MLFlow active run does not match environment run id

Question

I am trying to perform an MLFlow run but stuck with the following error after trying a lot of things.


run = mlflow.active_run()
if run:
    print("Active run_id: {}".format(run.info.run_id))
    mlflow.end_run()

mlflow.set_experiment('TNF_EXP') 
mlflow.set_tracking_uri("http://localhost:5000/") # Actual Server URI instead of localhost
experiment = mlflow.get_experiment_by_name("TNF_EXP")

with mlflow.start_run(experiment_id=experiment.experiment_id) as run:
...
...

mlflow.end_run()

Error -

File "/.../ModelTrainer.py", line 108, in train
    with mlflow.start_run(experiment_id=experiment.experiment_id) as run:
  File "/usr/local/lib/python3.6/site-packages/mlflow/tracking/fluent.py", line 207, in start_run
    "arguments".format(existing_run_id)
mlflow.exceptions.MlflowException: Cannot start run with ID e9953eb5918845bb9be1xxxxxx because active run ID does not match environment run ID. Make sure --experiment-name or --experiment-id matches experiment set with set_experiment(), or just use command-line arguments
2021/02/11 09:41:36 ERROR mlflow.cli: === Run (ID 'e9953eb5918845bb9be1xxxxxx') failed ===

I noticed I had an active run earlier so I included the first if block to end that run. The code ran successfully and I was able to log the data on MLFlow UI but now when I run it I start getting the same issue. There are no active runs found before starting a new run currently.

FYI, I am running the code on Azure server with the respective tracking URI mentioned in the code.

However the code runs fine if I include an argument --experiment-name="TNF_EXP" in the mlflow run command on the CLI

score 3 · Accepted Answer · answered Mar 25 '21 at 07:27

That is primarily because you have started a run with default experiment name and then you are trying to set the experiment_name as "TNF_EXP".

Will suggest you to make use of mlflow.run(..., experiment_name="TNF_EXP") python method then running it from the CLI.

You can find more information here.

score 0 · Answer 2 · answered Dec 15 '21 at 21:21

The current supported behavior for MLflow projects is to define the experiment name or id (if you know the id) using mlflow cli.

It would require systemic change in the way MLflow projects are executed since mlflow run CLI command will create a main run (under --experiment-name argument or default). The nested run created in main with start_run needs to belong to the same parent experiment.

The only supported models is use --experiment-name or --experiment-id in CLI command if you want log under a specific experiment.

You can now run it as : mlflow run . --experiment-name test

Welcome to stackoverflow! Could you add a source or link to the documentation of the behavior you describe? — anestv, Dec 16 '21 at 01:30

MLFlow active run does not match environment run id

2 Answers2