Python sockets buffering

Question

Let's say I want to read a line from a socket, using the standard socket module:

def read_line(s):
    ret = ''

    while True:
        c = s.recv(1)

        if c == '\n' or c == '':
            break
        else:
            ret += c

    return ret

What exactly happens in s.recv(1)? Will it issue a system call each time? I guess I should add some buffering, anyway:

For best match with hardware and network realities, the value of bufsize should be a relatively small power of 2, for example, 4096.

http://docs.python.org/library/socket.html#socket.socket.recv

But it doesn't seem easy to write efficient and thread-safe buffering. What if I use file.readline()?

# does this work well, is it efficiently buffered?
s.makefile().readline()

Answer 1

The recv() call is handled directly by calling the C library function.

It will block waiting for the socket to have data. In reality it will just let the recv() system call block.

file.readline() is an efficient buffered implementation. It is not threadsafe, because it presumes it's the only one reading the file. (For example by buffering upcoming input.)

If you are using the file object, every time read() is called with a positive argument, the underlying code will recv() only the amount of data requested, unless it's already buffered.

It would be buffered if:

• you had called readline(), which reads a full buffer

• the end of the line was before the end of the buffer

Thus leaving data in the buffer. Otherwise the buffer is generally not overfilled.

The goal of the question is not clear. if you need to see if data is available before reading, you can select() or set the socket to nonblocking mode with s.setblocking(False). Then, reads will return empty, rather than blocking, if there is no waiting data.

Are you reading one file or socket with multiple threads? I would put a single worker on reading the socket and feeding received items into a queue for handling by other threads.

Suggest consulting Python Socket Module source and C Source that makes the system calls.

Answer 2

If you are concerned with performance and control the socket completely (you are not passing it into a library for example) then try implementing your own buffering in Python -- Python string.find and string.split and such can be amazingly fast.

def linesplit(socket):
    buffer = socket.recv(4096)
    buffering = True
    while buffering:
        if "\n" in buffer:
            (line, buffer) = buffer.split("\n", 1)
            yield line + "\n"
        else:
            more = socket.recv(4096)
            if not more:
                buffering = False
            else:
                buffer += more
    if buffer:
        yield buffer

If you expect the payload to consist of lines that are not too huge, that should run pretty fast, and avoid jumping through too many layers of function calls unnecessarily. I'd be interesting in knowing how this compares to file.readline() or using socket.recv(1).

Answer 3

def buffered_readlines(pull_next_chunk, buf_size=4096):
  """
  pull_next_chunk is callable that should accept one positional argument max_len,
  i.e. socket.recv or file().read and returns string of up to max_len long or
  empty one when nothing left to read.

  >>> for line in buffered_readlines(socket.recv, 16384):
  ...   print line
    ...
  >>> # the following code won't read whole file into memory
  ... # before splitting it into lines like .readlines method
  ... # of file does. Also it won't block until FIFO-file is closed
  ...
  >>> for line in buffered_readlines(open('huge_file').read):
  ...   # process it on per-line basis
        ...
  >>>
  """
  chunks = []
  while True:
    chunk = pull_next_chunk(buf_size)
    if not chunk:
      if chunks:
        yield ''.join(chunks)
      break
    if not '\n' in chunk:
      chunks.append(chunk)
      continue
    chunk = chunk.split('\n')
    if chunks:
      yield ''.join(chunks + [chunk[0]])
    else:
      yield chunk[0]
    for line in chunk[1:-1]:
      yield line
    if chunk[-1]:
      chunks = [chunk[-1]]
    else:
      chunks = []