SpamAssassin via Python not always working?

Question

I've setup a mail server managed by a Python script via smtpd and since I receive some spam, I decided to connect SpamAssassin to it.

Since I couldn't find a Python code that connects to SpamAssassin to get the score, I built it myself using a few pieces I found online. Here's the code:

# -*- config:utf-8 -*-

import socket, select, re, logging
from io import BytesIO


divider_pattern = re.compile(br'^(.*?)\r?\n(.*?)\r?\n\r?\n', re.DOTALL)
first_line_pattern = re.compile(br'^SPAMD/[^ ]+ 0 EX_OK$')


# @see https://github.com/slimta/python-slimta/blob/master/slimta/policy/spamassassin.py
class SpamAssassin(object):
    def __init__(self, message, timeout=15):
        self.score = None
        self.symbols = None

        # Connecting
        client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        client.settimeout(timeout)
        client.connect(('127.0.0.1', 783))

        # Sending
        client.sendall(self._build_message(message))
        client.shutdown(socket.SHUT_WR)

        # Reading
        resfp = BytesIO()
        while True:
            ready = select.select([client], [], [], timeout)
            if ready[0] is None:
                # Kill with Timeout!
                logging.info('[SpamAssassin] - Timeout ({0}s)!'.format(str(timeout)))
                break

            data = client.recv(4096)
            if data == b'':
                break

            resfp.write(data)

        # Closing
        client.close()
        client = None

        self._parse_response(resfp.getvalue())

    def _build_message(self, message):
        reqfp = BytesIO()
        data_len = str(len(message)).encode()
        reqfp.write(b'SYMBOLS SPAMC/1.2\r\n')
        reqfp.write(b'Content-Length: ' + data_len + b'\r\n')
        reqfp.write(b'User: cx42\r\n\r\n')
        reqfp.write(message)
        return reqfp.getvalue()

    def _parse_response(self, response):
        if response == b'':
            logging.info("[SPAM ASSASSIN] Empty response")
            return None

        match = divider_pattern.match(response)
        if not match:
            logging.error("[SPAM ASSASSIN] Response error:")
            logging.error(response)
            return None

        first_line = match.group(1)
        headers = match.group(2)
        body = response[match.end(0):]

        # Checking response is good
        match = first_line_pattern.match(first_line)
        if not match:
            logging.error("[SPAM ASSASSIN] invalid response:")
            logging.error(first_line)
            return None

        self.symbols = [s.decode('ascii').strip() for s in body.strip().split(',')]

        headers = headers.replace(' ', '').replace(':', ';').replace('/', ';').split(';')
        self.score = float(headers[2])

    def get_score(self):
        return self.score

    def get_symbols(self):
        return self.symbols

    def is_spam(self, level=5):
        return self.score is None or self.score >= level

And on my server script, I have the following part to check for spam:

# data is the mail body received from smtpd
assassin = SpamAssassin(data)
if assassin.is_spam():
    logging.info('SpamAssassin rejected. Score of {0}'.format(assassin.get_score()))
    return '554 Command rejected for policy reasons.'

When the code don't go in the if condition, the mail is then sent.

The huge problem I'm having is that some emails are sent, even though they are considered as spam from SpamAssassin.

I know it because I've built a second script that loads the queue from postfix (via postqueue -j for JSON format), and do a SpamAssassin check from my code below. Quite a few emails are then detected as spam. (In order to not show too many code, the one to load the Postfix queue and clean it is here).

I don't know what is wrong in the code shown here, how is it possible for my Python code to allow spams to be sent.

I checked the logs and I don't have any exceptions from my code (like Timeouts from SpamAssassin or anything else).

For me, the issue is that the condition if assassin.is_spam(), returns False whereas it should return True in some case, but I don't know how/why/when, so I'm turning to you for help.

My theories are:

• Maybe the socket is re-using some cached version that returned False for SpamAssassin, on a new email, and fails the correct check
• Some odd stuff going on with socket/spamd.py file, since it's the only place that seems to not work.
• Maybe some concurrency issues? Since there is many requests made on the server, maybe one socket is opened for multiple incoming requests, and the first result is read for all the incoming requests, accepting mail that shouldn't?