SNOW-966444: Fix date binding issue (#1822)

Author: sfc-gh-aling

src/snowflake/connector/converter.py

@@ -360,6 +360,18 @@ def _str_to_snowflake_bindings(self, _, value: str) -> str:
         # NOTE: str type is always taken as a text data and never binary
         return str(value)
 
+    def _date_to_snowflake_bindings_in_bulk_insertion(self, value: date) -> str:
+        # notes: this is for date type bulk insertion, it's different from non-bulk date type insertion flow
+        milliseconds = _convert_date_to_epoch_milliseconds(value)
+        # according to https://docs.snowflake.com/en/sql-reference/functions/to_date
+        # through test, value in seconds will lead to wrong date
+        # millisecond and nanoarrow second are good
+        # if the milliseconds is beyond the range of 31536000000000, we switch to use nanoseconds
+        # otherwise we will hit overflow error in snowflake
+        if int(milliseconds) < 31536000000000:
+            return milliseconds
+        return _convert_date_to_epoch_nanoseconds(value)
+
     _int_to_snowflake_bindings = _str_to_snowflake_bindings
     _long_to_snowflake_bindings = _str_to_snowflake_bindings
     _float_to_snowflake_bindings = _str_to_snowflake_bindings
@@ -378,15 +390,9 @@ def _nonetype_to_snowflake_bindings(self, *_) -> None:
         return None
 
     def _date_to_snowflake_bindings(self, _, value: date) -> str:
-        milliseconds = _convert_date_to_epoch_milliseconds(value)
-        # according to https://docs.snowflake.com/en/sql-reference/functions/to_date
-        # through test, value in seconds will lead to wrong date
-        # millisecond and nanoarrow second are good
-        # if the milliseconds is beyond the range of 31536000000000, we switch to use nanoseconds
-        # otherwise we will hit overflow error in snowflake
-        if int(milliseconds) < 31536000000000:
-            return milliseconds
-        return _convert_date_to_epoch_nanoseconds(value)
+        # this is for date type non-bulk insertion, it's different from bulk date type insertion flow
+        # milliseconds
+        return _convert_date_to_epoch_milliseconds(value)
 
     def _time_to_snowflake_bindings(self, _, value: dt_t) -> str:
         # nanoseconds
@@ -662,6 +668,11 @@ def to_csv_bindings(self, value: tuple[str, Any] | Any) -> str | None:
         else:
             if isinstance(value, (dt_t, timedelta)):
                 val = self.to_snowflake(value)
+            elif isinstance(value, date) and not isinstance(value, datetime):
+                # FIX SNOW-770678 and SNOW-966444
+                # bulk insertion congestion is different from non-bulk insertion
+                # to_csv_bindings is only used in bulk insertion logic
+                val = self._date_to_snowflake_bindings_in_bulk_insertion(value)
             else:
                 _type = self.snowflake_type(value)
                 val = self.to_snowflake_bindings(_type, value)

test/integ/test_bindings.py

@@ -449,6 +449,8 @@ def test_binding_bulk_insert_date(conn_cnx, db_parameters):
             c = cnx.cursor()
             cnx._session_parameters[CLIENT_STAGE_ARRAY_BINDING_THRESHOLD] = 1
             dates = [
+                [date.fromisoformat("1750-05-09")],
+                [date.fromisoformat("1969-01-01")],
                 [date.fromisoformat("1970-01-01")],
                 [date.fromisoformat("2023-05-12")],
                 [date.fromisoformat("2999-12-31")],
@@ -459,27 +461,14 @@ def test_binding_bulk_insert_date(conn_cnx, db_parameters):
             assert c.rowcount == len(dates)
             ret = c.execute(f'SELECT c1 from {db_parameters["name"]}').fetchall()
             assert ret == [
+                (date(1750, 5, 9),),
+                (date(1969, 1, 1),),
                 (date(1970, 1, 1),),
                 (date(2023, 5, 12),),
                 (date(2999, 12, 31),),
                 (date(3000, 12, 31),),
                 (date(9999, 12, 31),),
             ]
-            cnx.cursor().execute(f"TRUNCATE TABLE {db_parameters['name']}")
-            # TODO: bulk insert mixing date < 1970-01-01 and >= 1970-01-01 can return wrong result
-            #  here we split the insertion to make sure the client logic is correct
-            #  this could be a bug in snowflake
-            dates = [
-                [date.fromisoformat("1750-05-09")],
-                [date.fromisoformat("1969-12-31")],
-            ]
-            c.executemany(f'INSERT INTO {db_parameters["name"]}(c1) VALUES (?)', dates)
-            assert c.rowcount == len(dates)
-            ret = c.execute(f'SELECT c1 from {db_parameters["name"]}').fetchall()
-            assert ret == [
-                (date(1750, 5, 9),),
-                (date(1969, 12, 31),),
-            ]
     finally:
         with conn_cnx() as cnx:
             cnx.cursor().execute(

test/integ/test_cursor.py

@@ -816,15 +816,20 @@ def test_executemany_qmark_types(conn, db_parameters):
             cur.execute(f"create temp table {table_name} (birth_date date)")
 
             insert_qy = f"INSERT INTO {table_name} (birth_date) values (?)"
-            date_1, date_2 = date(1969, 2, 7), date(1969, 1, 1)
+            date_1, date_2, date_3, date_4 = (
+                date(1969, 2, 7),
+                date(1969, 1, 1),
+                date(2999, 12, 31),
+                date(9999, 1, 1),
+            )
 
             # insert two dates, one in tuple format which specifies
             # the snowflake type similar to how we support it in this
             # example:
             # https://docs.snowflake.com/en/user-guide/python-connector-example.html#using-qmark-or-numeric-binding-with-datetime-objects
             cur.executemany(
                 insert_qy,
-                [[date_1], [("DATE", date_2)]],
+                [[date_1], [("DATE", date_2)], [date_3], [date_4]],
                 # test that kwargs get passed through executemany properly
                 _statement_params={
                     PARAMETER_PYTHON_CONNECTOR_QUERY_RESULT_FORMAT: "json"
@@ -835,7 +840,7 @@ def test_executemany_qmark_types(conn, db_parameters):
             )
 
             cur.execute(f"select * from {table_name}")
-            assert {row[0] for row in cur} == {date_1, date_2}
+            assert {row[0] for row in cur} == {date_1, date_2, date_3, date_4}
 
 
 @pytest.mark.skipolddriver