#!/usr/bin/env python """Threaded Application Server The AppServer is the main process of WebKit. It handles requests for servlets from webservers. ThreadedAppServer uses a threaded model for handling multiple requests. At one time there were other experimental execution models for AppServer, but none of these were successful and have been removed. The ThreadedAppServer/AppServer distinction is thus largely historical. ThreadedAppServer takes the following command line arguments: start: start the AppServer (default argument) stop: stop the currently running Apperver daemon: run as a daemon ClassName.SettingName=value: change configuration settings When started, the app server records its pid in appserver.pid. """ import threading, Queue, select, socket, errno, traceback from threading import Thread, currentThread from marshal import dumps, loads try: from ctypes import pythonapi, py_object except ImportError: py_object = pythonapi = None try: PyThreadState_SetAsyncExc = pythonapi.PyThreadState_SetAsyncExc except (TypeError, AttributeError): PyThreadState_SetAsyncExc = None from Common import * import AppServer as AppServerModule from PidFile import ProcessRunning from AutoReloadingAppServer import AutoReloadingAppServer as AppServer from ASStreamOut import ASStreamOut, ConnectionAbortedError from HTTPExceptions import HTTPServiceUnavailable from WebUtils.Funcs import requestURI debug = False defaultConfig = { 'Host': 'localhost', # same as '127.0.0.1' 'EnableAdapter': True, # enable WebKit adapter 'AdapterPort': 8086, 'EnableMonitor': False, # disable status monitoring 'SCGIPort': 8084, 'EnableSCGI': False, # disable SCGI adapter 'MonitorPort': 8085, 'EnableHTTP': True, # enable built-in HTTP server 'HTTPPort': 8080, 'StartServerThreads': 10, # initial number of server threads 'MinServerThreads': 5, # minimum number 'MaxServerThreads': 20, # maxium number 'MaxRequestTime': 300, # maximum request execution time in seconds 'RequestQueueSize': 0, # means twice the maximum number of threads 'RequestBufferSize': 8*1024, # 8 kBytes 'ResponseBufferSize': 8*1024, # 8 kBytes 'AddressFiles': '%s.address', # %s stands for the protocol name # @@ the following setting is not yet implemented # 'SocketType': 'inet', # inet, inet6, unix } # Need to know this value for communications # (note that this limits the size of the dictionary we receive # from the AppServer to 2,147,483,647 bytes): intLength = len(dumps(int(1))) # Initialize global variables server = None exitStatus = 0 class NotEnoughDataError(Exception): pass class ProtocolError(Exception): pass class ThreadAbortedError(HTTPServiceUnavailable): pass class RequestAbortedError(ThreadAbortedError): pass class RequestTooLongError(RequestAbortedError): pass class ServerShutDownError(ThreadAbortedError): pass class WorkerThread(Thread): """Base class for Webware worker threads that can be aborted. (Idea taken from: http://sebulba.wikispaces.com/recipe+thread2) """ _canAbort = PyThreadState_SetAsyncExc is not None def threadID(self): """Return the thread's internal id.""" try: return self._threadID except AttributeError: for threadID, t in threading._active.items(): if t is self: self._threadID = threadID return threadID def abort(self, exception=ThreadAbortedError): """Abort the current thread by raising an exception in its context. A return value of one means the thread was successfully aborted, a value of zero means the thread could not be found, any other value indicates that an error has occurred. """ if not self._canAbort: if debug: print "Error: Aborting threads is not possible" return -1 if debug: print "Aborting worker thread..." try: processing = self.isAlive() and self._processing except AttributeError: processing = False if not processing: if debug: print "Error: Thread is not working." threadID = self.threadID() if threadID is None: if debug: print "Error: Worker thread id not found" return 0 if debug: print "Worker thread id is", threadID try: ret = PyThreadState_SetAsyncExc(threadID, py_object(exception)) # If it returns a number greater than one, we're in trouble, # and should call it again with exc=NULL to revert the effect if ret > 1: