def test_nuid_sequence_rollover(self): nuid = NUID() seq_a = nuid._seq inc_a = nuid._inc nuid_a = nuid.next() seq_b = nuid._seq inc_b = nuid._inc self.assertTrue(seq_a < seq_b) self.assertEqual(seq_b, seq_a + inc_a) nuid_b = nuid.next() self.assertEqual(nuid_a[:PREFIX_LENGTH], nuid_b[:PREFIX_LENGTH]) # Force the sequence to rollover, prefix should now change nuid._seq = seq_c = MAX_SEQ + 1 nuid_c = nuid.next() self.assertNotEqual(nuid_a[:PREFIX_LENGTH], nuid_c[:PREFIX_LENGTH])
def test_nuid_are_very_unique(self): nuid = NUID() entries = [nuid.next().decode() for i in range(1000000)] counted_entries = Counter(entries) repeated = [ entry for entry, count in counted_entries.items() if count > 1 ] self.assertEqual(len(repeated), 0)
def test_nuid_length(self): nuid = NUID() self.assertEqual(len(nuid.next()), TOTAL_LENGTH)
class Client(object): """ Tornado based client for NATS. """ DISCONNECTED = 0 CONNECTED = 1 CLOSED = 2 RECONNECTING = 3 CONNECTING = 4 def __repr__(self): return "<nats client v{}>".format(__version__) def __init__(self): self.options = {} # INFO that we get upon connect from the server. self._server_info = {} self._max_payload_size = DEFAULT_MAX_PAYLOAD_SIZE # Client connection state and clustering. self.io = None self._socket = None self._status = Client.DISCONNECTED self._server_pool = [] self._current_server = None self._pending = [] self._pending_size = 0 self._loop = None self.stats = { 'in_msgs': 0, 'out_msgs': 0, 'in_bytes': 0, 'out_bytes': 0, 'reconnects': 0, 'errors_received': 0 } # Storage and monotonically increasing index for subscription callbacks. self._subs = {} self._ssid = 0 # Parser with state for processing the wire protocol. self._ps = Parser(self) self._err = None self._flush_queue = None # New style request/response self._resp_sub = None self._resp_map = None self._resp_sub_prefix = None self._nuid = NUID() # Ping interval to disconnect from unhealthy servers. self._ping_timer = None self._pings_outstanding = 0 self._pongs_received = 0 self._pongs = [] self._error_cb = None self._close_cb = None self._disconnected_cb = None self._reconnected_cb = None @tornado.gen.coroutine def connect(self, servers=[], verbose=False, pedantic=False, name=None, ping_interval=DEFAULT_PING_INTERVAL, max_outstanding_pings=MAX_OUTSTANDING_PINGS, dont_randomize=False, allow_reconnect=True, close_cb=None, error_cb=None, disconnected_cb=None, reconnected_cb=None, io_loop=None, max_read_buffer_size=DEFAULT_READ_BUFFER_SIZE, max_write_buffer_size=DEFAULT_WRITE_BUFFER_SIZE, read_chunk_size=DEFAULT_READ_CHUNK_SIZE, tcp_nodelay=False, connect_timeout=DEFAULT_CONNECT_TIMEOUT, max_reconnect_attempts=MAX_RECONNECT_ATTEMPTS, reconnect_time_wait=RECONNECT_TIME_WAIT, tls=None): """ Establishes a connection to a NATS server. Examples: # Configure pool of NATS servers. nc = nats.io.client.Client() yield nc.connect({ 'servers': ['nats://192.168.1.10:4222', 'nats://192.168.2.10:4222'] }) # User and pass are to be passed on the uri to authenticate. yield nc.connect({ 'servers': ['nats://*****:*****@192.168.1.10:4222'] }) """ self.options["servers"] = servers self.options["verbose"] = verbose self.options["pedantic"] = pedantic self.options["name"] = name self.options["max_outstanding_pings"] = max_outstanding_pings self.options["max_reconnect_attempts"] = max_reconnect_attempts self.options["reconnect_time_wait"] = reconnect_time_wait self.options["dont_randomize"] = dont_randomize self.options["allow_reconnect"] = allow_reconnect self.options["tcp_nodelay"] = tcp_nodelay # In seconds self.options["connect_timeout"] = connect_timeout self.options["ping_interval"] = ping_interval # TLS customizations if tls is not None: self.options["tls"] = tls self._close_cb = close_cb self._error_cb = error_cb self._disconnected_cb = disconnected_cb self._reconnected_cb = reconnected_cb self._loop = io_loop if io_loop else tornado.ioloop.IOLoop.instance() self._max_read_buffer_size = max_read_buffer_size self._max_write_buffer_size = max_write_buffer_size self._read_chunk_size = read_chunk_size if len(self.options["servers"]) < 1: srv = Srv(urlparse("nats://127.0.0.1:4222")) self._server_pool.append(srv) else: for srv in self.options["servers"]: self._server_pool.append(Srv(urlparse(srv))) while True: try: s = self._next_server() if s is None: raise ErrNoServers # Check when was the last attempt and back off before reconnecting if s.last_attempt is not None: now = time.time() if (now - s.last_attempt ) < self.options["reconnect_time_wait"]: yield tornado.gen.sleep( self.options["reconnect_time_wait"]) # Mark that we have attempted to connect s.reconnects += 1 s.last_attempt = time.time() yield self._server_connect(s) self._current_server = s s.did_connect = True # Established TCP connection at least and about # to send connect command, which might not succeed # in case TLS required and handshake failed. self._status = Client.CONNECTING yield self._process_connect_init() self.io.set_close_callback(self._unbind) break except (socket.error, tornado.iostream.StreamClosedError) as e: self._status = Client.DISCONNECTED self._err = e if self._error_cb is not None: self._error_cb(ErrServerConnect(e)) if not self.options["allow_reconnect"]: raise ErrNoServers # Flush pending data before continuing in connected status. # FIXME: Could use future here and wait for an error result # to bail earlier in case there are errors in the connection. yield self._flush_pending() # First time connecting to NATS so if there were no errors, # we can consider to be connected at this point. self._status = Client.CONNECTED # Prepare the ping pong interval. self._ping_timer = tornado.ioloop.PeriodicCallback( self._send_ping, self.options["ping_interval"] * 1000) self._ping_timer.start() @tornado.gen.coroutine def _server_connect(self, s): """ Sets up a TCP connection to the server. """ self._socket = socket.socket(socket.AF_INET, socket.SOCK_STREAM) self._socket.setblocking(0) self._socket.settimeout(1.0) if self.options["tcp_nodelay"]: self._socket.setsockopt(socket.IPPROTO_TCP, socket.TCP_NODELAY, 1) self.io = tornado.iostream.IOStream( self._socket, max_buffer_size=self._max_read_buffer_size, max_write_buffer_size=self._max_write_buffer_size, read_chunk_size=self._read_chunk_size) # Connect to server with a deadline future = self.io.connect((s.uri.hostname, s.uri.port)) yield tornado.gen.with_timeout( timedelta(seconds=self.options["connect_timeout"]), future) @tornado.gen.coroutine def _send_ping(self, future=None): if self._pings_outstanding > self.options["max_outstanding_pings"]: yield self._unbind() else: yield self.send_command(PING_PROTO) yield self._flush_pending() if future is None: future = tornado.concurrent.Future() self._pings_outstanding += 1 self._pongs.append(future) def connect_command(self): ''' Generates a JSON string with the params to be used when sending CONNECT to the server. ->> CONNECT {"verbose": false, "pedantic": false, "lang": "python2" } ''' options = { "verbose": self.options["verbose"], "pedantic": self.options["pedantic"], "lang": __lang__, "version": __version__, "protocol": PROTOCOL } if "auth_required" in self._server_info: if self._server_info["auth_required"] == True: # In case there is no password, then consider handle # sending a token instead. if self._current_server.uri.password is None: options["auth_token"] = self._current_server.uri.username else: options["user"] = self._current_server.uri.username options["pass"] = self._current_server.uri.password if self.options["name"] is not None: options["name"] = self.options["name"] args = json.dumps(options, sort_keys=True) return CONNECT_PROTO.format(CONNECT_OP, args, _CRLF_) @tornado.gen.coroutine def send_command(self, cmd, priority=False): """ Flushes a command to the server as a bytes payload. """ if priority: self._pending.insert(0, cmd) else: self._pending.append(cmd) self._pending_size += len(cmd) if self._pending_size > DEFAULT_PENDING_SIZE: yield self._flush_pending() @tornado.gen.coroutine def _publish(self, subject, reply, payload, payload_size): payload_size_bytes = ("%d" % payload_size).encode() pub_cmd = b''.join([ PUB_OP, _SPC_, subject.encode(), _SPC_, reply, _SPC_, payload_size_bytes, _CRLF_, payload, _CRLF_ ]) self.stats['out_msgs'] += 1 self.stats['out_bytes'] += payload_size yield self.send_command(pub_cmd) @tornado.gen.coroutine def _flush_pending(self, check_connected=True): if not self.is_connected and check_connected: return yield self._flush_queue.put(None) @tornado.gen.coroutine def publish(self, subject, payload): """ Sends a PUB command to the server on the specified subject. ->> PUB hello 5 ->> MSG_PAYLOAD: world <<- MSG hello 2 5 """ yield self.publish_request(subject, _EMPTY_, payload) @tornado.gen.coroutine def publish_request(self, subject, reply, payload): """ Publishes a message tagging it with a reply subscription which can be used by those receiving the message to respond: ->> PUB hello _INBOX.2007314fe0fcb2cdc2a2914c1 5 ->> MSG_PAYLOAD: world <<- MSG hello 2 _INBOX.2007314fe0fcb2cdc2a2914c1 5 """ payload_size = len(payload) if payload_size > self._max_payload_size: raise ErrMaxPayload if self.is_closed: raise ErrConnectionClosed yield self._publish(subject, reply, payload, payload_size) if self._flush_queue.empty(): yield self._flush_pending() @tornado.gen.coroutine def flush(self, timeout=60): """ Flush will perform a round trip to the server and return True when it receives the internal reply or raise a Timeout error. """ if self.is_closed: raise ErrConnectionClosed yield self._flush_timeout(timeout) @tornado.gen.coroutine def _flush_timeout(self, timeout): """ Takes a timeout and sets up a future which will return True once the server responds back otherwise raise a TimeoutError. """ future = tornado.concurrent.Future() yield self._send_ping(future) try: result = yield tornado.gen.with_timeout(timedelta(seconds=timeout), future) except tornado.gen.TimeoutError: # Set the future to False so it can be ignored in _process_pong. future.set_result(False) raise raise tornado.gen.Return(result) @tornado.gen.coroutine def request(self, subject, payload, timeout=0.5, expected=1, cb=None): """ Implements the request/response pattern via pub/sub using an unique reply subject and an async subscription. If cb is None, then it will wait and return a single message using the new request/response style that is less chatty over the network. ->> SUB _INBOX.BF6zPVxvScfXGd4VyUMJyo.* 1 ->> PUB hello _INBOX.BF6zPVxvScfXGd4VyUMJyo.BF6zPVxvScfXCh4VyUMJyo 5 ->> MSG_PAYLOAD: hello <<- MSG hello 2 _INBOX.BF6zPVxvScfXGd4VyUMJyo.BF6zPVxvScfXCh4VyUMJyo 5 ->> PUB _INBOX.BF6zPVxvScfXGd4VyUMJyo.BF6zPVxvScfXCh4VyUMJyo 5 ->> MSG_PAYLOAD: world <<- MSG _INBOX.BF6zPVxvScfXGd4VyUMJyo.BF6zPVxvScfXCh4VyUMJyo 1 5 If a cb is passed, then it will use auto unsubscribe functionality and expect a limited number of messages which will be handled asynchronously in the callback. ->> SUB _INBOX.gnKUg9bmAHANjxIsDiQsWO 90 ->> UNSUB 90 1 ->> PUB hello _INBOX.gnKUg9bmAHANjxIsDiQsWO 5 ->> MSG_PAYLOAD: world <<- MSG hello 2 _INBOX.gnKUg9bmAHANjxIsDiQsWO 5 """ # If callback given then continue to use old style. if cb is not None: next_inbox = INBOX_PREFIX[:] next_inbox.extend(self._nuid.next()) inbox = str(next_inbox) sid = yield self.subscribe( inbox, queue=_EMPTY_, cb=cb, max_msgs=expected, ) yield self.auto_unsubscribe(sid, expected) yield self.publish_request(subject, inbox, payload) raise tornado.gen.Return(sid) if self._resp_sub_prefix is None: self._resp_map = {} # Create a prefix and single wildcard subscription once. self._resp_sub_prefix = str(INBOX_PREFIX[:]) self._resp_sub_prefix += self._nuid.next() self._resp_sub_prefix += b'.' resp_mux_subject = str(self._resp_sub_prefix[:]) resp_mux_subject += b'*' sub = Subscription(subject=str(resp_mux_subject)) # FIXME: Allow setting pending limits for responses mux subscription. sub.pending_msgs_limit = DEFAULT_SUB_PENDING_MSGS_LIMIT sub.pending_bytes_limit = DEFAULT_SUB_PENDING_BYTES_LIMIT sub.pending_queue = tornado.queues.Queue( maxsize=sub.pending_msgs_limit) # Single task for handling the requests @tornado.gen.coroutine def wait_for_msgs(): while True: sub = wait_for_msgs.sub if sub.closed: break msg = yield sub.pending_queue.get() if msg is None: break token = msg.subject[INBOX_PREFIX_LEN:] try: fut = self._resp_map[token] fut.set_result(msg) del self._resp_map[token] except KeyError: # Future already handled so drop any extra # responses which may have made it. continue wait_for_msgs.sub = sub self._loop.spawn_callback(wait_for_msgs) # Store the subscription in the subscriptions map, # then send the protocol commands to the server. self._ssid += 1 sid = self._ssid sub.sid = sid self._subs[sid] = sub # Send SUB command... sub_cmd = b''.join([ SUB_OP, _SPC_, sub.subject.encode(), _SPC_, ("%d" % sid).encode(), _CRLF_ ]) yield self.send_command(sub_cmd) yield self._flush_pending() # Use a new NUID for the token inbox and then use the future. token = self._nuid.next() inbox = self._resp_sub_prefix[:] inbox.extend(token) future = tornado.concurrent.Future() self._resp_map[token.decode()] = future yield self.publish_request(subject, str(inbox), payload) msg = yield tornado.gen.with_timeout(timedelta(seconds=timeout), future) raise tornado.gen.Return(msg) @tornado.gen.coroutine def timed_request(self, subject, payload, timeout=0.5): """ Implements the request/response pattern via pub/sub using an ephemeral subscription which will be published with a limited interest of 1 reply returning the response or raising a Timeout error. ->> SUB _INBOX.E9jM2HTirMXDMXPROSQmSd 90 ->> UNSUB 90 1 ->> PUB hello _INBOX.E9jM2HTirMXDMXPROSQmSd 5 ->> MSG_PAYLOAD: world <<- MSG hello 2 _INBOX.E9jM2HTirMXDMXPROSQmSd 5 """ next_inbox = INBOX_PREFIX[:] next_inbox.extend(self._nuid.next()) inbox = str(next_inbox) future = tornado.concurrent.Future() sid = yield self.subscribe(subject=inbox, queue=_EMPTY_, cb=None, future=future, max_msgs=1) yield self.auto_unsubscribe(sid, 1) yield self.publish_request(subject, inbox, payload) msg = yield tornado.gen.with_timeout(timedelta(seconds=timeout), future) raise tornado.gen.Return(msg) @tornado.gen.coroutine def subscribe( self, subject="", queue="", cb=None, future=None, max_msgs=0, is_async=False, pending_msgs_limit=DEFAULT_SUB_PENDING_MSGS_LIMIT, pending_bytes_limit=DEFAULT_SUB_PENDING_BYTES_LIMIT, ): """ Sends a SUB command to the server. Takes a queue parameter which can be used in case of distributed queues or left empty if it is not the case, and a callback that will be dispatched message for processing them. """ if self.is_closed: raise ErrConnectionClosed self._ssid += 1 sid = self._ssid sub = Subscription( subject=subject, queue=queue, cb=cb, future=future, max_msgs=max_msgs, is_async=is_async, sid=sid, ) self._subs[sid] = sub if cb is not None: sub.pending_msgs_limit = pending_msgs_limit sub.pending_bytes_limit = pending_bytes_limit sub.pending_queue = tornado.queues.Queue( maxsize=pending_msgs_limit) @tornado.gen.coroutine def wait_for_msgs(): while True: sub = wait_for_msgs.sub err_cb = wait_for_msgs.err_cb try: sub = wait_for_msgs.sub if sub.closed: break msg = yield sub.pending_queue.get() if msg is None: break sub.pending_size -= len(msg.data) if sub.max_msgs > 0 and sub.received >= sub.max_msgs: # If we have hit the max for delivered msgs, remove sub. self._remove_subscription(sub) # Invoke depending of type of handler. if sub.is_async: # NOTE: Deprecate this usage in a next release, # the handler implementation ought to decide # the concurrency level at which the messages # should be processed. self._loop.spawn_callback(sub.cb, msg) else: # Call it and take the possible future in the loop. yield sub.cb(msg) except Exception as e: # All errors from