Run /metrics on another port

[severo]

See #260 (comment)

See also how it's done in Go for the tensorboard launcher: https://github.com/huggingface/tensorboard-launcher/blob/46df45821e8311095e824bc39b9acdadaf99634c/launcher/pkg/cmd/launcher/main.go#L61

[severo]

Note that one issue is that we want to get metrics about the main app using the PrometheusMiddleware (https://github.com/huggingface/datasets-server/blob/cc763a31d588b88ce56a0ac6eccc6fd923a8a9eb/services/api/src/api/app.py#L48). Running the /metrics endpoint on a totally separate process would not give access to these metrics. That's why we run the endpoint inside the same app for now. Uvicorn does not (and won't) support binding apps to different ports.

[severo]

An option is to rewrite the api service in another language, maybe Go or Node.js: #493

Another one is to setup the reverse proxy.

[severo]

See a report from hackerone here: #496

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.

[severo]

not needed anymore (at least for now). Maybe take into account if we migrate to node (#493)