mlflow/mlflow · error · MlflowException

Failed to get credentials for DBFS; they are read from the D

Error message

Failed to get credentials for DBFS; they are read from the Databricks CLI credentials or MLFLOW_TRACKING* environment variables.

What it means

_get_host_creds_from_default_store retrieves credentials from the currently configured MLflow tracking store. If the active store is not a RestStore (e.g. a FileStore when tracking_uri is a local path), there is no way to obtain Databricks host credentials, so it raises MlflowException telling the user DBFS credentials must come from Databricks CLI config or MLFLOW_TRACKING* environment variables.

Source

Thrown at mlflow/store/artifact/dbfs_artifact_repo.py:191

                return []
            is_dir = dbfs_file["is_dir"]
            artifact_size = None if is_dir else dbfs_file["file_size"]
            infos.append(FileInfo(stripped_path, is_dir, artifact_size))
        return sorted(infos, key=lambda f: f.path)

    def _download_file(self, remote_file_path, local_path):
        self._dbfs_download(
            output_path=local_path, endpoint=self._get_dbfs_endpoint(remote_file_path)
        )

    def delete_artifacts(self, artifact_path=None):
        raise MlflowException("Not implemented yet")


def _get_host_creds_from_default_store():
    store = utils._get_store()
    if not isinstance(store, RestStore):
        raise MlflowException(
            "Failed to get credentials for DBFS; they are read from the "
            + "Databricks CLI credentials or MLFLOW_TRACKING* environment "
            + "variables."
        )
    return store.get_host_creds


def dbfs_artifact_repo_factory(
    artifact_uri: str, tracking_uri: str | None = None, registry_uri: str | None = None
):
    """
    Returns an ArtifactRepository subclass for storing artifacts on DBFS.

    This factory method is used with URIs of the form ``dbfs:/<path>``. DBFS-backed artifact
    storage can only be used together with the RestStore.

    In the special case where the URI is of the form
    `dbfs:/databricks/mlflow-tracking/<Exp-ID>/<Run-ID>/<path>',

View on GitHub (pinned to 6a27f2decc)

Solutions

Set the Databricks profile in ~/.databrickscfg (via `databricks configure`) or export MLFLOW_TRACKING_URI=databricks (plus DATABRICKS_HOST/DATABRICKS_TOKEN) before constructing the DBFS artifact repo.
Export MLFLOW_TRACKING_HOST and MLFLOW_TRACKING_TOKEN environment variables so HostCreds can be built without a RestStore.
Ensure the active tracking store is a RestStore (a server/databricks URI), not a local file: path, when using dbfs://profile@databricks/ artifact URIs.

Example fix

// before
mlflow.set_tracking_uri("./mlruns")  # FileStore
repo = get_artifact_repository("dbfs://profile@databricks/mnt/data")  # raises

// after
os.environ["MLFLOW_TRACKING_HOST"] = "https://adb-xxx.azuredatabricks.net"
os.environ["MLFLOW_TRACKING_TOKEN"] = "dapi..."
# or: mlflow.set_tracking_uri("databricks")

Defensive patterns

Strategy: validation

Validate before calling

import os
assert (os.environ.get("MLFLOW_TRACKING_HOST") and os.environ.get("MLFLOW_TRACKING_TOKEN")) or os.environ.get("MLFLOW_TRACKING_URI") == "databricks" or os.path.exists(os.path.expanduser("~/.databrickscfg")), "Configure Databricks credentials before using dbfs artifact repos"

Type guard

def has_dbfs_credentials() -> bool:
    import os
    return bool(os.environ.get("MLFLOW_TRACKING_HOST") and os.environ.get("MLFLOW_TRACKING_TOKEN")) or os.environ.get("MLFLOW_TRACKING_URI") == "databricks"

Try / catch

try:
    repo = get_artifact_repository(dbfs_uri)
except MlflowException as e:
    if "Failed to get credentials for DBFS" in str(e):
        os.environ["MLFLOW_TRACKING_HOST"] = host
        os.environ["MLFLOW_TRACKING_TOKEN"] = token
        repo = get_artifact_repository(dbfs_uri)

Prevention

Set MLFLOW_TRACKING_HOST/MLFLOW_TRACKING_TOKEN or run `databricks configure` once per environment.
Avoid mixing a local file-store tracking URI with dbfs://profile@databricks artifact URIs in the same process.
In CI, provision Databricks secrets as env vars before MLflow imports.

When it happens

Trigger: Creating a DbfsArtifactRepo (dbfs://profile@databricks/...) or calling host-cred-dependent DBFS operations while mlflow.set_tracking_uri points to a local store (e.g. './mlruns' file store), so utils._get_store() returns a FileStore instead of a RestStore.

Common situations: Local scripts that set tracking_uri to a local directory but log artifacts to dbfs:// URIs; tests or notebooks mixing local tracking with Databricks artifact storage; missing Databricks CLI profile and missing MLFLOW_TRACKING_HOST/MLFLOW_TRACKING_TOKEN env vars.

Related errors

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/028382cd92108c52. Report an issue: GitHub.