SNOW-770678: Fix incorrect date conversion in bulk insertion (#1562)

Author: sfc-gh-aling

DESCRIPTION.md

@@ -17,6 +17,7 @@ Source code is also available at: https://github.com/snowflakedb/snowflake-conne
   - Fixed a bug when `_prefetch_hook()` was not called before yielding results of `execute_async()`.
   - Fixed a bug where some ResultMetadata fields were marked as required when they were optional.
   - Bumped pandas dependency from <1.6.0,>=1.0.0 to >=1.0.0,<2.1.0
+  - Fixed a bug where bulk insert converts date incorrectly.
 
 - v3.0.3(April 20, 2023)
   - Fixed a bug that prints error in logs for GET command on GCS.

src/snowflake/connector/connection.py

@@ -57,7 +57,7 @@
     OCSPMode,
     QueryStatus,
 )
-from .converter import SnowflakeConverter
+from .converter import SnowflakeConverter, _adjust_bind_type
 from .cursor import LOG_MAX_QUERY_LENGTH, SnowflakeCursor
 from .description import (
     CLIENT_NAME,
@@ -1261,13 +1261,13 @@ def _process_params_qmarks(
                 if all(param_data.type == first_type for param_data in all_param_data):
                     snowflake_type = first_type
                 processed_params[str(idx + 1)] = {
-                    "type": snowflake_type,
+                    "type": _adjust_bind_type(snowflake_type),
                     "value": [param_data.binding for param_data in all_param_data],
                 }
             else:
                 snowflake_type, snowflake_binding = get_type_and_binding(v)
                 processed_params[str(idx + 1)] = {
-                    "type": snowflake_type,
+                    "type": _adjust_bind_type(snowflake_type),
                     "value": snowflake_binding,
                 }
         if logger.getEffectiveLevel() <= logging.DEBUG:

src/snowflake/connector/converter.py

@@ -101,10 +101,6 @@ def _convert_datetime_to_epoch_nanoseconds(dt: datetime) -> str:
     return f"{convert_datetime_to_epoch(dt):f}".replace(".", "") + "000"
 
 
-def _convert_date_to_epoch_milliseconds(dt: datetime) -> str:
-    return f"{(dt - ZERO_EPOCH_DATE).total_seconds():.3f}".replace(".", "")
-
-
 def _convert_time_to_epoch_nanoseconds(tm: dt_t) -> str:
     return (
         str(tm.hour * 3600 + tm.minute * 60 + tm.second)
@@ -366,8 +362,8 @@ def _nonetype_to_snowflake_bindings(self, *_) -> None:
         return None
 
     def _date_to_snowflake_bindings(self, _, value: date) -> str:
-        # milliseconds
-        return _convert_date_to_epoch_milliseconds(value)
+        # we are binding "TEXT" value for DATE, check function _adjust_bind_type
+        return value.isoformat()
 
     def _time_to_snowflake_bindings(self, _, value: dt_t) -> str:
         # nanoseconds
@@ -735,3 +731,11 @@ def create_timestamp_from_string(
         if not tz:
             return datetime.utcfromtimestamp(seconds) + timedelta(microseconds=fraction)
         return datetime.fromtimestamp(seconds, tz=tz) + timedelta(microseconds=fraction)
+
+
+def _adjust_bind_type(input_type: str | None) -> str | None:
+    # This is to address SNOW-7706788, binding "DATE" value can not go beyond date 2969-05-03 (31536000000)
+    # https://docs.snowflake.com/en/sql-reference/functions/to_date#usage-notes
+    # https://docs.snowflake.com/en/developer-guide/sql-api/submitting-requests
+    # to correctly bind DATE value, we adjust to use "TEXT" type to bind value
+    return input_type if input_type != "DATE" else "TEXT"

test/integ/test_bindings.py

@@ -431,6 +431,55 @@ def test_binding_bulk_insert(conn_cnx, db_parameters, num_rows):
             )
 
 
+@pytest.mark.skipolddriver
+def test_binding_bulk_insert_date(conn_cnx, db_parameters):
+    """Bulk insert test."""
+    with conn_cnx() as cnx:
+        cnx.cursor().execute(
+            """
+create or replace table {name} (
+    c1 date
+)
+""".format(
+                name=db_parameters["name"]
+            )
+        )
+    try:
+        with conn_cnx(paramstyle="qmark") as cnx:
+            c = cnx.cursor()
+            cnx._session_parameters[CLIENT_STAGE_ARRAY_BINDING_THRESHOLD] = 1
+            dates = [
+                [date.fromisoformat("1750-05-09")],
+                [date.fromisoformat("1969-12-31")],
+                [date.fromisoformat("1970-01-01")],
+                [date.fromisoformat("2023-05-12")],
+                [date.fromisoformat("2999-12-31")],
+                [date.fromisoformat("3000-12-31")],
+                [date.fromisoformat("9999-12-31")],
+            ]
+            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),),
+                (date(1970, 1, 1),),
+                (date(2023, 5, 12),),
+                (date(2999, 12, 31),),
+                (date(3000, 12, 31),),
+                (date(9999, 12, 31),),
+            ]
+    finally:
+        with conn_cnx() as cnx:
+            cnx.cursor().execute(
+                """
+drop table if exists {name}
+""".format(
+                    name=db_parameters["name"]
+                )
+            )
+
+
 @pytest.mark.skipolddriver
 def test_bulk_insert_binding_fallback(conn_cnx):
     """When stage creation fails, bulk inserts falls back to server side binding and disables stage optimization."""