This module is useful for finding unused network ports on a host. If you need legacy Python 2 support, use the 1.3.x releases.
This module provides a pure Python
pick_unused_port() function. It can also be
called via the command line for use in shell scripts.
If your code can accept a bound TCP socket rather than a port number consider
socket.bind(('localhost', 0)) to bind atomically to an available port
rather than using this library at all.
There is a race condition between picking a port and your application code binding to it. The use of a port server by all of your test code to avoid that problem is recommended on loaded test hosts running many tests at a time.
Unless you are using a port server, subsequent calls to
obtain an additional port are not guaranteed to return a unique port.
A port server is intended to be run as a daemon, for use by all processes
running on the host. It coordinates uses of network ports by anything using a
portpicker library. If you are using hosts as part of a test automation cluster,
each one should run a port server as a daemon. You should set the
PORTSERVER_ADDRESS=@unittest-portserver environment variable on all of your
test runners so that portpicker makes use of it.
A sample port server is included. This portserver implementation works but has not spent time in production. If you use it with good results please report back so that this statement can be updated to reflect that. :)
A port server listens on a unix socket, reads a pid from a new connection, tests the ports it is managing and replies with a port assignment port for that pid. A port is only reclaimed for potential reassignment to another process after the process it was originally assigned to has died. Processes that need multiple ports can simply issue multiple requests and are guaranteed they will each be unique.
import portpicker test_port = portpicker.pick_unused_port()
Or from the command line:
Or, if portpicker is installed as a library on the system Python interpreter:
TEST_PORT=`python3 -m portpicker $$`
This is not an official Google product (experimental or otherwise), it is just code that happens to be owned by Google.