| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2020-06-18 | |||
| 03:55:45 | brinzhang | I dont similar this scenario for you test error now :(, sorry shaohe_feng | |
| 03:58:37 | Yumeng | is shaohe_feng still around? | |
| 03:58:40 | shaohe_feng | yes | |
| 03:58:53 | shaohe_feng | I have read efried note | |
| 03:59:23 | shaohe_feng | the glance image handle the endpoint is different from others | |
| 03:59:41 | shaohe_feng | it will join version twice | |
| 04:00:00 | shaohe_feng | such as: /image/v2/v2/balabala | |
| 04:00:48 | Yumeng | sounds interesting.. | |
| 04:01:34 | shaohe_feng | the first time efried write a special code to get endpoint for image | |
| 04:01:56 | shaohe_feng | if error, it will try another way to get the endpoint | |
| 04:02:05 | Yumeng | shaohe_feng: does this need a real image to test? can we test this with fake image and fake driver? | |
| 04:02:20 | shaohe_feng | these two logic I have past in above | |
| 04:02:39 | Yumeng | aha. cool. | |
| 04:02:46 | shaohe_feng | https://github.com/openstack/cyborg/blob/6dab512af226677c9065f85f264a691f3aeb6eab/cyborg/common/utils.py#L185 | |
| 04:02:55 | shaohe_feng | https://github.com/openstack/cyborg/blob/6dab512af226677c9065f85f264a691f3aeb6eab/cyborg/common/utils.py#L200 | |
| 04:03:11 | shaohe_feng | ^ Yumeng, you can have a look at it | |
| 04:03:19 | brinzhang | I remember download image need to steps, should combine the glance api, but I am not sure | |
| 04:03:28 | shaohe_feng | the logic it not complex. | |
| 04:04:02 | shaohe_feng | Yumeng let me find the bug description about image | |
| 04:05:22 | shaohe_feng | Yumeng: https://bugs.launchpad.net/python-glanceclient/+bug/1707995 | |
| 04:05:22 | openstack | Launchpad bug 1707995 in Glance Client "Can't use versioned endpoints" [Low,Triaged] | |
| 04:05:34 | shaohe_feng | ^ it is here | |
| 04:05:54 | shaohe_feng | But it think it is not the root cause fro my issue. | |
| 04:06:20 | shaohe_feng | not sure | |
| 04:06:31 | shaohe_feng | I can not get the endpoint from keystone adaptor | |
| 04:06:53 | shaohe_feng | maybe I need dive into it | |
| 04:07:07 | shaohe_feng | just want to know anyone know it | |
| 04:07:25 | Yumeng | ok. I will take a look. but I don't have an real env to test the program. :( | |
| 04:07:40 | shaohe_feng | or who introduce glance API into cyborg | |
| 04:07:52 | shaohe_feng | may know better about it. | |
| 04:08:30 | Yumeng | maybe you can also discuss with shogo. he may know something. | |
| 04:09:23 | Yumeng | I will also take a look at this issue. ^^ | |
| 04:09:50 | Yumeng | just ping in wechat, so that we all can see that. | |
| 04:10:13 | Yumeng | ops. we seems run out of time. | |
| 04:10:28 | Yumeng | #topic AOB | |
| 04:11:11 | Yumeng | shaohe_feng: or ML is better? glance guys can see it and discuss | |
| 04:11:44 | shaohe_feng | ok | |
| 04:11:50 | shaohe_feng | Yumeng thanks | |
| 04:11:52 | Yumeng | is there anything else for today? | |
| 04:12:18 | Yumeng | if nothing else let's wrap up today's meeting. Thank you all. | |
| 04:12:19 | Yumeng | bye | |
| 04:12:29 | Yumeng | #endmeeting | |
| 04:12:31 | openstack | Meeting ended Thu Jun 18 04:12:29 2020 UTC. Information about MeetBot at http://wiki.debian.org/MeetBot . (v 0.1.4) | |
| 04:12:32 | openstack | Minutes: http://eavesdrop.openstack.org/meetings/openstack_cyborg/2020/openstack_cyborg.2020-06-18-03.01.html | |
| 04:12:33 | openstack | Minutes (text): http://eavesdrop.openstack.org/meetings/openstack_cyborg/2020/openstack_cyborg.2020-06-18-03.01.txt | |
| 04:12:34 | openstack | Log: http://eavesdrop.openstack.org/meetings/openstack_cyborg/2020/openstack_cyborg.2020-06-18-03.01.log.html | |
| 09:26:57 | efried | Yumeng, shaohe_feng, brinzhang: If you need help with that glance endpoint thing, I might be able to remember the details, if you catch me during US Central daytime. | |
| 09:27:18 | efried | But IIRC what happened here was that cyborg copied in a bunch of nova code, warts and all. | |
| 09:27:56 | efried | I can't remember if we since fixed up the nova side in this area. We might have. That would be the first thing to check. Then you could assimilate whatever nova is doing to get around this. | |
| 09:27:57 | efried | But | |
| 09:28:05 | shaohe_feng | efried thanks you | |
| 09:28:17 | efried | what would be better is if cyborg used openstacksdk instead. | |
| 09:28:19 | shaohe_feng | Yes, need your help | |
| 09:29:14 | efried | If you jump in #openstack-sdks, they might be able to help you get that spun up. | |
| 09:29:15 | shaohe_feng | do you means use openstacksdk instead of keystone adapter? | |
