mlflow/mlflow · error · Exception

Run with UUID {} is already active. To start a new run, firs

Error message

Run with UUID {} is already active. To start a new run, first end the current run with mlflow.end_run(). To start a nested run, call start_run with nested=True

What it means

mlflow.start_run() maintains an active-run stack; a plain (non-nested) start_run while a run is already active on the stack is ambiguous and rejected. The error names the currently active run UUID and points to end_run() or nested=True.

Source

Thrown at mlflow/tracking/fluent.py:567

    .. code-block:: text
        :caption: Output

        parent run:
        run_id: 8979459433a24a52ab3be87a229a9cdf
        description: starting a parent for experiment 7
        version tag value: v1
        priority tag value: P1
        --
        child runs:
                                     run_id params.child tags.mlflow.runName
        0  7d175204675e40328e46d9a6a5a7ee6a          yes           CHILD_RUN
    """
    active_run_stack = _active_run_stack.get()
    _validate_experiment_id_type(experiment_id)
    # back compat for int experiment_id
    experiment_id = str(experiment_id) if isinstance(experiment_id, int) else experiment_id
    if len(active_run_stack) > 0 and not nested:
        raise Exception(
            (
                "Run with UUID {} is already active. To start a new run, first end the "
                + "current run with mlflow.end_run(). To start a nested "
                + "run, call start_run with nested=True"
            ).format(active_run_stack[0].info.run_id)
        )
    client = MlflowClient()
    sgc_job_run_id_tag_key: str | None = None
    if run_id:
        existing_run_id = run_id
    elif run_id := MLFLOW_RUN_ID.get():
        existing_run_id = run_id
        del os.environ[MLFLOW_RUN_ID.name]
    # Get SGC job run ID tag key for run resumption if applicable
    elif sgc_job_run_id_tag_key := _get_sgc_job_run_id_tag_key():
        existing_run_id = _get_sgc_mlflow_run_id_for_resumption(
            client, experiment_id, sgc_job_run_id_tag_key
        )

View on GitHub (pinned to 6a27f2decc)

Solutions

Use `with mlflow.start_run(nested=True):` for the inner run.
Call mlflow.end_run() before starting a new top-level run, or use separate `with mlflow.start_run():` blocks.
Restructure so each run is properly closed (e.g., don't call start_run in a callback that runs inside an active run).
If a stale run is stuck from a crashed process, it no longer affects a new process; just ensure this process's stack is clean.

Example fix

// before
with mlflow.start_run():
    with mlflow.start_run():  # raises
        train()
// after
with mlflow.start_run():
    with mlflow.start_run(nested=True):
        train()

Defensive patterns

Strategy: validation

Validate before calling

if mlflow.active_run() is not None:
    mlflow.end_run()  # or use nested=True

Type guard

def has_active_run() -> bool:
    return mlflow.active_run() is not None

Try / catch

try:
    mlflow.start_run()
except Exception as e:
    if "already active" in str(e):
        mlflow.end_run()
        run = mlflow.start_run()
    else:
        raise

Prevention

Always use context managers (with mlflow.start_run():) so runs auto-close
Use nested=True for inner runs
Don't call start_run inside callbacks that run within an active run

When it happens

Trigger: Calling mlflow.start_run() (or the context manager) while _active_run_stack is non-empty and nested is not True — e.g., nested start_run calls in a loop, or re-entering start_run inside an active `with mlflow.start_run():` block.

Common situations: Calling start_run twice in the same script without ending the first run; wrapping training steps in start_run inside an outer start_run without nested=True; a callback (e.g., on_evaluate_start) starting a run while the framework already has one active.

Related errors

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/05662c1b6742f931. Report an issue: GitHub.