56 |
| notmorgan | really we should have stuck with a very restrictive list. | 21:57 |
| notmorgan | an endpoint should support any number of interfaces, but it shouldn't be a new "endpoint" for each "interface" | 21:57 |
| notmorgan | it should have been a one-to-many relationship | 21:57 |
| samueldmq | yeah, it's technically the same endpoint (entity), but just multiple interfaces | 21:58 |
| samueldmq | I agree with you | 21:58 |
| notmorgan | no it isn't | 21:58 |
| notmorgan | not in keystone | 21:58 |
| notmorgan | it is a totally different entity | 21:58 |
| notmorgan | we smush things in weird ways in the catalog itself | 21:58 |
| samueldmq | yes I am saying in the real world it'd make sense to be | 21:58 |
| notmorgan | but it is not the same endpoint at all | 21:58 |
| samueldmq | and in keystone it's not, I am in agreement with you | 21:58 |
| notmorgan | oh yeah, it should be | 21:58 |
| notmorgan | like i said, v3 catalog api is a large amount worse than v2 | 21:59 |
| *** gouthamr has quit IRC | 22:03 | |
| *** gouthamr has joined #openstack-shade | 22:29 | |
| *** ioggstream has quit IRC | 22:41 | |
| notmorgan | FTR: I'm up to ~393 lines changed, for endpoints, and i'm about 1/2 way done. | 23:42 |
| mordred | yah. endpoints suck | 23:42 |
| notmorgan | role_assignments are going to be awful | 23:42 |
| notmorgan | also... ftr: this is the single most tedious bit of code refactoring i've ever done. | 23:43 |
| openstackgerrit | Morgan Fainberg proposed openstack-infra/shade master: change test_endpoints to use requests mock https://review.openstack.org/444541 | 23:44 |
| notmorgan | mordred: ^ you can take a look. | 23:44 |
| notmorgan | it's about 1/2 done | 23:44 |
| notmorgan | it can't pass because i know the code is broken | 23:44 |
| notmorgan | but... w/e | 23:45 |
| mordred | notmorgan: yay w/e! | 23:52 |
| mordred | notmorgan: remind me to get you drunk sometime and tell you stories about mysql->drizzle refactoring | 23:52 |
| mordred | many of those involved figuring out how to walk back places where they were doing pointer arithmetic into classes | 23:53 |
| mordred | (tl;dr - don't do that) | 23:53 |
Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!