Is it possible to dynamically pass a value into all metaflow k8s jobs? I have users in JupyterHub running on GKE, who are using Metaflow to do data processing and model training. In the model training jobs, they are utilizing MLflow to log model experiment details. MLflow populates their

user

field by the

MLFLOW_USERNAME

environment variable. In JupyterHub, I have the user pods dynamically setting this environment variable so I can track who is training what models. However when users launch jobs using

python file.py run --with kubernetes

, their MLflow user reports back as

root

. What I would like to see is their user (i.e. dynamically change the

MLFLOW_USERNAME

value on their jobs) to be

zbloss

for example.

it's similar to this previous questions of yours, right? you could have a custom decorator as outlined in that thread setting

MFLOW_USERNAME

Yeah it maybe it's worth throwing that config into a configmap that gets auto-mounted

you could do that, or you can just use Metaflow configs

That's what I was thinking. Putting these configs into a configmap then using @environment to point to that configuration