CODEWITHSUNDEEP
pythonmultithreadingnetwork-programming
Tom
Tom

Reputation: 3183

Python Interpreter blocks Multithreaded DNS requests?

I just played around a little bit with python and threads, and realized even in a multithreaded script, DNS requests are blocking. Consider the following script:

from threading import Thread import socket

class Connection(Thread):
    def __init__(self, name, url):
        Thread.__init__(self)
        self._url = url
        self._name = name

    def run(self):
        print "Connecting...", self._name
        try:
            s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
            s.setblocking(0)
            s.connect((self._url, 80))
        except socket.gaierror:
            pass #not interested in it
        print "finished", self._name


if __name__ == '__main__':
    conns = []
    # all invalid addresses to see how they fail / check times
    conns.append(Connection("conn1", "www.2eg11erdhrtj.com"))
    conns.append(Connection("conn2", "www.e2ger2dh2rtj.com"))
    conns.append(Connection("conn3", "www.eg2de3rh1rtj.com"))
    conns.append(Connection("conn4", "www.ege2rh4rd1tj.com"))
    conns.append(Connection("conn5", "www.ege52drhrtj1.com"))

    for conn in conns:
        conn.start()

I dont know exactly how long the timeout is, but when running this the following happens:

  1. All Threads start and I get my printouts
  2. Every xx seconds, one thread displays finished, instead of all at once
  3. The Threads finish sequentially, not all at once (timeout = same for all!)

So my only guess is that this has to do with the GIL? Obviously the threads do not perform their task concurrently, only one connection is attempted at a time.

Does anyone know a way around this?

(asyncore doesnt help, and I'd prefer not to use twisted for now) Isn't it possible to get this simple little thing done with python?

Greetings, Tom

edit:

I am on MacOSX, I just let my friend run this on linux, and he actually does get the results I wished to get. His socket.connects()'s return immediately, even in a non Threaded environment. And even when he sets the sockets to blocking, and timeout to 10 seconds, all his Threads finish at the same time.

Can anyone explain this?

Upvotes: 8

Views: 5229

Answers (3)

Joe Koberg
Joe Koberg

Reputation: 26699

if it's suitable you could use the multiprocessing module to enable process-based parallelism

import multiprocessing, socket

NUM_PROCESSES = 5

def get_url(url):
    try:
        s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
        s.setblocking(0)
        s.connect((url, 80))
    except socket.gaierror:
        pass #not interested in it
    return 'finished ' + url


def main(url_list):
    pool = multiprocessing.Pool( NUM_PROCESSES )
    for output in pool.imap_unordered(get_url, url_list):
        print output

if __name__=="__main__":
    main("""
             www.2eg11erdhrtj.com 
             www.e2ger2dh2rtj.com
             www.eg2de3rh1rtj.com
             www.ege2rh4rd1tj.com
             www.ege52drhrtj1.com
          """.split())

Upvotes: 2

jfs
jfs

Reputation: 414149

Send DNS requests asynchronously using Twisted Names: 

import sys
from twisted.internet import reactor
from twisted.internet import defer
from twisted.names    import client
from twisted.python   import log

def process_names(names):
    log.startLogging(sys.stderr, setStdout=False)

    def print_results(results):
        for name, (success, result) in zip(names, results):
            if success:
                print "%s -> %s" % (name, result)
            else:
                print >>sys.stderr, "error: %s failed. Reason: %s" % (
                    name, result)

    d = defer.DeferredList(map(client.getHostByName, names), consumeErrors=True)
    d.addCallback(print_results)
    d.addErrback(defer.logError)
    d.addBoth(lambda _: reactor.stop())

reactor.callWhenRunning(process_names, """
    google.com
    www.2eg11erdhrtj.com 
    www.e2ger2dh2rtj.com
    www.eg2de3rh1rtj.com
    www.ege2rh4rd1tj.com
    www.ege52drhrtj1.com
    """.split())
reactor.run()

Upvotes: 2

Martin v. Löwis
Martin v. Löwis

Reputation: 127447

On some systems, getaddrinfo is not thread-safe. Python believes that some such systems are FreeBSD, OpenBSD, NetBSD, OSX, and VMS. On those systems, Python maintains a lock specifically for the netdb (i.e. getaddrinfo and friends).

So if you can't switch operating systems, you'll have to use a different (thread-safe) resolver library, such as twisted's.

Upvotes: 15

Related Questions