It would be nice to run mlflow tracking and mlflow ui as a central service to allow multiple user send their data to tracking API and view and share their results from central service.
For each user, it would be nice to give each of them an 'namespace', so each user only log to their own namespace. When accessing the UI, they pass the 'namespace' as a parameter and then the UI only show their own experiments.
Users can share their experiment result with a URL with have their namespace in it. Proper access control can be added later.
Are there plans to add multi-tenants support to MLFlow or is it out of scope ?
Sorry for the delayed response! We don't immediately plan to add this in open-source, but if there is a pluggable way to do this without requiring all implementors of the REST API / all client APIs to add a new parameter, we'd be open to hearing proposals (see the new feature request template).
I would also be interested in a solution for this. Are there any updates?
I don't know the existing architecture of mlflow in detail to design a good solution. If it is not possible to add multi-user support and access management to the existing components a gateway/proxy may could help.
At least we have to add a Authorization-Header to the REST API (or something similar) to identify the requesting user (= authentication). Also we have to categorize everything in namespaces (or name it "tenant", "space", "bucket", "context" ...) to avoid collisions (= authorization).
Nice to see this is at least considered. I was wondering, since i saw this:
We don't immediately plan to add this in open-source, but...
Is this actually something that enterprise plans include?
Any update regarding the multi-tenancy feature?
Most helpful comment
Any update regarding the multi-tenancy feature?