SQLalchemy rowcount always -1 for statements

Question

I was playing around with SQLalchemy and Microsoft SQL Server to get a hang of the functions when I came across a strange behavior. I was taught that the attribute rowcount on the result proxy object will tell how many rows were effected by executing a statement. However, when I select or insert single or multiple rows in my test database, I always get -1. How could this be and how can I fix this to reflect the reality?

connection = engine.connect()
metadata = MetaData()

# Ex1: select statement for all values
student = Table('student', metadata, autoload=True, autoload_with=engine)
stmt = select([student])
result_proxy = connection.execute(stmt)
results = result_proxy.fetchall()
print(result_proxy.rowcount)

# Ex2: inserting single values
stmt = insert(student).values(firstname='Severus', lastname='Snape')
result_proxy = connection.execute(stmt)
print(result_proxy.rowcout)
 
# Ex3: inserting multiple values 
stmt = insert(student)
values_list = [{'firstname': 'Rubius', 'lastname': 'Hagrid'},
               {'firstname': 'Minerva', 'lastname': 'McGonogall'}]
result_proxy = connection.execute(stmt, values_list)
print(result_proxy.rowcount)

The print function for each block seperately run example code prints -1. The Ex1 successfully fetches all rows and both insert statements successfully write the data to the database.

According to the following issue, the rowcount attribute isn't always to be trusted. Is that true here as well? And when, how can I compensate with a Count statement in a SQLalcehmy transaction? PDO::rowCount() returning -1

Answer 1

The single-row INSERT … VALUES ( … ) is trivial: If the statement succeeds then one row was affected, and if it fails (throws an error) then zero rows were affected.

For a multi-row INSERT simply perform it inside a transaction and rollback if an error occurs. Then the number of rows affected will either be zero or len(values_list) .

To get the number of rows that a SELECT will return, wrap the select query in a SELECT count(*) query and run that first, for example:

select_stmt = sa.select([Parent])
count_stmt = sa.select([sa.func.count(sa.text("*"))]).select_from(
    select_stmt.alias("s")
)
with engine.connect() as conn:
    conn.execution_options(isolation_level="SERIALIZABLE")
    rows_found = conn.execute(count_stmt).scalar()
    print(f"{rows_found} row(s) found")
    results = conn.execute(select_stmt).fetchall()
for item in results:
    print(item.id)