Swagger - custom root/base url

[gregd33]

Is your feature request related to a problem? Please describe.

So I have a setup like this:

• one service running via docker-compose. For example, MyBentoService running on port 5555. In theory, I could have many services running on many ports.
• I have a server called e.g myservices.com
• I have nginx running which proxy_pass's /bento_services/MyBentoService/ to localhost:5555

So, if I call myservices.com/bento_services/MyBentoService/predict via python/requests, it all works fine.

If I go to myservices.com/bento_services/MyBentoService/, the Swagger UI pops up.

However, if I try to predict via the UI, it fails. The reason is that the Swagger page is sending the request to myservices.com/predict

Describe the solution you'd like

I don't know enough about how the Swagger page is generated to know how it might be fixed. But I can think of two possibilities:

• the swagger page can generate the URL by simply appending "predict" to the URL it is being accessed at (not sure if this is doable)
• an argument is passed to the serve-gunicorn which gives the base url (e.g. myservices.com/bento_services/MyBentoService/) so it knows how to construct the url

[yubozhao]

@gregd33 This is a good suggestion to enhance the experience. I think the second option is better at handling it for this case.

Can you tell me more about your current workflow? Are you using the docker-compose and nginx proxy setup in development? If you are using this setup in production, what role is swagger UI playing in that?

[gregd33]

The Swagger component is not that important at all - real processing would be done via the REST end points (which work fine).

The Swagger UI is mostly nice for demoing how a service works, especially since the service is on an image so its easy to upload one and get some results.

So this is low priority.

[gregd33]

Also if you can point me in the direction of how/where the Swagger page is generated, I can try to tackle the issue myself

[yubozhao]

The Swagger UI is mostly nice for demoing how a service works, especially since the service is on an image so its easy to upload one and get some results.

Yeah, that makes sense.

Also if you can point me in the direction of how/where the Swagger page is generated, I can try to tackle the issue myself

Yes, I am happy to show you.

To render the swagger UI on the '/' page, it went through these steps:

1. The index route will default to display swagger UI if the user didn't have any customized HTML. https://github.com/bentoml/BentoML/blob/439412fe394116c265a5199c5983ce60eb45e85b/bentoml/server/api_server.py#L250

2. Inside the render swagger UI function, we send the client a HTML that is based on a template with dynamically generated open API definition in JSON. https://github.com/bentoml/BentoML/blob/439412fe394116c265a5199c5983ce60eb45e85b/bentoml/server/api_server.py#L159

In the template, we use Swagger-UI to display the interactive elements based on the open API definition https://github.com/bentoml/BentoML/blob/439412fe394116c265a5199c5983ce60eb45e85b/bentoml/server/api_server.py#L46-L60

The open API definition(docs.json) is generated here: https://github.com/bentoml/BentoML/blob/439412fe394116c265a5199c5983ce60eb45e85b/bentoml/server/open_api.py#L20

[gregd33]

I was able to get a start of this and a PR is started in https://github.com/bentoml/BentoML/pull/1411

I still need to update the docs and add tests, if possible. I also need to make sure that I don't have extra changes - I originally went down a path of changes that changed the endpoint that was hosting the service not what Swagger called.

But I thought I'd share progress here. Hopefully will finish it up soon.

[yubozhao]

@gregd33 Awesome. Thank you for contributing! Let me know if I can help in anyway

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[parano]

Ultimately we decided to not support this use case for now. In BentoML 1.0, we have standardized how kubernetes deployments are created via Yatai, which will handle all the deployment endpoints management and ingress resources, making sure all swagger pages are available.