What causes "Workspace '{workspace.name}' already exists." in mlflow/mlflow?

Call list_workspaces or get_workspace first and skip creation if the name already exists

mlflow/mlflow · error · MlflowException

RESOURCE_ALREADY_EXISTS

RESOURCE_ALREADY_EXISTS

Error message

Workspace '{workspace.name}' already exists.

What it means

RestWorkspaceStore.create_workspace expects HTTP 201; when the server answers with RESOURCE_ALREADY_EXISTS, the client re-raises it as an MlflowException with code RESOURCE_ALREADY_EXISTS, preserving the server's message when present. It means a workspace with the same name already exists on the server.

Source

Thrown at mlflow/store/workspace/rest_store.py:88

        if workspace.default_artifact_root is not None:
            request_message.default_artifact_root = workspace.default_artifact_root
        if workspace.trace_archival_location is not None:
            request_message.trace_archival_config.location = workspace.trace_archival_location
        if workspace.trace_archival_retention is not None:
            request_message.trace_archival_config.retention = workspace.trace_archival_retention
        try:
            proto = call_endpoint(
                host_creds=self.get_host_creds(),
                endpoint=WORKSPACES_ENDPOINT,
                method="POST",
                json_body=message_to_json(request_message),
                response_proto=CreateWorkspace.Response(),
                expected_status=201,
            )
        except RestException as exc:
            if exc.error_code == databricks_pb2.ErrorCode.Name(RESOURCE_ALREADY_EXISTS):
                message = exc.message or f"Workspace '{workspace.name}' already exists."
                raise MlflowException(message, RESOURCE_ALREADY_EXISTS) from exc
            raise

        return self._workspace_from_proto(proto)

    def update_workspace(self, workspace: Workspace) -> Workspace:
        request_message = UpdateWorkspace()
        if workspace.description is not None:
            request_message.description = workspace.description
        if workspace.default_artifact_root is not None:
            request_message.default_artifact_root = workspace.default_artifact_root
        if workspace.trace_archival_location is not None:
            request_message.trace_archival_config.location = workspace.trace_archival_location
        if workspace.trace_archival_retention is not None:
            request_message.trace_archival_config.retention = workspace.trace_archival_retention
        proto = call_endpoint(
            host_creds=self.get_host_creds(),
            endpoint=f"{WORKSPACES_ENDPOINT}/{_quote_workspace(workspace.name)}",
            method="PATCH",

View on GitHub (pinned to 6a27f2decc)

Solutions

Call list_workspaces or get_workspace first and skip creation if the name already exists
Catch MlflowException with error_code RESOURCE_ALREADY_EXISTS and treat it as a no-op if creation is meant to be idempotent
Use a unique workspace name (e.g. with a team/env suffix) to avoid collisions

Example fix

// before
create_workspace(Workspace(name="team-a"))  # raises if exists
// after
try:
    create_workspace(Workspace(name="team-a"))
except MlflowException as e:
    if e.error_code != "RESOURCE_ALREADY_EXISTS":
        raise

Defensive patterns

Strategy: try-catch

Validate before calling

existing = {w.name for w in store.list_workspaces()}
if name in existing:
    skip_creation = True

Try / catch

try:
    ws = store.create_workspace(Workspace(name=name))
except MlflowException as e:
    if e.error_code == "RESOURCE_ALREADY_EXISTS":
        ws = store.get_workspace(name)  # treat as idempotent
    else:
        raise

Prevention

Make provisioning scripts idempotent by checking existence first
Use get-before-create in CI/CD workspace setup
Adopt unique naming conventions per team/environment

When it happens

Trigger: POSTing a CreateWorkspace request for a name that is already registered on the tracking server; concurrent creation races where two processes create the same workspace name.

Common situations: Re-running idempotency-broken setup scripts that create workspaces; CI provisioning colliding with an existing workspace; teams sharing a server and independently creating the same workspace name.

Related errors

AI-assisted analysis of mlflow/mlflow@6a27f2decc (2026-08-29). Data as JSON: /api/errors/6aac58621556d5c9. Report an issue: GitHub.