Question

I wrote a script that accessed a bunch of servers using nc on the command line, and originally I was using Python’s commands module and calling commands.getoutput(). The script ran in about 45 seconds. Since commands is deprecated, I want to change everything over to using the subprocess module, but now the script takes 2 min 45 secs to run. Why would this be?

What I had before:

output = commands.getoutput("echo get file.ext | nc -w 1 server.com port_num")

Now I have

p = Popen('echo get file.ext | nc -w 1 server.com port_num', shell=True, stdout=PIPE)
output = p.communicate()[0]

Answer 1

I would expect subprocess to be slower than command. Without meaning to suggest that this is the only reason your script is running slowly, you should take a look at the commands source code. There are fewer than 100 lines, and most of the work is delegated to functions from os, many of which are taken straight from C POSIX libraries (at least in POSIX systems). Note that commands is Unix-only, so it doesn’t have to do any extra work to ensure cross-platform compatibility.

Now take a look at subprocess. There are more than 1500 lines, all pure Python, doing all sorts of checks to ensure consistent cross-platform behavior. Based on this, I would expect subprocess to run slower than commands.

I timed the two modules, and on something quite basic, subprocess was almost twice as slow as commands.

>>> %timeit commands.getoutput('echo "foo" | cat')
100 loops, best of 3: 3.02 ms per loop
>>> %timeit subprocess.check_output('echo "foo" | cat', shell=True)
100 loops, best of 3: 5.76 ms per loop

Swiss suggests some good improvements that will help your script’s performance. But even after applying them, note that subprocess is still slower.

>>> %timeit commands.getoutput('echo "foo" | cat')
100 loops, best of 3: 2.97 ms per loop
>>> %timeit Popen('cat', stdin=PIPE, stdout=PIPE).communicate('foo')[0]
100 loops, best of 3: 4.15 ms per loop

Assuming you are performing the above command many times in a row, this will add up, and account for at least some of the performance difference.

In any case, I am interpreting your question as being about the relative performance of subprocess and command, rather than being about how to speed up your script. For the latter question, Swiss’s answer is better.