fix(datasource): handle numpy array tags from MariaDB in get_dataframe_by_tag

MariaDB round-trips tags through json/gzip/LONGBLOB, producing numpy
arrays instead of Python lists. The isinstance(cell, list) check
returned False, causing all rows to be filtered out. Extract tags via
_extract_tags helper that handles both list and np.ndarray.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: Sudip Sinha <Sudip.Sinha@RedHat.com>

Author: SudipSinha

src/service/data/datasources/data_source.py

@@ -3,6 +3,7 @@
 import logging
 import os
 
+import numpy as np
 import pandas as pd
 
 from src.service.constants import (
@@ -196,10 +197,14 @@ async def get_dataframe_by_tag(self, model_id: str, tag: str) -> pd.DataFrame:
             if tags_col < 0:
                 return pd.DataFrame()
 
-            mask = [
-                tag in (row[tags_col] if isinstance(row[tags_col], list) else [])
-                for row in metadata
-            ]
+            def _extract_tags(cell: object) -> list:
+                if isinstance(cell, np.ndarray):
+                    return cell.tolist()
+                if isinstance(cell, list):
+                    return cell
+                return []
+
+            mask = [tag in _extract_tags(row[tags_col]) for row in metadata]
             filtered_input = input_data[mask]
 
             df_data: dict[str, object] = {}

tests/service/data/datasources/test_datasource.py

@@ -524,3 +524,38 @@ async def test_returns_empty_when_no_data(
         df = await data_source.get_dataframe_by_tag("test-model", "TRAINING")
 
         assert len(df) == 0
+
+    @patch("src.service.data.datasources.data_source.ModelData")
+    @pytest.mark.asyncio
+    async def test_handles_numpy_array_tags_from_mariadb(
+        self, mock_model_data_class: Mock, data_source: DataSource
+    ) -> None:
+        """Tags stored as numpy arrays (MariaDB round-trip) are handled."""
+        mock = Mock(spec=ModelData)
+        mock.column_names = AsyncMock(
+            return_value=(
+                np.array(["feature1"]),
+                np.array(["output"]),
+                np.array(["id", "iso_time", "unix_timestamp", "tags"]),
+            ),
+        )
+        mock.data = AsyncMock(
+            return_value=(
+                np.array([[1.0], [2.0], [3.0]]),
+                np.array([[0.0], [1.0], [0.0]]),
+                np.array(
+                    [
+                        ["id_0", "t0", 0.0, np.array(["TRAINING"])],
+                        ["id_1", "t1", 1.0, np.array(["unlabeled"])],
+                        ["id_2", "t2", 2.0, np.array(["TRAINING"])],
+                    ],
+                    dtype=object,
+                ),
+            ),
+        )
+        mock_model_data_class.return_value = mock
+
+        df = await data_source.get_dataframe_by_tag("test-model", "TRAINING")
+
+        assert len(df) == 2  # noqa: PLR2004
+        assert df["feature1"].tolist() == [1.0, 3.0]