| 09:29:20 | efried | yes | |
| 09:29:40 | efried | We started doing that cutover in nova. It was a pretty simple transition. | |
| 09:29:51 | efried | We did it for placement, and got most of the way there for ironic. | |
| 09:30:04 | efried | let me find those patches... | |
| 09:30:13 | shaohe_feng | sounds good. | |
| 09:31:36 | shaohe_feng | currently, I can not get the image endpoint from keystone adapter | |
| 09:32:15 | shaohe_feng | And for the service_catalog is empty. | |
| 09:34:07 | efried | Here's a good place to start poking: https://github.com/openstack/nova/blame/master/nova/utils.py#L1053 | |
| 09:35:32 | shaohe_feng | let me read up ti. | |
| 09:36:45 | shaohe_feng | is it convenience to change the current keystone adapter to openstacksdk | |
| 09:36:55 | shaohe_feng | how many effort? | |
| 09:37:15 | shaohe_feng | such as how many days? | |
| 09:37:26 | efried | "It depends" :P | |
| 09:38:03 | efried | The transition from ksa should be very easy, ~1d (of course blown out across many calendar days for review cycle etc) | |
| 09:38:33 | efried | When we cut placement over, the transition involved very little code. | |
| 09:39:33 | shaohe_feng | sounds good. does placement call glance? | |
| 09:40:10 | shaohe_feng | I see your ksa code, there are some special handle for image? | |
| 09:40:29 | efried | No, sorry, what I mean is: | |
| 09:40:29 | efried | For a given $service that $your_project calls via the API, if today you're using ksa to get a handle on that API, transitioning to calling $service via sdk is easy. | |
| 09:40:55 | efried | We did this for $service=placement from $your_project=nova. | |
| 09:40:55 | efried | I'm suggesting you do it for $service=glance from $your_project=glance | |
| 09:41:04 | efried | sorry | |
| 09:41:16 | efried | I'm suggesting you do it for $service=glance from $your_project=cyborg | |
| 09:41:50 | efried | https://github.com/openstack/nova/commit/5601c82915a2d745ee600a4ca4c0e7215206dbb6 <== Here's what I was referring to earlier. Nova got rid of that chunk of ugliness. Your shortest path may be to subsume the same change. | |
| 09:43:02 | efried | You'll need to bump your ksa minimum to 3.3.0 (which I guess had already happened by the time ^ went in) | |
| 09:44:08 | shaohe_feng | OK, any special extra configure in conf file for it? | |
| 09:45:14 | efried | I don't think so. In fact, I think you should be able to leave the [glance] section *empty* and it should "just work" | |
| 09:45:52 | efried | Here's the commit that cut nova over to using the sdk for placement: https://github.com/openstack/nova/commit/2a2379a63cab4414676882c068d25163874cf08e | |
| 09:45:52 | efried | It was built on top of the commit(s) that introduced get_sdk_adapter, which I linked above. But you can see, once that was in place, this change was very simple. | |
| 09:46:13 | shaohe_feng | that's good. this is more easy for deployer with less configure. | |
| 09:46:23 | efried | In fact, on the code side, it consisted only of s/get_ksa_adapter/get_sdk_adapter/ -- the rest was just test fixups. | |
| 09:46:52 | shaohe_feng | https://github.com/openstack/cyborg/blob/6dab512af226677c9065f85f264a691f3aeb6eab/cyborg/common/utils.py#L185 | |
| 09:47:31 | shaohe_feng | ^ in your early code, you do some extra handle for image. | |
| 09:47:47 | efried | Oh, yeah, that's super stale. See the commit above for things you can do at a minimum to clean that up. | |
| 09:47:56 | shaohe_feng | after sdk, can we remove it? | |
| 09:48:09 | shaohe_feng | OK | |
| 09:48:33 | efried | Yeah, if you cut over to sdk, you wouldn't need get_endpoint *or* the things that call it. | |
| 09:49:01 | shaohe_feng | sounds good. let me try. | |
| 09:49:13 | efried | Hey, cyborg already has get_sdk_adapter :P https://github.com/openstack/cyborg/blob/6dab512af226677c9065f85f264a691f3aeb6eab/cyborg/common/utils.py#L135 | |
| 09:49:35 | efried | you may want to just make sure it's in sync with what's in nova. | |
| 09:50:37 | efried | by the way, long-term, you'll want to use the higher-level interfaces that sdk provides to talk to glance, rather than the REST primitives. Depending how much bandwidth you have, and how disruptive whatever you're working on is going to be, you may be able to do some of that now. | |
| 09:51:59 | efried | https://docs.openstack.org/openstacksdk/latest/user/resources/image/v2/image.html | |
| 09:52:23 | efried | Again, the folks in #openstack-sdks would be delighted to help you with this. | |
| 09:53:12 | shaohe_feng | good thanks | |
| 09:53:36 | efried | mordred (US) and gtema (EU I think) would be good resources there. | |