Handling of ties in row_number in Pyspark vs SQL

I have a table containing following columns:
year, subject, marks, city

Let's say it contains following values:

year  subject marks student name
2023  Maths   91    Jon
2023  Maths   71    Dany
2023  Maths   71    Rob
2023  Maths   85    Joffery

Lets say i perform

.withColumn('Ranking', f.row_number().over(\
        Window.partitionBy('year', 'subject').orderBy(f.col('marks').asc())))

As Dany and Rob have same marks, who will get 'Ranking' 1

Will it yeild different result if i run multiple times?

How such "ties" are handled in PySpak vs redshift SQL

In Apache Spark, if the row number and order by value are the same, the order of the rows is not guaranteed. Spark does not provide a deterministic order when the order by values are identical.

The reason for this behavior is that Spark's processing is distributed across multiple nodes, and the order of data processing and aggregation is not guaranteed to be consistent across different partitions or nodes. As a result, when multiple rows have the same order by value, their relative order might vary between different executions or runs of the same query.

If you need a consistent order for rows with the same order by value, you should include additional columns in the order by clause to break the tie. By specifying additional columns, you can ensure a deterministic order. For example:

.withColumn('Ranking', f.row_number().over(\
        Window.partitionBy('year', 'subject').orderBy(f.col('marks'), f.col('name'))))

By including an additional column(s) in the order by clause, you can achieve a well-defined order for rows with the same order by value.