When running df.join, the different values of the how argument aren't always handled correclty:

df_1 = pd.DataFrame(
    {
        "A": ["A", "A", "B", "C"],
        "my_id": [1, 2, 3, 3],
    }
).to_dict("records")

df_2 = pd.DataFrame(
    {
        "B": ["A", "A", "B", "C"],
        "my_id": [1, 2, 3, 4],
    }
).to_dict("records")


df_1 = session.createDataFrame(df_1)
df_2 = session.createDataFrame(df_2)

• Default side

When default side isn't provided, it can cause an Exception downstream. For instance using DuckDB'qs backend, we get a Parser Error: syntax error at or near "OUTER" (Since no side is provided, which is mandatory here).
It may be fixed in sqlglot, but we can also map all correct values for how to values with default side added

• Output column selection

When running SEMI or ANTI join, the output selected columns fails since it includes columns from both tables.
In can be fixed by handling this case by only adding the self columns

I suggested an implementation here, but didn't test it thoroughly yet !