Should I only use partition columns inside predicate with mode="overwrite" & engine="rust"?

[nmishturak-gl]

Hey, I have two INT partition columns TypeId, Year as well as INT GroupId (non-partitioned).
I'm trying to add new data / replace existing by these three columns with a predicate.

 predicate = f"""
        TypeId = CAST({type_id} AS INT)
        AND Year = CAST({year} AS INT)
        AND GroupId = CAST({group_id} as INT)
    """

It seems to add new data for TypeId=30, Year=2024,GroupId=123 and TypeId=20, Year=2024,GroupId=123 (both get 152 rows).
But when I try to replace data for TypeId=30, Year=2024,GroupId=123 the number of rows increases, and the old data is not overwritten. Instead it's added back for some reason.

Here how parts of the log looks like

0000000000 - add - TypeId=30/Year=2024/part-00001-cbf3e582-5365-4530-8024-8596342a70a2-c000.snappy.parquet
(added some data for TypeId = 30)

0000000001 - add - TypeId=20/Year=2024/part-00001-f225a81c-6946-45b8-9677-593c832480bc-c000.snappy.parquet
(added some data for TypeId = 20)

0000000002 - add - TypeId=30/Year=2024/part-00001-a529c30b-5ad2-4dfe-a9b1-cd55af962670-c000.snappy.parquet
(updated data for TypeId = 30)

0000000002 - add - TypeId=30/Year=2024/part-00001-c4483eee-6eb3-4d26-87b3-c2b704ec911b-c000.snappy.parquet
(contains old data for TypeId = 30 - added back for some reason)

0000000002 - remove - TypeId=30/Year=2024/part-00001-cbf3e582-5365-4530-8024-8596342a70a2-c000.snappy.parquet
(contains old data for Type=30 - removed correctly)

Do I have to use overwrite/replaceWhere only for entire partitions?