Question

I am using cPickle and psycopg2 to store some vectors into database.
This is my code to store binary data

binary_vec = cPickle.dumps(vec, -1)
db.cur.execute('''
               INSERT INTO feature_vector (vector, id) 
               VALUES (%s, %s);
               ''', (psycopg2.Binary(binary_vec), thread_id)
db.conn.commit()

However when I use fetchall() to load my data back, the type is buffer. I can’t find how how to restore this buffer object back to a list (vec).

This is how I fetch the data

db.cur.execute("SELECT * FROM feature_vector;")
m = db.cur.fetchall()

The result looks like this

[(3169187, <read-only buffer for 0x1002b0f10, size 3462, offset 0 at 0x1004a7430>), 
(3169275, <read-only buffer for 0x1002b0f50, size 3462, offset 0 at 0x1004a7570>), 
(3169406, <read-only buffer for 0x1002b0f70, size 3462, offset 0 at 0x10140b0b0>), 
(3169541, <read-only buffer for 0x10141c030, size 3462, offset 0 at 0x10140b2b0>), 
(3169622, <read-only buffer for 0x10141c050, size 3462, offset 0 at 0x10140b3f0>),...

When I try to use cPickle.loads(m[0][1]), it will return the error message

Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: must be string, not buffer

Answer 1

You can create a customized typecaster to automatically convert pickled values to Python:

import cPickle
obj = {'a': 10}
data = cPickle.dumps(obj, -1)

import psycopg2

def cast_pickle(data, cur):
    if data is None: return None
    return cPickle.loads(str(psycopg2.BINARY(data, cur)))

psycopg2.extensions.register_type(
    psycopg2.extensions.new_type(
        psycopg2.BINARY.values, 'BINARY-PICKLE', cast_pickle))

cnn = psycopg2.connect('')
cur = cnn.cursor()
cur.execute("select %s::bytea", [psycopg2.Binary(data)])
cur.fetchone()
# ({'a': 10},)