[SETUP-BUG]

[MohamedKHALILRouissi]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): 22.04
• MLflow installed from (source or binary): source
• MLflow version (run mlflow --version): 2.14.2
• Python version: 3.8

Code to reproduce issue

yarn build serve -s build -l 3000

Describe the problem

hello i have successfully isolated the component of mlflow ui and mlflow api into 2 separated container and 1 pg database for data persistent , after running the dockerfile mlflow ui using the simple way copy all the folder of mlflow/server/js to the container and run yarn install & yarn start , this cause the container to consume 3.9 gb of memory and at booting the cpu exceed 450% for long duration , so i decided to work with the optimized build production , after building the ui using the yarn build , i have also installed serve as global package , and executed server -s build -l 3000 ( when running the command am at this folder mlflow/server/js , the ui doesn't show any thing it blank white page

i need help how to run the ui using the build instead of yarn start and thanks ,

Other info / logs

No response

[WeichenXu123]

so i decided to work with the optimized build production , after building the ui using the yarn build , i have also installed serve as global package , and executed server -s build -l 3000

I guess the modified process causes the issue. @daniellok-db do you have some ideas ?

[MohamedKHALILRouissi]

another issue

[MohamedKHALILRouissi]

logs

[daniellok-db]

If you want to do this, I think you'll probably need to set up a lightweight server that defines some proxies.

Requests for /static-files/* should be proxied to the build folder (or you can just rename build to static-files and serve from mlflow/server/js instead)

You'll have to proxy the following routes to wherever you're serving the actual mlflow server, otherwise the HTTP requests that the frontend sends to fetch data will fail:

https://github.com/mlflow/mlflow/blob/399cef1e03008497cc1c47ca67636ca78f05f604/mlflow/server/js/src/setupProxy.js#L10

[MohamedKHALILRouissi]

[daniellok-db]

did you set up the proxies? this won't work with just serve -s build

[MohamedKHALILRouissi]

can you please further explain the proxy functionality ?

[MohamedKHALILRouissi]

renamed the folder to static-files and nothing happen

[daniellok-db]

A few important points:

1. The site is built assuming that the files are hosted in a directory called static-files. So if your website is hosted at 192.168.1.121:3000, then 192.168.1.121:3000/static-files/ should correspond to the folder mlflow/server/js/build/. You can see how the MLflow Flask server is set up to serve the frontend here: https://github.com/mlflow/mlflow/blob/1ae83e4e4b733e7cb16c5b486c7f1157df5541d6/mlflow/server/__init__.py#L128-L133 You'll need to set up something similar.
2. Your backend and frontend are not deployed in the same server, but the UI will be making requests to your frontend server. In order to make this work, your frontend server needs to redirect requests from 192.168.1.121:3000/ajax-api/* to <backend server address>/ajax-api/*. Please see the following for all routes you need to redirect. https://github.com/mlflow/mlflow/blob/399cef1e03008497cc1c47ca67636ca78f05f604/mlflow/server/js/src/setupProxy.js#L13-L33 You can also do this in your lightweight Flask app (or whatever web framework you want to use).

[github-actions[bot]]

@mlflow/mlflow-team Please assign a maintainer and start triaging this issue.