[SPARK-40835] to_utc_timestamp creates null column

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 3.2.0
• Component/s: PySpark
• Assignee: Unassigned
• Reporter: Rohan Barman
• Created: 18/Oct/22 17:22
• Updated: 19/Oct/22 02:03
• Resolved: 19/Oct/22 02:03

Description

We are in the process of migrating our PySpark applications from Spark version 3.1.2 to Spark version 3.2.0. 

This bug is present in version 3.2.0. We do not see this issue in version 3.1.2.

 

Minimal example to reproduce bug

Below is a minimal example of applying to_utc_timestamp() on String column with timestamp data

from pyspark.sql.types import StringType
from pyspark.sql.functions import *

# Source data
columns = ["id","timestamp_field"]
data = [("1", "2022-10-17T00:00:00+0000"), ("2", "2022-10-17T00:00:00+0000")]
source_df = spark.createDataFrame(data).toDF(*columns)
source_df.createOrReplaceTempView("source")
print("Source:")
print(source_df.show())

# Execute query
query = """
SELECT
    id, 
    timestamp_field as original,  
    to_utc_timestamp(timestamp_field, 'UTC')     AS received_timestamp
FROM source
"""
df = spark.sql(query)
print("Transformed:")
print(df.show())
print(df.count()) 

Post Execution

The source data has a column called timestamp_field which is a string type.

Source:
+---+--------------------+                                                      
| id|     timestamp_field|
+---+--------------------+
|  1|2022-10-17T00:00:...|
|  2|2022-10-17T00:00:...|
+---+--------------------+

The query applies to_utc_timestamp() on timestamp_field to create a new column. The new column is null.

Transformed:
+---+--------------------+------------------+
| id|            original|received_timestamp|
+---+--------------------+------------------+
|  1|2022-10-16T00:00:...|              null|
|  2|2022-10-16T00:00:...|              null|
+---+--------------------+------------------+ 

–

 

Questions

• Did the to_utc_timestamp function get any new changes in spark version 3.2.0? We don't see this issue in spark 3.1.2
• Can we apply any spark settings to resolve this?
• Is there a new preferred function in spark 3.2.0 that replaces to_utc_timestamp?

 

Issue Links

• duplicates: SPARK-37067 DateTimeUtils.stringToTimestamp() incorrectly rejects timezone without colon