X-Git-Url: https://jfr.im/git/z_archive/twitter.git/blobdiff_plain/221bc078584e2131e6a38f6d3fd7b6bb2249d192..f560656d5b2e57c6d264d073566848fd71c9000e:/twitter/stream.py?ds=sidebyside diff --git a/twitter/stream.py b/twitter/stream.py index a0df98e..c7d9292 100644 --- a/twitter/stream.py +++ b/twitter/stream.py @@ -1,10 +1,10 @@ try: import urllib.request as urllib_request import urllib.error as urllib_error - import io except ImportError: import urllib2 as urllib_request import urllib2 as urllib_error +import io import json from ssl import SSLError import socket @@ -12,109 +12,152 @@ import sys, select, time from .api import TwitterCall, wrap_response, TwitterHTTPError -python26 = sys.version_info < (2, 7) -def recv_chunk(sock): # -> bytearray: +PY_3_OR_HIGHER = sys.version_info >= (3, 0) - header = sock.recv(8) # Scan for an up to 16MiB chunk size (0xffffff). - crlf = header.find(b'\r\n') # Find the HTTP chunk size. +CRLF = b'\r\n' - if crlf > 0: # If there is a length, then process it +Timeout = {'timeout': True} +Hangup = {'hangup': True} +HeartbeatTimeout = {'heartbeat_timeout': True, 'hangup': True} - size = int(header[:crlf], 16) # Decode the chunk size. Rarely exceeds 8KiB. - chunk = bytearray(size) - start = crlf + 2 # Add in the length of the header's CRLF pair. +class ChunkDecodeError(Exception): + pass - if size <= 3: # E.g. an HTTP chunk with just a keep-alive delimiter or end of stream (0). - chunk[:size] = header[start:start + size] - # There are several edge cases (size == [4-6]) as the chunk size exceeds the length - # of the initial read of 8 bytes. With Twitter, these do not, in practice, occur. The - # shortest JSON message starts with '{"limit":{'. Hence, it exceeds in size the edge cases - # and eliminates the need to address them. - else: # There is more to read in the chunk. - end = len(header) - start - chunk[:end] = header[start:] - if python26: # less efficient for python2.6 compatibility - chunk[end:] = sock.recv(max(0, size - end)) - else: # When possible, use less memory by reading directly into the buffer. - buffer = memoryview(chunk)[end:] # Create a view into the bytearray to hold the rest of the chunk. - sock.recv_into(buffer) - sock.recv(2) # Read the trailing CRLF pair. Throw it away. +class EndOfStream(Exception): + pass - return chunk +range = range if PY_3_OR_HIGHER else xrange - return bytearray() +class SocketShim(io.IOBase): + """ + Adapts a raw socket to fit the IO protocol. + """ + def __init__(self, sock): + self.sock = sock + def readable(self): + return True + def read(self, size): + return self.sock.read(size) + def readinto(self, buf): + return self.sock.recv_into(buf) + +def recv_chunk(reader): # -> bytearray: + for headerlen in range(12): + header = reader.peek(headerlen)[:headerlen] + if header.endswith(CRLF): + break + else: + raise ChunkDecodeError() + + size = int(header, 16) # Decode the chunk size + reader.read(headerlen) # Ditch the header + + if size == 0: + raise EndOfStream() + + chunk = bytearray() + while len(chunk) < size: + remainder = size - len(chunk) + chunk.extend(reader.read(remainder)) + + reader.read(2) # Ditch remaining CRLF + + return chunk + + +class Timer(object): + def __init__(self, timeout): + # If timeout is None, we never expire. + self.timeout = timeout + self.reset() + + def reset(self): + self.time = time.time() + + def expired(self): + """ + If expired, reset the timer and return True. + """ + if self.timeout is None: + return False + elif time.time() - self.time > self.timeout: + self.reset() + return True + return False class TwitterJSONIter(object): - def __init__(self, handle, uri, arg_data, block=True, timeout=None): + def __init__(self, handle, uri, arg_data, block, timeout, heartbeat_timeout): self.handle = handle self.uri = uri self.arg_data = arg_data self.block = block - self.timeout = timeout + self.timeout = float(timeout) if timeout else None + self.heartbeat_timeout = float(heartbeat_timeout) if heartbeat_timeout else None def __iter__(self): - sock = self.handle.fp.raw._sock if sys.version_info >= (3, 0) else self.handle.fp._sock.fp._sock + actually_block = self.block and not self.timeout + sock_timeout = min(self.timeout or 1000000, self.heartbeat_timeout) if actually_block else None + sock = self.handle.fp.raw._sock if PY_3_OR_HIGHER else self.handle.fp._sock.fp._sock sock.setsockopt(socket.SOL_SOCKET, socket.SO_KEEPALIVE, 1) - sock.setblocking(self.block and not self.timeout) + sock.setblocking(actually_block) + reader = io.BufferedReader(SocketShim(sock)) buf = '' - json_decoder = json.JSONDecoder() - timer = time.time() + raw_decode = json.JSONDecoder().raw_decode + timer = Timer(self.timeout) + heartbeat_timer = Timer(self.heartbeat_timeout) while True: + buf = buf.lstrip() # Remove any keep-alive delimiters try: - buf = buf.lstrip() - res, ptr = json_decoder.raw_decode(buf) + res, ptr = raw_decode(buf) buf = buf[ptr:] + except ValueError: + if not self.block and not self.timeout: + yield None + else: yield wrap_response(res, self.handle.headers) + timer.reset() + heartbeat_timer.reset() continue - except ValueError