Improve overall user experience of model service

[kyujin-cho]

Main idea

Since the birth of Backend.AI Model Service, main concern of the feature is that it is too hard to utilize for majority of users those who want to serve their own model. To overcome this problem we decided to add several new features on both Core and WebUI, which will potentially enhance overall experience of the Model Service feature.

• Core: New "Dry Run" API
  This new API should validate actual whole lifecycle of the inference session. Its request schema will be identical with the model service creation API. The implementation should first read model-definition.yml, create a new inference session accordingly but without the bound routing and endpoint, wait until model server loads, perform a health check (if defined at model definition), and finally terminate the created session. The API should report the whole progress to callee with the help of SSE.

Alternative ideas

No response

Anything else?

No response

Tasks

Preview Give feedback

1. feat: try-start model service API #1754
   comp:client comp:manager size:XL +3
   
2. Expose model-definition validator as a static validation schema (perhaps as a format of JSONSchema)
3. (on WebUI) Implement live model-definition.yml editor
4. (on WebUI) Introduce on-the-fly model-definition.yml schema validator for the editor