| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2018-08-22 | |||
| 14:24:34 | Coco_gao | Sundar, I don't know well on traits, so I don't know my OVO can meet the placement requirements. | |
| 14:25:04 | Sundar | Li_Liu: We should first agree on the os-acc flow, right? | |
| 14:25:44 | Sundar | Coco_gao: NP, I can help you. Power architecture developers want to support custom traits from the driver, as I mentione din os-acc spec | |
| 14:25:51 | Sundar | *sorry, discovery spec | |
| 14:26:19 | Coco_gao | By now, we don't have DB to store traits on Cyborg side. | |
| 14:26:40 | wangzhh | Sundar, not only flow but also how to handle different hypervisor, different platform, etc. | |
| 14:27:21 | Li_Liu | Sundar, but currently feature I am implementing does not involve os-acc at all, its the REST api to allow fpga programming | |
| 14:27:27 | Sundar | Coco_gao: at least for now, we can store custom traits as attributes | |
| 14:27:58 | Sundar | Li+Liu: So, there is no expectation that we should use the exact same flow for os-acc, right? | |
| 14:28:00 | Li_Liu | Coco_gao, yes, put them in the attributes list | |
| 14:28:00 | Coco_gao | OK. | |
| 14:28:02 | Sundar | wangzhh: Agreed | |
| 14:29:23 | wangzhh | We need to resolve: 1.report data to placement 2. os-acc flow(include some other things) 3. nova intergration | |
| 14:30:08 | Li_Liu | Sundar, yes, you can say that. but I think we could reuse that piece of code in os-acc anyways | |
| 14:30:23 | Li_Liu | wangzhh, exactly | |
| 14:30:31 | xinran | wangzhh yes | |
| 14:30:59 | shaohe_feng_robo | xinran: good at all 1, 2, 3 | |
| 14:31:30 | Li_Liu | 1.report data to placement -- Coco_gao 2. os-acc flow(include some other things) -- wangzhh 3. nova intergration -- xinran... are these assignments right? | |
| 14:32:09 | shaohe_feng_robo | Li_Liu: all can be xinran . | |
| 14:32:37 | Li_Liu | lol | |
| 14:32:42 | wangzhh | Wow,. | |
| 14:32:43 | Coco_gao | I think so | |
| 14:32:48 | xinran | 😂 | |
| 14:33:10 | xinran | I am starting design a new API for nova intergration | |
| 14:33:34 | shaohe_feng_robo | for she already do these things, we can avoid duplicated work. | |
| 14:33:35 | xinran | with lock and batch alloc support | |
| 14:33:37 | Li_Liu | wangzhh, can you help xinran when you have some time? | |
| 14:33:51 | wangzhh | Of course. | |
| 14:33:56 | sum12 | are there any implementation specific reviews of os-acc already ? | |
| 14:34:13 | shaohe_feng_robo | xinran: great | |
| 14:34:32 | Li_Liu | shaohe_feng_robo, it's great that xinran is the expert on all of these :) | |
| 14:34:49 | zhipeng | plz do remember that we develop and implement what we agreed for Rocky, at the moment | |
| 14:35:04 | zhipeng | anything beyond that is for design discussion starting at the Denver PTG | |
| 14:35:14 | shaohe_feng_robo | Li_Liu: she's new api is for the intergration | |
| 14:35:20 | wangzhh | sum12, Yes, but It's piecemeal. | |
| 14:35:31 | Li_Liu | so I am assuming there will a batch of patch reviews coming in from you right? xinran | |
| 14:35:37 | zhipeng | do not start development on those design specs that are not merged | |
| 14:35:57 | wangzhh | zhipeng, Agree. | |
| 14:36:00 | shaohe_feng_robo | but pleacement report depend on driver | |
| 14:36:08 | xinran | I just trying to understand if all these should be done in Rocky or not? | |
| 14:36:30 | shaohe_feng_robo | not sure all drivers are ready. | |
| 14:36:32 | sum12 | wangzhh: couple you point me to a piece ? I was expecting it to be a separate project | |
| 14:36:45 | shaohe_feng_robo | Sundar: what's the status of the new driver? | |
| 14:37:07 | xinran | yes placement report depend on driver, did coco already have some works on it ? | |
| 14:37:57 | shaohe_feng_robo | Li_Liu: for interaction, edleafe can help on nova side, | |
| 14:38:24 | sum12 | found it | |
| 14:39:01 | xinran | Li_Liu: yes, is it urgent? | |
| 14:39:21 | wangzhh | sum12, OK. :) | |
| 14:39:35 | Li_Liu | xinran, just do it on your own pace, but do you have a eta tho? | |
| 14:40:17 | openstackgerrit | Nguyen Hai proposed openstack/python-cyborgclient master: Update reno for stable/rocky https://review.openstack.org/586830 | |
| 14:40:19 | shaohe_feng_robo | Li_Liu: who report to placement? | |
| 14:40:22 | Coco_gao | what kind of data placement need? I just store all of the data. | |
| 14:40:29 | shaohe_feng_robo | agent or conductor? | |
| 14:40:39 | Sundar | shaohe_feng: The driver developer is apparently busy with other stuff. He has reached a stage where rpm and deb packages for OPAE libraries get packaged with the tree. I'll sync up with him and help in testing. | |
| 14:41:00 | Coco_gao | besides, the driver-specific traits is not support right now. | |
| 14:41:02 | openstackgerrit | Nguyen Hai proposed openstack/python-cyborgclient master: import zuul job settings from project-config https://review.openstack.org/592840 | |
| 14:41:11 | shaohe_feng_robo | Sundar: good. thanks . | |
| 14:41:11 | openstackgerrit | Nguyen Hai proposed openstack/python-cyborgclient master: switch documentation job to new PTI https://review.openstack.org/592841 | |
| 14:41:16 | openstackgerrit | Nguyen Hai proposed openstack/python-cyborgclient master: add python 3.6 unit test job https://review.openstack.org/592842 | |
| 14:41:21 | xinran | Li_Liu: i am afraid that it's quite hard to implement all in this release ... | |
| 14:41:54 | sum12 | xinran: agreed | |
| 14:41:57 | Coco_gao | yeah, I think the work to heavy for xinran | |
| 14:42:01 | Coco_gao | too | |
| 14:42:22 | Li_Liu | shaohe_feng_robo, agent report to cyborg, then cyborg uses the placement client to report to placement, right? | |
| 14:42:24 | Coco_gao | since the time is limited before Rocky | |
| 14:42:33 | xinran | Coco_gao: have you already do some work on placement report which related to your current work? | |
| 14:42:41 | shaohe_feng_robo | Li_Liu: OK. | |
| 14:43:20 | xinran | Li_Liu: yes, cyborg use placement client to report resource | |
| 14:43:21 | Li_Liu | xinran, I know, as I said, just do it on your own pace. but could you give estimations on them? | |
| 14:43:22 | Coco_gao | I didn't report to placement by now. I | |
| 14:44:37 | Coco_gao | I just collected the driver data, store in db. | |
| 14:45:20 | shaohe_feng_ | every card is a sub-provider | |
| 14:45:25 | wangzhh | Coco, After diff. :) | |
| 14:45:50 | xinran | Li_Liu: I will probably submit first version of new api allowing nova/cyborg interaction next week | |
| 14:46:29 | Li_Liu | Coco_gao, that should be good for now. next we will figure out during the reporting step, how to invoke placement client apis | |
| 14:46:31 | shaohe_feng_ | Li_Liu: the placement report does not need many code | |
| 14:46:46 | Li_Liu | xinran, sounds good | |
| 14:46:47 | shaohe_feng_ | just need to know the deloyable well. | |
| 14:47:07 | Li_Liu | shaohe_feng_, yup | |
| 14:47:11 | xinran | to be clear, did we agreed to improve the current API, I mean, add a new API like DeployableActions to do alloc and dealloc? | |
| 14:47:56 | wangzhh | xinran, I suggest to imporve them in S release. | |
| 14:47:59 | shaohe_feng_ | xinran: you can introduce your new API briefly | |
| 14:48:47 | Coco_gao | Li_Liu, where can I store the acc_topology data? | |
| 14:49:10 | xinran | wangzhh: yes that what i desired | |
| 14:49:17 | xinran | shaohe_feng_: sure | |
| 14:50:52 | Sundar | Folks, I want to clarify that all commits now should be considered provisional. We may need to change them substantially to meet the use cases and Nova requirements | |
| 14:51:04 | Li_Liu | Coco_gao, the acc_topology is not to be saved tho. It helps you to structure the deployable relationships | |
| 14:51:39 | sum12 | wangzhh: do you plan to pull in any patches for the Rocky release for os-acc ? | |
| 14:52:19 | Li_Liu | Coco_gao, in case it's not clear to you, we can discuss it further once you feel better | |
| 14:52:26 | shaohe_feng_ | Sundar: will you work on the image upload ? auto parser the bitstream files, and generate the glance metadata? | |
| 14:52:46 | wangzhh | sum12: No. | |
| 14:53:49 | xinran | I am trying to add a new API representing all actions of deployables e.g. allocation/deallocation. for example: def post(self, body) with body like {"FPGA":"vendor_id":"","type":""...,"GPU","vendor_id":""...}. And in this method, deployable_list will be call with filters. | |
| 14:54:54 | wangzhh | shaohe_feng_, Before upload image, we need download first, in agent. Do you think so? | |
| 14:54:57 | Sundar | Shaohe: Sure. | |
| 14:56:03 | xinran | To support batch alloc/dealloc, there will be a loop to call deployable_list(...,filters) that's my current solution. Does anyone of you have some better suggestion? | |
| 14:57:54 | wangzhh | xinran, for batch alloc/dealloc, we should have new APIs, and I think it's better to have a spec. | |
| 14:58:18 | wangzhh | We should disscuss how to handle failure. | |
| 14:58:45 | Li_Liu | wangzhh, you mean the looping case? | |
| 14:59:17 | wangzhh | Li_Liu, No, for the new design. | |
| 14:59:59 | Coco_gao | LI_Liu, I just mess up,can we list the remaining work and the timeline for that? | |