feat(client): add methods to langfuse client to fetch and delete dataset runs

Author: marliessophie

langfuse/_client/client.py

@@ -2462,7 +2462,7 @@ def get_dataset(
             raise e
 
     def get_dataset_run(
-        self, dataset_name: str, *, run_name: str
+        self, *, dataset_name: str, run_name: str
     ) -> DatasetRunWithItems:
         """Fetch a dataset run by dataset name and run name.
 
@@ -2475,8 +2475,8 @@ def get_dataset_run(
         """
         try:
             return self.api.datasets.get_run(
-                dataset_name=self._url_encode(dataset_name),
-                run_name=self._url_encode(run_name),
+                dataset_name=self._url_encode(dataset_name, is_url_param=True),
+                run_name=self._url_encode(run_name, is_url_param=True),
                 request_options=None,
             )
         except Error as e:
@@ -2485,8 +2485,8 @@ def get_dataset_run(
 
     def get_dataset_runs(
         self,
-        dataset_name: str,
         *,
+        dataset_name: str,
         page: Optional[int] = None,
         limit: Optional[int] = None,
     ) -> PaginatedDatasetRuns:
@@ -2502,7 +2502,7 @@ def get_dataset_runs(
         """
         try:
             return self.api.datasets.get_runs(
-                dataset_name=self._url_encode(dataset_name),
+                dataset_name=self._url_encode(dataset_name, is_url_param=True),
                 page=page,
                 limit=limit,
                 request_options=None,
@@ -2512,7 +2512,7 @@ def get_dataset_runs(
             raise e
 
     def delete_dataset_run(
-        self, dataset_name: str, *, run_name: str
+        self, *, dataset_name: str, run_name: str
     ) -> DeleteDatasetRunResponse:
         """Delete a dataset run and all its run items. This action is irreversible.
 
@@ -2525,8 +2525,8 @@ def delete_dataset_run(
         """
         try:
             return self.api.datasets.delete_run(
-                dataset_name=self._url_encode(dataset_name),
-                run_name=self._url_encode(run_name),
+                dataset_name=self._url_encode(dataset_name, is_url_param=True),
+                run_name=self._url_encode(run_name, is_url_param=True),
                 request_options=None,
             )
         except Error as e:

tests/test_datasets.py

@@ -418,3 +418,54 @@ def execute_dataset_item(item, run_name):
         assert "args" in trace.input
         assert trace.input["args"][0] == expected_input
         assert trace.output == expected_input
+
+
+def test_dataset_runs_with_special_characters():
+    """Test that dataset runs work correctly with special characters in names."""
+    langfuse = Langfuse(debug=False)
+
+    # Test with various special characters that need URL encoding
+    dataset_name = f"test/dataset with spaces & special chars {create_uuid()[:5]}"
+    run_name = f"run/name with % and # {create_uuid()[:5]}"
+
+    langfuse.create_dataset(name=dataset_name)
+    input_data = json.dumps({"input": "Test data"})
+    langfuse.create_dataset_item(dataset_name=dataset_name, input=input_data)
+
+    dataset = langfuse.get_dataset(dataset_name)
+    assert len(dataset.items) == 1
+
+    # Create a dataset run with special characters in the run name
+    for item in dataset.items:
+        with item.run(
+            run_name=run_name,
+            run_metadata={"test": "value"},
+            run_description="Test run with special chars",
+        ):
+            pass
+
+    langfuse.flush()
+    time.sleep(1)
+
+    # Test get_dataset_runs with special characters in dataset name
+    runs = langfuse.get_dataset_runs(dataset_name=dataset_name)
+    assert len(runs.data) == 1
+    assert runs.data[0].name == run_name
+    assert runs.data[0].metadata == {"test": "value"}
+    assert runs.data[0].description == "Test run with special chars"
+
+    # Test get_dataset_run with special characters in both dataset and run name
+    run = langfuse.get_dataset_run(dataset_name=dataset_name, run_name=run_name)
+    assert run.run_name == run_name
+    assert run.dataset_name == dataset_name
+    assert len(run.dataset_run_items) == 1
+
+    # Test delete_dataset_run with special characters
+    delete_response = langfuse.delete_dataset_run(
+        dataset_name=dataset_name, run_name=run_name
+    )
+    assert delete_response.deleted_run_items_count == 1
+
+    # Verify the run was deleted
+    runs_after_delete = langfuse.get_dataset_runs(dataset_name=dataset_name)
+    assert len(runs_after_delete.data) == 0