PyThreadState_SetAsyncExc(threadID, 0) except Exception: ret = -1 if debug: if ret == 0: print "Error: Could not find thread", threadID elif ret != 1: print "Error: Could not abort thread", threadID return ret class ThreadedAppServer(AppServer): """Threaded Application Server. `ThreadedAppServer` accepts incoming socket requests, spawns a new thread or reuses an existing one, then dispatches the request to the appropriate handler (e.g., an Adapter handler, HTTP handler, etc., one for each protocol). The transaction is connected directly to the socket, so that the response is sent directly (if streaming is used, like if you call `response.flush()`). Thus the ThreadedAppServer packages the socket/response, rather than value being returned up the call chain. """ ## Init ## def __init__(self, path=None): """Setup the AppServer. Create an initial thread pool (threads created with `spawnThread`), and the request queue, record the PID in a file, and add any enabled handlers (Adapter, HTTP, Monitor). """ self._threadPool = [] self._threadCount = 0 self._threadUseCounter = [] self._addr = {} self._requestID = 0 self._socketHandlers = {} self._handlerCache = {} self._threadHandler = {} self._sockets = {} self._defaultConfig = None AppServer.__init__(self, path) try: threadCount = self.setting('StartServerThreads') self._maxServerThreads = self.setting('MaxServerThreads') self._minServerThreads = self.setting('MinServerThreads') self._requestQueueSize = self.setting('RequestQueueSize') if not self._requestQueueSize: # if not set, make queue size twice the max number of threads self._requestQueueSize = 2 * self._maxServerThreads elif self._requestQueueSize < self._maxServerThreads: # otherwise do not make it smaller than the max number of threads self._requestQueueSize = self._maxServerThreads self._requestBufferSize = self.setting('RequestBufferSize') self._responseBufferSize = self.setting('ResponseBufferSize') self._requestQueue = Queue.Queue(self._requestQueueSize) maxRequestTime = self.setting('MaxRequestTime') or None if maxRequestTime and not self._canAbortRequest: print "Warning: MaxRequestTime setting ineffective" \ " (cannot abort requests)" maxRequestTime = None self._maxRequestTime = maxRequestTime self._checkRequestTime = None out = sys.stdout out.write('Creating %d threads' % threadCount) for i in range(threadCount): self.spawnThread() if not debug: out.write(".") out.flush() out.write("\n") if self.setting('EnableAdapter'): self.addSocketHandler(AdapterHandler) if self.setting('EnableMonitor'): self.addSocketHandler(MonitorHandler) if self.setting('EnableSCGI'): self.addSocketHandler(SCGIHandler) if self.setting('EnableHTTP'): from HTTPServer import HTTPAppServerHandler self.addSocketHandler(HTTPAppServerHandler) self.readyForRequests() if maxRequestTime: self._checkRequestTime = time.time() + maxRequestTime except: AppServer.initiateShutdown(self) raise def addSocketHandler(self, handlerClass, serverAddress=None): """Add socket handler. Adds a socket handler for `serverAddress` -- `serverAddress` is a tuple ``(host, port)``, where ``host`` is the interface to connect to (for instance, the IP address on a machine with multiple IP numbers), and ``port`` is the port (e.g. HTTP is on 80 by default, and Webware adapters use 8086 by default). The `handlerClass` is a subclass of `Handler`, and is used to handle the actual request -- usually returning control back to ThreadedAppServer in some fashion. See `Handler` for more. """ if serverAddress is None: serverAddress = self.address(handlerClass.settingPrefix) sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM) sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1) try: sock.bind(serverAddress) sock.listen(1024) except Exception: print "Error: Can not listen for %s on %s" % ( handlerClass.settingPrefix, str(serverAddress)) sys.stdout.flush() raise serverAddress = sock.getsockname() # resolve/normalize self._socketHandlers[serverAddress] = handlerClass self._handlerCache[serverAddress] = [] self._sockets[serverAddress] = sock adrStr = ':'.join(map(str, serverAddress)) print "Listening for %s on %s" % (handlerClass.settingPrefix, adrStr) # write text file with server