calling an async subscriber # handler are async errors. if err_cb is not None: yield err_cb(e) finally: if sub.max_msgs > 0 and sub.received >= sub.max_msgs: # If we have hit the max for delivered msgs, remove sub. self._remove_subscription(sub) break # Bind the subscription and error cb if present wait_for_msgs.sub = sub wait_for_msgs.err_cb = self._error_cb self._loop.spawn_callback(wait_for_msgs) elif future is not None: # Used to handle the single response from a request # based on auto unsubscribe. sub.future = future # Send SUB command... sub_cmd = b''.join([ SUB_OP, _SPC_, sub.subject.encode(), _SPC_, sub.queue.encode(), _SPC_, ("%d" % sid).encode(), _CRLF_ ]) yield self.send_command(sub_cmd) yield self._flush_pending() raise tornado.gen.Return(sid) @tornado.gen.coroutine def subscribe_async(self, subject, **kwargs): """ Schedules callback from subscription to be processed asynchronously in the next iteration of the loop. """ kwargs["is_async"] = True sid = yield self.subscribe(subject, **kwargs) raise tornado.gen.Return(sid) @tornado.gen.coroutine def unsubscribe(self, ssid, max_msgs=0): """ Takes a subscription sequence id and removes the subscription from the client, optionally after receiving more than max_msgs, and unsubscribes immediatedly. """ if self.is_closed: raise ErrConnectionClosed sub = None try: sub = self._subs[ssid] except KeyError: # Already unsubscribed. return # In case subscription has already received enough messages # then announce to the server that we are unsubscribing and # remove the callback locally too. if max_msgs == 0 or sub.received >= max_msgs: self._subs.pop(ssid, None) self._remove_subscription(sub) # We will send these for all subs when we reconnect anyway, # so that we can suppress here. if not self.is_reconnecting: yield self.auto_unsubscribe(ssid, max_msgs) def _remove_subscription(self, sub): # Mark as invalid sub.closed = True # Remove the pending queue if sub.pending_queue is not None: try: # Send empty msg to signal cancellation # and stop the msg processing loop. sub.pending_queue.put_nowait(None) except tornado.queues.QueueFull: # Skip error return @tornado.gen.coroutine def auto_unsubscribe(self, sid, limit=1): """ Sends an UNSUB command to the server. Unsubscribe is one of the basic building blocks in order to be able to define request/response semantics via pub/sub by announcing the server limited interest a priori. """ b_limit = b'' if limit > 0: b_limit = ("%d" % limit).encode() b_sid = ("%d" % sid).encode() unsub_cmd = b''.join([UNSUB_OP, _SPC_, b_sid, _SPC_, b_limit, _CRLF_]) yield self.send_command(unsub_cmd) yield self._flush_pending() @tornado.gen.coroutine def _process_ping(self): """ The server will be periodically sending a PING, and if the the client does not reply a PONG back a number of times, it will close the connection sending an `-ERR 'Stale Connection'` error. """ yield self.send_command(PONG_PROTO) if self._flush_queue.empty(): yield self._flush_pending() @tornado.gen.coroutine def _process_pong(self): """ The client will send a PING soon after CONNECT and then periodically to the server as a failure detector to close connections to unhealthy servers. For each PING the client sends, we will add a respective PONG future. Here we want to find the oldest PONG future that is still running. If the flush PING-PONG already timed out, then just drop those old items. """ while len(self._pongs) > 0: future = self._pongs.pop(0) self._pongs_received += 1 self._pings_outstanding -= 1 # Only exit loop if future still running (hasn't exceeded flush timeout). if future.running(): future.set_result(True) break @tornado.gen.coroutine def _process_msg(self, sid, subject, reply, data): """ Dispatches the received message to the stored subscription. It first tries to detect whether the message should be dispatched to a passed callback. In case there was not a callback, then it tries to set the message into a future. """ payload_size = len(data) self.stats['in_msgs'] += 1 self.stats['in_bytes'] += payload_size msg = Msg(subject=subject.decode(), reply=reply.decode(), data=data) # Don't process the message if the subscription has been removed sub = self._subs.get(sid) if sub is None: raise tornado.gen.Return() sub.received += 1 if sub.max_msgs > 0 and sub.received >= sub.max_msgs: # Enough messages so can throwaway subscription now. self._subs.pop(sid, None) # Check if it is an old style request. if sub.future is not None: sub.future.set_result(msg) # Discard subscription since done self._remove_subscription(sub) raise tornado.gen.Return() # Let subscription wait_for_msgs coroutine process the messages, # but in case sending to the subscription task would block, # then consider it to be an slow consumer and drop the message. try: sub.pending_size += payload_size if sub.pending_size >= sub.pending_bytes_limit: # Substract again the bytes since throwing away # the message so would not be pending data. sub.pending_size -= payload_size if self._error_cb is not None: yield self._error_cb(ErrSlowConsumer()) raise tornado.gen.Return() yield sub.pending_queue.put_nowait(msg) except tornado.queues.QueueFull: if self._error_cb is not None: yield self._error_cb(ErrSlowConsumer()) @tornado.gen.coroutine def _process_connect_init(self): """ Handles the initial part of the NATS protocol, moving from the (RE)CONNECTING to CONNECTED states when establishing a connection with the server. """ # INFO {...} line = yield self.io.read_until(_CRLF_, max_bytes=None) _, args = line.split(INFO_OP + _SPC_, 1) self._server_info = tornado.escape.json_decode((args)) self._max_payload_size = self._server_info["max_payload"] # Check whether we need to upgrade to TLS first of all if 'tls_required' in self._server_info and self._server_info[ 'tls_required']: # Detach and prepare for upgrading the TLS connection. self._loop.remove_handler(self._socket.fileno()) tls_opts = {} if "tls" in self.options: # Allow customizing the TLS version though default # to one that the server supports at least. tls_opts = self.options["tls"] # Rewrap using a TLS connection, can't do handshake on connect # as the socket is non blocking. self._socket = ssl.wrap_socket(self._socket, do_handshake_on_connect=False, **tls_opts) # Use the TLS stream instead from now self.io = tornado.iostream.SSLIOStream(self._socket, io_loop=self._loop) self.io.set_close_callback(self._unbind) self.io._do_ssl_handshake() # CONNECT {...} cmd = self.connect_command() yield self.io.write(cmd) # Refresh state of the parser upon reconnect. if self.is_reconnecting: self._ps.reset() # Send a PING expecting a PONG to make a roundtrip to the server # and assert that sent messages sent this far have been processed. yield self.io.write(PING_PROTO) # FIXME: Add readline timeout for these. next_op = yield self.io.read_until(_CRLF_, max_bytes=MAX_CONTROL_LINE_SIZE) if self.options["verbose"] and OK_OP in next_op: next_op = yield self.io.read_until(_CRLF_, max_bytes=MAX_CONTROL_LINE_SIZE) if ERR_OP in next_op: err_line = next_op.decode() _, err_msg = err_line.split(_SPC_, 1) # FIXME: Maybe handling could be more special here, # checking for ErrAuthorization for example. # yield from self._process_err(err_msg) raise NatsError("nats: " + err_msg.rstrip('\r\n')) if PONG_PROTO in next_op: self._status = Client.CONNECTED # Parser reads directly from the same IO as the client. self._loop.spawn_callback(self._read_loop) # Queue and flusher for coalescing writes to the server. self._flush_queue = tornado.queues.Queue(maxsize=1024) self._loop.spawn_callback(self._flusher_loop) def _process_info(self, info_line): """ Process INFO lines sent by the server to reconfigure client with latest updates from cluster to enable server discovery. """ info = tornado.escape.json_decode(info_line.decode()) if 'connect_urls' in info: if info['connect_urls']: connect_urls = [] for connect_url in