as e: - if self.block: pass - else: yield None + + if heartbeat_timer.expired(): + yield HeartbeatTimeout + break + if timer.expired(): + yield Timeout + try: - buf = buf.lstrip() # Remove any keep-alive delimiters to detect hangups. - if self.timeout and not buf: # This is a non-blocking read. - ready_to_read = select.select([sock], [], [], self.timeout) - if not ready_to_read[0] and time.time() - timer > self.timeout: - yield {'timeout': True} + if sock_timeout: + ready_to_read = select.select([sock], [], [], sock_timeout)[0] + if not ready_to_read: continue - timer = time.time() - buf += recv_chunk(sock).decode('utf-8') - if not buf: - yield {'hangup': True} - break + received = recv_chunk(reader) + buf += received.decode('utf-8') + if received: + heartbeat_timer.reset() + except (ChunkDecodeError, EndOfStream): + yield Hangup + break except SSLError as e: - # Error from a non-blocking read of an empty buffer. - if (not self.block or self.timeout) and (e.errno == 2): pass - else: raise + # Code 2 is error from a non-blocking read of an empty buffer. + if e.errno != 2: + raise -def handle_stream_response(req, uri, arg_data, block, timeout=None): +def handle_stream_response(req, uri, arg_data, block, timeout, heartbeat_timeout): try: handle = urllib_request.urlopen(req,) except urllib_error.HTTPError as e: raise TwitterHTTPError(e, uri, 'json', arg_data) - return iter(TwitterJSONIter(handle, uri, arg_data, block, timeout=timeout)) - -class TwitterStreamCallWithTimeout(TwitterCall): - def _handle_response(self, req, uri, arg_data, _timeout=None): - return handle_stream_response(req, uri, arg_data, block=True, timeout=self.timeout) - -class TwitterStreamCall(TwitterCall): - def _handle_response(self, req, uri, arg_data, _timeout=None): - return handle_stream_response(req, uri, arg_data, block=True) - -class TwitterStreamCallNonBlocking(TwitterCall): - def _handle_response(self, req, uri, arg_data, _timeout=None): - return handle_stream_response(req, uri, arg_data, block=False) + return iter(TwitterJSONIter(handle, uri, arg_data, block, timeout, heartbeat_timeout)) -class TwitterStream(TwitterStreamCall): +class TwitterStream(TwitterCall): """ - The TwitterStream object is an interface to the Twitter Stream API - (stream.twitter.com). This can be used pretty much the same as the - Twitter class except the result of calling a method will be an - iterator that yields objects decoded from the stream. For - example:: + The TwitterStream object is an interface to the Twitter Stream + API. This can be used pretty much the same as the Twitter class + except the result of calling a method will be an iterator that + yields objects decoded from the stream. For example:: twitter_stream = TwitterStream(auth=OAuth(...)) iterator = twitter_stream.statuses.sample() @@ -122,28 +165,38 @@ class TwitterStream(TwitterStreamCall): for tweet in iterator: ...do something with this tweet... - The iterator will yield tweets forever and ever (until the stream - breaks at which point it raises a TwitterHTTPError.) - - The `block` parameter controls if the stream is blocking. Default - is blocking (True). When set to False, the iterator will - occasionally yield None when there is no available message. + The iterator will yield until the TCP connection breaks. When the + connection breaks, the iterator yields `{'hangup': True}`, and + raises `StopIteration` if iterated again. + + Similarly, if the stream does not produce heartbeats for more than + 90 seconds, the iterator yields `{'hangup': True, + 'heartbeat_timeout': True}`, and raises `StopIteration` if + iterated again. + + The `timeout` parameter controls the maximum time between + yields. If it is nonzero, then the iterator will yield either + stream data or `{'timeout': True}` within the timeout period. This + is useful if you want your program to do other stuff in between + waiting for tweets. + + The `block` parameter sets the stream to be fully non-blocking. In + this mode, the iterator always yields immediately. It returns + stream data, or `None`. Note that `timeout` supercedes this + argument, so it should also be set `None` to use this mode. """ - def __init__( - self, domain="stream.twitter.com", secure=True, auth=None, - api_version='1.1', block=True, timeout=None): - uriparts = () - uriparts += (str(api_version),) - - if block: - if timeout: - call_cls = TwitterStreamCallWithTimeout - else: - call_cls = TwitterStreamCall - else: - call_cls = TwitterStreamCallNonBlocking - - TwitterStreamCall.__init__( + def __init__(self, domain="stream.twitter.com", secure=True, auth=None, + api_version='1.1', block=True, timeout=None, + heartbeat_timeout=90.0): + uriparts = (str(api_version),) + + class TwitterStreamCall(TwitterCall): + def _handle_response(self, req, uri, arg_data, _timeout=None): + return handle_stream_response( + req, uri, arg_data, block, + _timeout or timeout, heartbeat_timeout) + + TwitterCall.__init__( self, auth=auth, format="json", domain=domain, - callable_cls=call_cls, + callable_cls=TwitterStreamCall, secure=secure, uriparts=uriparts, timeout=timeout, gzip=False)