Python and Windows Named Pipes

Question

In order to connect to an existing named pipe you can utilize the CreateFile API provided through the pywin32 package. Since it took me a while to put a working base together here is an example client/server which works fine for me (python 3.6.5, pywin32 223 on Windows 10 Pro x64):

import time
import sys
import win32pipe, win32file, pywintypes


def pipe_server():
    print("pipe server")
    count = 0
    pipe = win32pipe.CreateNamedPipe(
        r'\\.\pipe\Foo',
        win32pipe.PIPE_ACCESS_DUPLEX,
        win32pipe.PIPE_TYPE_MESSAGE | win32pipe.PIPE_READMODE_MESSAGE | win32pipe.PIPE_WAIT,
        1, 65536, 65536,
        0,
        None)
    try:
        print("waiting for client")
        win32pipe.ConnectNamedPipe(pipe, None)
        print("got client")

        while count < 10:
            print(f"writing message {count}")
            # convert to bytes
            some_data = str.encode(f"{count}")
            win32file.WriteFile(pipe, some_data)
            time.sleep(1)
            count += 1

        print("finished now")
    finally:
        win32file.CloseHandle(pipe)


def pipe_client():
    print("pipe client")
    quit = False

    while not quit:
        try:
            handle = win32file.CreateFile(
                r'\\.\pipe\Foo',
                win32file.GENERIC_READ | win32file.GENERIC_WRITE,
                0,
                None,
                win32file.OPEN_EXISTING,
                0,
                None
            )
            res = win32pipe.SetNamedPipeHandleState(handle, win32pipe.PIPE_READMODE_MESSAGE, None, None)
            if res == 0:
                print(f"SetNamedPipeHandleState return code: {res}")
            while True:
                resp = win32file.ReadFile(handle, 64*1024)
                print(f"message: {resp}")
        except pywintypes.error as e:
            if e.args[0] == 2:
                print("no pipe, trying again in a sec")
                time.sleep(1)
            elif e.args[0] == 109:
                print("broken pipe, bye bye")
                quit = True


if __name__ == '__main__':
    if len(sys.argv) < 2:
        print("need s or c as argument")
    elif sys.argv[1] == "s":
        pipe_server()
    elif sys.argv[1] == "c":
        pipe_client()
    else:
        print(f"no can do: {sys.argv[1]}")

Example output client

> python pipe_test.py c
pipe client
no pipe, trying again in a sec
no pipe, trying again in a sec
no pipe, trying again in a sec
message: (0, b'0')
message: (0, b'1')
message: (0, b'2')
message: (0, b'3')
message: (0, b'4')
message: (0, b'5')
message: (0, b'6')
message: (0, b'7')
message: (0, b'8')
message: (0, b'9')
broken pipe, bye bye

Example output server

> python pipe_test.py s
pipe server
waiting for client
got client
writing message 0
writing message 1
writing message 2
writing message 3
writing message 4
writing message 5
writing message 6
writing message 7
writing message 8
writing message 9
finished now

Obviously you’d need some error checking around the various calls but that should work.

Additional side note: A colleague of mine ran into trouble with the pipe being closed the moment the client tried to perform I/O on it (exception claiming that “all pipe instances are busy”). It turned out that he was using os.path.exists in the client code to test whether the named pipe already existed before running CreateFile on it. This somehow breaks the pipe. So using the approach above (CreateFile wrapped in a try-except) is the safe way of trying to connect to a pipe until it has been created by the server end.

Leave a Comment Cancel reply