It seems that if you want to use S3 storage, as well as configuring the documented parameters you need to ensure that the boto3 dependency is installed on your MLFlow server yourself.
See: mlflow/mlflow#1970 (comment)
@mirekphd the lack of boto3 dependency in core MLflow is indeed intentional to keep the set of MLflow dependencies minimal. Note that you can pip install mlflow[extras] and boto3 will be installed. Closing for now, but feel free to reopen - thanks!
See: mlflow/mlflow#8342 (comment)
Hi @asolimando, boto3 is deliberately excluded from the MLflow Server Docker image in order to keep dependencies minimal. Thank you for using MLflow!
It seems to be a design choice, but doesn't appear in the general documentation other than this one mention in a tutorial.
We can update the Docker deployment code to ensure that we install boto3 before starting the MLFlow server and we can follow up with a K8s specific PR too.
Originally posted by @peteski22 in #1276
Refs: #1279
Originally posted by @peteski22 in #1276
Refs: #1279