info['connect_urls']: uri = urlparse("nats://%s" % connect_url) srv = Srv(uri) srv.discovered = True # Filter for any similar server in the server pool already. should_add = True for s in self._server_pool: if uri.netloc == s.uri.netloc: should_add = False if should_add: connect_urls.append(srv) if self.options["dont_randomize"] is not True: shuffle(connect_urls) for srv in connect_urls: self._server_pool.append(srv) def _next_server(self): """ Chooses next available server to connect. """ if self.options["dont_randomize"]: server = self._server_pool.pop(0) self._server_pool.append(server) else: shuffle(self._server_pool) s = None for server in self._server_pool: if self.options["max_reconnect_attempts"] > 0 and ( server.reconnects > self.options["max_reconnect_attempts"]): continue else: s = server return s @property def is_closed(self): return self._status == Client.CLOSED @property def is_reconnecting(self): return self._status == Client.RECONNECTING @property def is_connected(self): return self._status == Client.CONNECTED @property def is_connecting(self): return self._status == Client.CONNECTING @tornado.gen.coroutine def _unbind(self): """ Unbind handles the disconnection from the server then attempts to reconnect if `allow_reconnect' is enabled. """ if self.is_connecting or self.is_closed or self.is_reconnecting: return if self._disconnected_cb is not None: self._disconnected_cb() if not self.options["allow_reconnect"]: self._process_disconnect() yield self._end_flusher_loop() return if self.is_connected: self._status = Client.RECONNECTING if self._ping_timer is not None and self._ping_timer.is_running(): self._ping_timer.stop() if self.io and not self.io.closed(): self.io.close() yield self._end_flusher_loop() while True: try: yield self._schedule_primary_and_connect() except ErrNoServers: self._process_disconnect() break try: yield self._process_connect_init() break except Exception as e: self._err = e yield self._close(Client.DISCONNECTED) # Replay all the subscriptions in case there were some. for ssid, sub in self._subs.items(): sub_cmd = SUB_PROTO.format(SUB_OP, sub.subject, sub.queue, ssid, _CRLF_) yield self.io.write(sub_cmd) # Restart the ping pong interval callback. self._ping_timer = tornado.ioloop.PeriodicCallback( self._send_ping, self.options["ping_interval"] * 1000) self._ping_timer.start() self._err = None self._pings_outstanding = 0 self._pongs = [] # Flush any pending bytes from reconnect if len(self._pending) > 0: yield self._flush_pending() # Reconnected at this point self._status = Client.CONNECTED # Roundtrip to the server to ensure connection # is healthy at this point. yield self.flush() @tornado.gen.coroutine def _schedule_primary_and_connect(self): """ Attempts to connect to an available server. """ while True: s = self._next_server() if s is None: raise ErrNoServers # For the reconnection logic, we need to consider # sleeping for a bit before trying to reconnect # too soon to a server which has failed previously. # Check when was the last attempt and back off before reconnecting if s.last_attempt is not None: now = time.time() if (now - s.last_attempt) < self.options["reconnect_time_wait"]: yield tornado.gen.sleep( self.options["reconnect_time_wait"]) s.reconnects += 1 s.last_attempt = time.time() self.stats['reconnects'] += 1 try: yield self._server_connect(s) self._current_server = s # Reset number of reconnects upon successful connection. s.reconnects = 0 self.io.set_close_callback(self._unbind) if self.is_reconnecting and self._reconnected_cb is not None: self._reconnected_cb() return except (socket.error, tornado.iostream.StreamClosedError) as e: if self._error_cb is not None: self._error_cb(ErrServerConnect(e)) # Continue trying to connect until there is an available server # or bail in case there are no more available servers. self._status = Client.RECONNECTING continue @tornado.gen.coroutine def _process_disconnect(self): """ Does cleanup