address adrFile = self.addressFileName(handlerClass) if os.path.exists(adrFile): print "Warning: %s already exists" % adrFile try: os.unlink(adrFile) except OSError: # we cannot remove the file if open(adrFile).read() == adrStr: return # same content, so never mind else: print "Error: Could not remove", adrFile sys.stdout.flush() raise try: f = open(adrFile, 'w') f.write(adrStr) f.close() except IOError: print "Error: Could not write", adrFile sys.stdout.flush() raise def isPersistent(self): return True def defaultConfig(self): """The default AppServer.config.""" if self._defaultConfig is None: self._defaultConfig = AppServer.defaultConfig(self).copy() # Update with ThreadedAppServer specific settings # as defined in defaultConfig on the module level: self._defaultConfig.update(defaultConfig) return self._defaultConfig _ignoreErrnos = [] # silently ignore these errors: for e in 'EAGAIN', 'EWOULDBLOCK', 'EINTR', 'ECONNABORTED', 'EPROTO': try: _ignoreErrnos.append(getattr(errno, e)) except AttributeError: pass def mainloop(self, timeout=1): """Main thread loop. This is the main thread loop that accepts and dispatches socket requests. It goes through a loop as long as ``self._running > 2``. Setting ``self._running = 2`` asks the the main loop to end. When the main loop is finished, it sets ``self._running = 1``. When the AppServer is completely down, it sets ``self._running = 0``. The loop waits for connections, then based on the connecting port it initiates the proper Handler (e.g., AdapterHandler, HTTPHandler). Handlers are reused when possible. The initiated handlers are put into a queue, and worker threads poll that queue to look for requests that need to be handled (worker threads use `threadloop`). Every so often (every 5 loops) it updates thread usage information (`updateThreadUsage`), and every ``MaxServerThreads * 2`` loops it it will manage threads (killing or spawning new ones, in `manageThreadCount`). """ threadCheckInterval = self._maxServerThreads * 2 threadUpdateDivisor = 5 # grab stat interval threadCheck = 0 self._running = 3 # server is in the main loop now try: while self._running > 2: # block for timeout seconds waiting for connections try: input, output, exc = select.select( self._sockets.values(), [], [], timeout) except select.error, e: if e[0] not in self._ignoreErrnos: raise if debug: print "Socket select error:", e continue for sock in input: try: client, addr = sock.accept() except select.error, e: if e[0] not in self._ignoreErrnos: raise if debug: print "Socket accept error:", e continue serverAddress = sock.getsockname() try: handler = self._handlerCache[serverAddress].pop() except IndexError: handler = self._socketHandlers[serverAddress](self, serverAddress) self._requestID += 1 handler.activate(client, self._requestID) self._requestQueue.put(handler) if threadCheck % threadUpdateDivisor == 0: self.updateThreadUsage() if threadCheck > threadCheckInterval: threadCheck = 0 self.manageThreadCount() else: threadCheck += 1 self.abortLongRequests() self.restartIfNecessary() finally: self._running = 1 ## Thread Management ## # These methods handle the thread pool. The AppServer pre-allocates # threads, and reuses threads for requests. So as more threads # are needed with varying load, new threads are spawned, and if there # are excess threads, then threads are removed. def updateThreadUsage(self): """Update the threadUseCounter list. Called periodically from `mainloop`. """ count = self.activeThreadCount() if len(self._threadUseCounter) > self._maxServerThreads: self._threadUseCounter.pop(0) self._threadUseCounter.append(count) def activeThreadCount(self): """Get a snapshot of the number of threads currently in use. Called from `updateThreadUsage`. """ count = 0 for t in self._threadPool: if t._processing: count += 1 return count def manageThreadCount(self): """Adjust the number of threads in use. From information gleened from `updateThreadUsage`, we see about how many threads are being used, to see if we have too many threads or too few. Based on this we create or absorb threads. """ # @@: This algorithm needs work. The edges (i.e. at the # minserverthreads) are tricky. When working