Different results parsing the same date in Windows and Linux - windows

Question

We're using pyspark to parse a dataset containing date columns converted to an UTC timestamp with a code that looks like this:
from datetime import datetime
from pyspark.sql import SparkSession
from pyspark.sql.functions import col, to_timestamp, to_utc_timestamp
session = (SparkSession
.builder
.appName('test_spark')
.master('local[*]')
.config('spark.sql.session.timeZone', 'UTC')
.getOrCreate())
date_df = session.createDataFrame([{'date': '2017-10-29'}])
result_df = (date_df
.withColumn('date', to_timestamp(col('date')))
.withColumn('date', to_utc_timestamp(col('date'), 'Europe/Madrid')))
expected_df = session.createDataFrame([{'date': datetime(2017, 10, 29, 0, 0)}])
print("Expected")
print(expected_df.collect())
print("Result")
print(result_df.collect())
In Linux, this code prints (Spark logging omitted):
Expected
[Row(date=datetime.datetime(2017, 10, 29, 0, 0))]
Result
[Row(date=datetime.datetime(2017, 10, 29, 0, 0))]
In Windows 10, however, the output is:
Expected
[Row(date=datetime.datetime(2017, 10, 29, 0, 0))]
Result
[Row(date=datetime.datetime(2017, 10, 29, 1, 0))]
As you can see, in Windows the date parsed have 1 hour more than the result in Linux.
I have tested this with both OpenJDK JRE and Oracle JRE (always version 8) but the result is the same.
What the hell is happening here? If I modify the year or the month of the dates, the "bug" no longer happens, looks like it only appears in 29 October of 2017, what makes this day so special?