of the client state and tears down the connection. """ self._status = Client.DISCONNECTED yield self.close() @tornado.gen.coroutine def close(self): """ Wraps up connection to the NATS cluster and stops reconnecting. """ yield self._close(Client.CLOSED) @tornado.gen.coroutine def _close(self, status, do_callbacks=True): """ Takes the status on which it should leave the connection and an optional boolean parameter to dispatch the disconnected and close callbacks if there are any. """ if self.is_closed: # If connection already closed, then just set status explicitly. self._status = status return self._status = Client.CLOSED if self._ping_timer is not None and self._ping_timer.is_running(): self._ping_timer.stop() if not self.io.closed(): self.io.close() # Cleanup subscriptions since not reconnecting so no need # to replay the subscriptions anymore. for ssid, sub in self._subs.items(): self._subs.pop(ssid, None) self._remove_subscription(sub) if do_callbacks: if self._disconnected_cb is not None: self._disconnected_cb() if self._close_cb is not None: self._close_cb() @tornado.gen.coroutine def _process_err(self, err=None): """ Stores the last received error from the server and dispatches the error callback. """ self.stats['errors_received'] += 1 if err == "'Authorization Violation'": self._err = ErrAuthorization elif err == "'Slow Consumer'": self._err = ErrSlowConsumer elif err == "'Stale Connection'": self._err = ErrStaleConnection else: self._err = Exception(err) if self._error_cb is not None: self._error_cb(err) def last_error(self): return self._err @property def connected_url(self): if self.is_connected: return self._current_server.uri else: return None @property def servers(self): servers = [] for srv in self._server_pool: servers.append(srv) return servers @property def discovered_servers(self): servers = [] for srv in self._server_pool: if srv.discovered: servers.append(srv) return servers @tornado.gen.coroutine def _read_loop(self, data=''): """ Read loop for gathering bytes from the server in a buffer of maximum MAX_CONTROL_LINE_SIZE, then received bytes are streamed to the parsing callback for processing. """ while True: if not self.is_connected or self.is_connecting or self.io.closed(): break try: yield self.io.read_bytes(DEFAULT_READ_CHUNK_SIZE, streaming_callback=self._ps.parse, partial=True) except tornado.iostream.StreamClosedError as e: self._err = e if self._error_cb is not None and not self.is_reconnecting and not self.is_closed: self._error_cb(e) break @tornado.gen.coroutine def _flusher_loop(self): """ Coroutine which continuously tries to consume pending commands and then flushes them to the socket. """ while True: pending = [] pending_size = 0 try: # Block and wait for the flusher to be kicked yield self._flush_queue.get() # Check whether we should bail first if not self.is_connected or self.is_connecting or self.io.closed( ): break # Flush only when we actually have something in buffer... if self._pending_size > 0: cmds = b''.join(self._pending) # Reset pending queue and store tmp in case write fails self._pending, pending = [], self._pending self._pending_size, pending_size = 0, self._pending_size yield self.io.write(cmds) except tornado.iostream.StreamBufferFullError: # Acumulate as pending data size and flush when possible. self._pending = pending + self._pending self._pending_size += pending_size except tornado.iostream.StreamClosedError as e: self._pending = pending + self._pending self._pending_size += pending_size self._err = e if self._error_cb is not None and not self.is_reconnecting: self._error_cb(e) yield self._unbind() @tornado.gen.coroutine def _end_flusher_loop(self): """ Let flusher_loop coroutine quit - useful when disconnecting. """ if not self.is_connected or self.is_connecting or self.io.closed(): if self._flush_queue is not None and self._flush_queue.empty(): self._flush_pending(check_connected=False) yield tornado.gen.moment