grundprinzip commented on code in PR #39469:
URL: https://github.com/apache/spark/pull/39469#discussion_r1065778475
##########
python/pyspark/sql/connect/session.py:
##########
@@ -215,7 +215,31 @@ def createDataFrame(
_inferred_schema: Optional[StructType] = None
if isinstance(data, pd.DataFrame):
- _table = pa.Table.from_pandas(data)
+ from pandas.api.types import ( # type: ignore[attr-defined]
+ is_datetime64_dtype,
+ is_datetime64tz_dtype,
+ )
+
+ # We need double conversions for the truncation, first truncate to
microseconds.
+ for col in data:
+ print("Checking", col)
+ if is_datetime64tz_dtype(data[col].dtype):
+ data[col] = data[col].astype("datetime64[us, UTC]")
+ elif is_datetime64_dtype(data[col].dtype):
+ data[col] = data[col].astype("datetime64[us]")
+
+ # Create a new schema and change the types to the truncated
microseconds.
+ pd_schema = pa.Schema.from_pandas(data)
Review Comment:
Even modifying the schema, the Pandas -> Arrow conversion complains that the
type is nanoseconds. The pandas operation allows the truncation to actually
proceed.
If I skip the pandas operation the error is:
```
pyarrow.lib.ArrowInvalid: ('Casting from timestamp[ns] to timestamp[us]
would lose data: 1546329600000000500', 'Conversion failed for column aware with
type datetime64[ns, UTC]')
```
If I skip the second one the error is:
```
pyspark.sql.connect.client.SparkConnectException:
(org.apache.spark.SparkUnsupportedOperationException) Unsupported data type:
Timestamp(NANOSECOND, null)
```
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
