supertable 2.3.0__tar.gz → 2.3.1__tar.gz

{supertable-2.3.0/supertable.egg-info → supertable-2.3.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: supertable
-Version: 2.3.0
+Version: 2.3.1
 Summary: SuperTable — versioned data lake library for SQL analytics on Parquet + Redis.
 Author: Levente Kupas
 Author-email: Levente Kupas <lkupas@kladnasoft.com>

{supertable-2.3.0 → supertable-2.3.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "supertable"
-version = "2.3.0"
+version = "2.3.1"
 description = "SuperTable — versioned data lake library for SQL analytics on Parquet + Redis."
 readme = "README.md"
 requires-python = ">=3.10"

{supertable-2.3.0 → supertable-2.3.1}/setup.py

@@ -19,7 +19,7 @@ long_description = readme.read_text(encoding="utf-8") if readme.exists() else ""
 
 setup(
     name="supertable",
-    version="2.3.0",
+    version="2.3.1",
     description="SuperTable — versioned data lake library for SQL analytics on Parquet + Redis.",
     long_description=long_description,
     long_description_content_type="text/markdown",

{supertable-2.3.0 → supertable-2.3.1}/supertable/__init__.py

@@ -25,7 +25,7 @@ See the ``supertable.demo`` package for runnable end-to-end demos and the
 project documentation for the full API surface.
 """
 
-__version__ = "2.3.0"
+__version__ = "2.3.1"
 
 # Re-export the core public surface so users can do ``from supertable import …``
 # instead of remembering submodule paths.

{supertable-2.3.0 → supertable-2.3.1}/supertable/data_reader.py

@@ -314,8 +314,14 @@ class DataReader:
                         if isinstance(payload, dict):
                             tomb_path = payload.get("tombstone")
                             if tomb_path:
+                                # Resolve the deletion-vector key exactly like the
+                                # data files (estimator._to_duckdb_path, see
+                                # data_estimator.py:426): the catalog stores a bare
+                                # object key, which DuckDB/Spark cannot read against
+                                # an object store and must be presigned. LOCAL
+                                # storage returns the key unchanged.
                                 reflection.tombstone_views[td.alias] = TombstoneDef(
-                                    tombstone_path=tomb_path,
+                                    tombstone_path=estimator._to_duckdb_path(tomb_path),
                                 )
                     except Exception as te:
                         logger.debug(self._lp(f"[tombstone] leaf lookup failed for {td.alias}: {te}"))

{supertable-2.3.0 → supertable-2.3.1}/supertable/engine/spark_thrift.py

@@ -239,7 +239,11 @@ def _spark_create_tombstone_view(
     has_rowid = "__rowid__" in src_cols
 
     if tomb_path and has_rowid:
-        escaped = tomb_path.replace("'", "''")
+        # The data files are converted to s3a:// for Spark (see the data-file
+        # loop in the executor); apply the same conversion to the already
+        # resolved deletion-vector pointer so Spark reads it via its own S3A
+        # client instead of choking on a presigned HTTP URL or bare key.
+        escaped = _to_s3a_path(tomb_path).replace("'", "''")
         sql = (
             f"CREATE OR REPLACE TEMPORARY VIEW {view_name} AS "
             f"SELECT {select_cols} FROM {source_table} AS src "

{supertable-2.3.0 → supertable-2.3.1}/supertable/tests/test_data_reader.py

@@ -388,6 +388,127 @@ class TestExecuteHappyPath:
         assert exec_call_kwargs[1]["engine"] == _Engine.DUCKDB_PRO or exec_call_kwargs[0][0] == _Engine.DUCKDB_PRO
 