with this, # remember thread creation is *cheap*. average = max = 0 if debug: print "ThreadUse Samples:", self._threadUseCounter for i in self._threadUseCounter: average += i if i > max: max = i average /= len(self._threadUseCounter) if debug: print "Average Thread Use: ", average print "Max Thread Use: ", max print "ThreadCount: ", self._threadCount if len(self._threadUseCounter) < self._maxServerThreads: return # not enough samples margin = self._threadCount / 2 # smoothing factor if debug: print "Margin:", margin if average > self._threadCount - margin and \ self._threadCount < self._maxServerThreads: # Running low: double thread count n = min(self._threadCount, self._maxServerThreads - self._threadCount) if debug: print "Adding %s threads" % n for i in range(n): self.spawnThread() elif average < self._threadCount - margin and \ self._threadCount > self._minServerThreads: n = min(self._threadCount - self._minServerThreads, self._threadCount - max) self.absorbThread(n) else: # cleanup any stale threads that we killed but haven't joined self.absorbThread(0) def spawnThread(self): """Create a new worker thread. Worker threads poll with the `threadloop` method. """ if debug: print "Spawning new thread" t = WorkerThread(target=self.threadloop) t._processing = False t.start() self._threadPool.append(t) self._threadCount += 1 if debug: print "New thread spawned, threadCount =", self._threadCount def absorbThread(self, count=1): """Absorb a thread. We do this by putting a None on the Queue. When a thread gets it, that tells it to exit. We also keep track of the threads, so after killing threads we go through all the threads and find the thread(s) that have exited, so that we can take them out of the thread pool. """ for i in range(count): self._requestQueue.put(None) # _threadCount is an estimate, just because we # put None in the queue, the threads don't immediately # disappear, but they will eventually. self._threadCount -= 1 for t in self._threadPool: # There may still be a None in the queue, and some # of the threads we want gone may not yet be gone. # But we'll pick them up later -- they'll wait. if not t.isAlive(): t.join() # Don't need a timeout, it isn't alive self._threadPool.remove(t) if debug: print "Thread absorbed, real threadCount =", len(self._threadPool) _canAbortRequest = WorkerThread._canAbort def abortRequest(self, requestID, exception=RequestAbortedError): """Abort a request by raising an exception in its worker thread. A return value of one means the thread was successfully aborted, a value of zero means the thread could not be found, any other value indicates that an error has occurred. """ verbose = self._verbose if verbose: print "Aborting request", requestID if not self._canAbortRequest: if verbose: print "Error: Cannot abort requests" return -1 for t, h in self._threadHandler.items(): try: handlerRequestID = h._requestID except AttributeError: handlerRequestID = None if requestID == handlerRequestID: t._abortHandler = h try: if self._threadHandler[t] is not h: # request already finished in the meantime raise KeyError ret = t.abort(exception) except Exception: ret = 0 t._abortHandler = None break else: ret = 0 if verbose: if ret == 0: print "Error: Could not find thread for this request" elif ret == 1: print "The worker thread for this request has been aborted" else: print "Error: Could not abort thread for this request" return ret def abortLongRequests(self): """Check for long-running requests and cancel these. The longest allowed execution time for requests is controlled by the MaxRequestTime setting. """ if self._checkRequestTime is None: return currentTime = time.time() if currentTime > self._checkRequestTime: if debug: print "Checking for long-running requests" verbose = self._verbose minRequestTime = currentTime - self._maxRequestTime for t, h in self._threadHandler.items(): try: requestDict = h._requestDict requestID = requestDict['requestID'] requestTime = requestDict['time'] except (AttributeError, KeyError): continue if requestTime < minRequestTime: t._abortHandler = h try: if self._threadHandler[t] is not h: # request already finished in the meantime raise KeyError if verbose: print "Aborting long-running request", requestID t.abort(RequestTooLongError) except Exception: pass t._abortHandler = None elif requestTime < currentTime: currentTime = requestTime self._checkRequestTime = currentTime + self._maxRequestTime ## Worker Threads ## def threadloop(self): """The main loop for worker threads. Worker threads poll the `_requestQueue` to find a request handler waiting to run. If they find a None in the queue, this thread has been selected to die, which is the way the loop ends. The handler object does all the work when its `handleRequest` method is called. `initThread` and `delThread` methods are called at the beginning and end of the thread loop, but they aren't being used for anything (future use as a hook). """ self.initThread() t = currentThread() t._processing = False t._abortHandler = None try: while 1: try: handler = self._requestQueue.get() except Queue.Empty: continue if handler is None: # None means time to quit break try: t._processing = True self._threadHandler[t] = handler try: handler.handleRequest() except ThreadAbortedError: print "Worker thread has been aborted" except Exception: print "Exception in worker thread" traceback.print_exc(file=sys.stderr) del self._threadHandler[t] t._processing = False finally: handler.close() while t._abortHandler is handler: # this handler is to be aborted, # so don't handle another request now time.sleep(0.1) finally: try: del self._threadHandler[t] t._processing = False except KeyError: pass self.delThread() if debug: print "Quitting", t def initThread(self): """Initialize thread. Invoked immediately by threadloop() as a hook for subclasses. This implementation does nothing and subclasses need not invoke super. """ pass def delThread(self): """Delete thread. Invoked immediately by threadloop() as a hook for subclasses. This implementation does nothing and subclasses need not invoke super. """ pass ## Shutting Down ## def shutDown(self): """Called on shutdown. Also calls `AppServer.shutDown`, but first closes all sockets and tells all the threads to die. """ print "ThreadedAppServer is shutting down..." if self._running > 2: self._running = 2 # ask main loop to finish self.awakeSelect() # unblock select call in mainloop() sys.stdout.flush() for i in range(30): # wait at most 3 seconds for shutdown if self._running < 2: break time.sleep(0.1) if self._sockets: # Close all sockets now: for sock in self._sockets.values(): sock.close() if self._socketHandlers: # Remove the text files with the server addresses: for handler in self._socketHandlers.values(): adrFile = self.addressFileName(handler) if os.path.exists(adrFile): try: os.unlink(adrFile) except OSError: print "Warning: Could not remove", adrFile # Tell all threads to end: for i in range(self._threadCount): self._requestQueue.put(None) if self._canAbortRequest: for t in self._threadHandler.keys(): t.abort(ServerShutDownError) for t in self._threadPool: try: t.join() except Exception: pass # Call super's shutdown: AppServer.shutDown(self) def awakeSelect(self): """Awake the select() call. The `select()` in `mainloop()` is blocking, so when we shut down we have to make a connect to unblock it. Here's where we do that. """ for host, port in self._sockets.keys(): if host == '0.0.0.0': # Can't connect to 0.0.0.0; use 127.0.0.1 instead host = '127.0.0.1' sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM) try: sock.connect((host, port)) sock.close() except Exception: pass ## Misc ## def address(self, settingPrefix): """Get host address. The address for the Adapter (Host/interface, and port), as taken from ``Configs/AppServer.config``, settings ``Host`` and ``AdapterPort``. """ try: return self._addr[settingPrefix] except KeyError: host = self.setting(settingPrefix + 'Host', self.setting('Host')) if settingPrefix == 'Adapter': # jdh 2004-12-01: # 'Port' has been renamed to 'AdapterPort'. However, we don't # want the the default AdapterPort in DefaultConfig above to # be used if a user still has 'Port' in their config file. # So for now, we prefer the 'Port' setting if it exists. # After a few releases we can remove this special case. port = self.setting('Port', None) if port is