Python SQLAlchemy memory leak on Linux

I wrote a script that iterates through a large database table. (~ 150K lines.) To avoid using too much memory, I use this windowed_query method . My script looks something like this:

query = db.query(Table)

count = 0
for row in windowed_query(query, Table.id, 1000):

    points = 0

    # +100 points for a logo
    if row.logo_id:
        points += 100

    # +10 points for each image
    points += 10 * len(row.images) #images is a SQLAlchemy one-to-many relationship

    #...The script continues with much of the same...

    row.points = points
    db.add(row)

    count += 1
    if count % 100 == 0:
        db.commit()
        print count

request.db.commit()

      

        
        
        
      

    

When you try to run it on a CentOS server, it does so 9000 lines before being killed by the kernel because it uses ~ 2GB of memory .

In my Mac development environment, it works like a charm, even though it runs on exactly the same version of Python (2.7.3), SQLAlchemy (0.7.8) and psycopg2 (2.4.5).

Using memory_profiler for some simple debugging: On Linux, every piece of code that queries the database has increased the amount of memory by a small amount, and the growth has never stopped. On Mac, the same thing happened, but after growing ~ 4MB, it flattened out. It's as if Linux doesn't collect anything. (I even tried to run gc.collect () every 100 lines. Did nothing.)

Does anyone know what's going on?