I'm currently experimenting with Tor via the Stem python lib and using process.launch_tor_with_config() with a pool of processes, each using a different Control Port.

However as I increase pool size from 3 and above I'm seeing multiple OSError: Process terminated: No, it's still there. Exiting stack traces. Each process in the pool connects, sends out a HTTP request and terminates the connection, going on to the next request.

Is there a max number of concurrent connections or is the terminate killing more than a single tor connection?

  • Can you post the code you're using? Is there a particular reason you need more than one tor process? Wouldn't passing all the traffic through the same instance work as well? Commented Apr 1, 2017 at 22:53

1 Answer 1


The issue appears to be that the DataDirectory is ~/.tor by default. It contains a lock file, which is why stem believes that Tor is already running.

You can change the DataDirectory like this:

process.launch_tor_with_config({…, 'DataDirectory': '~/.tor/custom1'})

I recommend that you run only one Tor instance, reducing complexity and resource use. The only reason I can think of why that you would want multiple instances is if you wanted to be sure that a different circuit is used for every connection. If you want that, you can simply set a different proxy password for every connection. As far as I know, Tor Browser still uses this technique to get a different circuit for every first party domain.

Here is a short example:

import itertools
from stem import process
import requests

with process.launch_tor_with_config({'ControlPort': '3421', 'SocksPort': '3420'}) as tor:
    session = requests.session()

    def receive(password):
        proxies = {
            'http': 'socks5://user:{}@localhost:3420'.format(password),
            'https': 'socks5://user:{}@localhost:3420'.format(password)
        return session.get('https://ipinfo.io', proxies=proxies)

    print('== use 10 different circuits ==')
    for response in map(receive, range(10)):

    print('== use 10 times same circuit (same password, same circuit) ==')
    for response in map(receive, itertools.repeat('secret', 10)):


How many local concurrent Tor connections can I run?

I'm not sure if there is a limit for clients. There probably is a limit per guard node. This usually shouldn't be an issue though since a guard won't know anything about the connections you have made to the other guards. So, if a guard is randomly picked, you shouldn't be able hit the limit easily.

If you're running a relay, the limit is 2 per IP.

  • While I'm here @P. Gerber, what is the difference between process terminate & kill? Commented Apr 2, 2017 at 10:42
  • 1
    Terminate sends signal SIGTERM to the process in which case Tor, according to its manpage, does this: "Tor will catch this, clean up and sync to disk if necessary, and exit." If it gets killed, the process seizes to exist without Tor having a change to do any clean-up work. You probably want to make sure caches are written to disk, so they can be reused on the next start. If so, use terminate. Commented Apr 2, 2017 at 11:05
  • Mentioned code doesn't work. Tested with requests, the get is never completed, hangs forever; and also Selenium/Chrome, displays a "Page cannot be reached" error.
    – Veehmot
    Commented Aug 29, 2017 at 11:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .