This is needed in order to ensure that created pods can access MLflow. This was initially suggested here by @misohu in the dss create PR.
What needs to get done
Write an integration test which will create a notebook and then it will run the script to create experiement in MLflow. e.g.
import mlflow
c = mlflow.MlflowClient() # You should not need to specify the tracking_uri - it should be populated automatically
print(c.tracking_uri) # should print the in-cluster url for dss's mlflow
c.create_experiment("test-experiment") # Should succeed
The idea is to create notebook and then run kubectl exec with that script. Or we could also execute a jupyter notebook from inside the pod (which could be git clone and run).
Something like this (from PR) could also be used to run command from the pod.
Why it needs to get done
This is needed in order to ensure that created pods can access MLflow. This was initially suggested here by @misohu in the
dss create
PR.What needs to get done
Write an integration test which will create a notebook and then it will run the script to create experiement in MLflow. e.g.
The idea is to create notebook and then run kubectl exec with that script. Or we could also execute a jupyter notebook from inside the pod (which could be
git clone
and run).Something like this (from PR) could also be used to run command from the pod.
When is the task considered done
There is integration test checking MLflow access.