 
+# ====================================================================
+# 5b. DataReader.execute — Tombstone deletion-vector path resolution
+# ====================================================================
+
+class TestExecuteTombstoneResolution:
+    """Regression: the tombstone deletion-vector pointer must be resolved
+    through the SAME path as the data files (``estimator._to_duckdb_path``,
+    see data_estimator.py:426). Data files are presigned for object stores;
+    the tombstone key used to be embedded raw, so DuckDB/Spark could not read
+    it. These lock in 'if the data is presigned, the tombstone is too'.
+    """
+
+    @patch(_PATCH_EXTEND_PLAN)
+    @patch(_PATCH_EXECUTOR)
+    @patch(_PATCH_DATA_ESTIMATOR)
+    @patch(_PATCH_QUERY_PLAN_MGR)
+    @patch(_PATCH_RESTRICT_READ)
+    @patch(_PATCH_PLAN_STATS)
+    @patch(_PATCH_TIMER)
+    @patch(_PATCH_GET_STORAGE)
+    @patch(_PATCH_SQL_PARSER)
+    def test_tombstone_pointer_resolved_through_same_resolver_as_data(
+        self, MockParser, mock_get_storage, MockTimer, MockPlanStats,
+        mock_restrict, MockQPM, MockEstimator, MockExecutor, mock_extend,
+        _mock_data_reader_redis,
+    ):
+        from types import SimpleNamespace
+
+        td = SimpleNamespace(alias="t", super_name="s", simple_name="tbl")
+        mock_parser = MagicMock()
+        mock_parser.get_table_tuples.return_value = [td]
+        mock_parser.original_query = "SELECT * FROM tbl"
+        MockParser.return_value = mock_parser
+
+        mock_get_storage.return_value = MagicMock()
+        MockTimer.return_value = MagicMock(timings=[])
+        MockPlanStats.return_value = MagicMock()
+        MockQPM.return_value = MagicMock(query_id="q", query_hash="h")
+
+        # Catalog leaf carries a deletion-vector pointer (a bare object key).
+        raw_tomb_key = "o/s/tables/tbl/tombstone/123_abc_deleted.parquet"
+        _mock_data_reader_redis.return_value.get_leaf.return_value = {
+            "payload": {"tombstone": raw_tomb_key},
+        }
+
+        # Estimator resolver presigns (object-store behaviour) — the SAME method
+        # the estimator applies to data files at data_estimator.py:426.
+        reflection = _make_reflection()
+        mock_est = MagicMock()
+        mock_est.estimate.return_value = reflection
+        mock_est._to_duckdb_path.side_effect = lambda k: f"https://signed/{k}"
+        MockEstimator.return_value = mock_est
+
+        mock_exec = MagicMock()
+        mock_exec.execute.return_value = (pd.DataFrame(), "duckdb_pinned")
+        MockExecutor.return_value = mock_exec
+
+        from supertable.data_reader import DataReader, engine
+        dr = DataReader("s", "o", "SELECT * FROM tbl")
+        dr.execute("admin", engine=engine.AUTO)
+
+        # The raw key was routed through the resolver (not embedded as-is) ...
+        mock_est._to_duckdb_path.assert_called_once_with(raw_tomb_key)
+        # ... and the RESOLVED (presigned) value is what lands in the reflection.
+        assert "t" in reflection.tombstone_views
+        assert (
+            reflection.tombstone_views["t"].tombstone_path
+            == f"https://signed/{raw_tomb_key}"
+        )
+
+    @patch(_PATCH_EXTEND_PLAN)
+    @patch(_PATCH_EXECUTOR)
+    @patch(_PATCH_DATA_ESTIMATOR)
+    @patch(_PATCH_QUERY_PLAN_MGR)
+    @patch(_PATCH_RESTRICT_READ)
+    @patch(_PATCH_PLAN_STATS)
+    @patch(_PATCH_TIMER)
+    @patch(_PATCH_GET_STORAGE)
+    @patch(_PATCH_SQL_PARSER)
+    def test_local_storage_tombstone_left_unchanged(
+        self, MockParser, mock_get_storage, MockTimer, MockPlanStats,
+        mock_restrict, MockQPM, MockEstimator, MockExecutor, mock_extend,
+        _mock_data_reader_redis,
+    ):
+        # LOCAL contract: no presign → _to_duckdb_path returns the key as-is,
+        # so the tombstone path stays readable from disk (golden-test path).
+        from types import SimpleNamespace
+
+        td = SimpleNamespace(alias="t", super_name="s", simple_name="tbl")
+        mock_parser = MagicMock()
+        mock_parser.get_table_tuples.return_value = [td]
+        mock_parser.original_query = "SELECT * FROM tbl"
+        MockParser.return_value = mock_parser
+
+        mock_get_storage.return_value = MagicMock()
+        MockTimer.return_value = MagicMock(timings=[])
+        MockPlanStats.return_value = MagicMock()
+        MockQPM.return_value = MagicMock(query_id="q", query_hash="h")
+
+        raw_tomb_key = "local/dir/tombstone/x_deleted.parquet"
+        _mock_data_reader_redis.return_value.get_leaf.return_value = {
+            "payload": {"tombstone": raw_tomb_key},
+        }
+
+        reflection = _make_reflection()
+        mock_est = MagicMock()
+        mock_est.estimate.return_value = reflection
+        mock_est._to_duckdb_path.side_effect = lambda k: k  # LOCAL no-op
+        MockEstimator.return_value = mock_est
+
+        mock_exec = MagicMock()
+        mock_exec.execute.return_value = (pd.DataFrame(), "duckdb_pinned")
+        MockExecutor.return_value = mock_exec
+
+        from supertable.data_reader import DataReader, engine
+        dr = DataReader("s", "o", "SELECT * FROM tbl")
+        dr.execute("admin", engine=engine.AUTO)
+
+        assert reflection.tombstone_views["t"].tombstone_path == raw_tomb_key
+
+
 # ====================================================================
 # 6.  DataReader.execute — RBAC Failure
 # ====================================================================

{supertable-2.3.0 → supertable-2.3.1/supertable.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: supertable
-Version: 2.3.0
+Version: 2.3.1
 Summary: SuperTable — versioned data lake library for SQL analytics on Parquet + Redis.
 Author: Levente Kupas
 Author-email: Levente Kupas <lkupas@kladnasoft.com>