None: port = self.setting(settingPrefix + 'Port') else: print "WARNING:", \ "The 'Port' setting has been renamed to 'AdapterPort'." print "Please update your AppServer.config file." else: port = self.setting(settingPrefix + 'Port') self._addr[settingPrefix] = (host, port) return self._addr[settingPrefix] def addressFileName(self, handlerClass): """Get the name of the text file with the server address.""" return self.serverSidePath( self.setting('AddressFiles') % handlerClass.protocolName) class Handler: """A very general socket handler. Handler is an abstract superclass -- specific protocol implementations will subclass this. A Handler takes a socket to interact with, and creates a raw request. Handlers will be reused. When a socket is received `activate` will be called -- but the handler should not do anything, as it is still running in the main thread. The handler is put into a queue, and a worker thread picks it up and runs `handleRequest`, which subclasses should override. Several methods are provided which are typically used by subclasses. """ def __init__(self, server, serverAddress): """Create a new socket handler. Each handler is attached to a specific host and port, and of course to the AppServer. """ self._server = server self._serverAddress = serverAddress self._verbose = server._verbose def activate(self, sock, requestID): """Activate the handler for processing the request. `sock` is the incoming socket that this handler will work with, and `requestID` is a serial number unique for each request. This isn't where work gets done -- the handler is queued after this, and work is done when `handleRequest` is called. """ self._requestID = requestID self._sock = sock def close(self): """Close the socket. Called when the handler is finished. Closes the socket and returns the handler to the pool of inactive handlers. """ self._sock = None self._server._handlerCache[self._serverAddress].append(self) def receiveDict(self): """Receive a dictionary from the socket. Utility function to receive a marshalled dictionary from the socket. Returns None if the request was empty. """ chunk = '' missing = intLength while missing > 0: block = self._sock.recv(missing) if not block: self._sock.close() if not chunk: # We probably awakened due to awakeSelect being called. return None # We got a partial request -- something went wrong. raise NotEnoughDataError, 'received only %d of %d bytes' \ ' when receiving dictLength' % (len(chunk), intLength) chunk += block missing -= len(block) try: dictLength = loads(chunk) except (ValueError, EOFError), msg: if chunk[:3] == 'GET': # Common error: client is speaking HTTP. while msg and len(chunk) < 8192: block = self._sock.recv(1) if not block: break chunk += block if chunk.endswith('\r\r') or chunk.endswith('\n\n') \ or chunk.endswith('\r\n\r\n'): msg = None if msg: print "ERROR:", msg else: print "ERROR: HTTP GET from WebKit adapter port." self._sock.sendall('''\ HTTP/1.0 505 HTTP Version Not Supported\r Content-type: text/plain\r \r Error: Invalid AppServer protocol.\r Sorry, I don't speak HTTP. You must connect via an adapter.\r See the Troubleshooting section of the WebKit Install Guide.\r''') self._sock.close() print " You can only connect to", self._serverAddress[1], \ "via an adapter like mod_webkit or wkcgi." return None if type(dictLength) != type(1): self._sock.close() raise ProtocolError, "Invalid AppServer protocol" chunk = '' missing = dictLength while missing > 0: block = self._sock.recv(missing) if not block: self._sock.close() raise NotEnoughDataError, 'received only %d of %d bytes' \ ' when receiving dict' % (len(chunk), dictLength) chunk += block missing -= len(block) return loads(chunk) def handleRequest(self): """Handle a raw request. This is where the work gets done. Subclasses should override. """ pass def startRequest(self, requestDict=None): """Track start of a raw request. Subclasses can use and override this method. """ requestDict = requestDict or {} requestID = self._requestID requestTime = requestDict.get('time') or time.time() requestDict['requestID'] = requestID requestDict['time'] = requestTime # The request object is stored for tracking/debugging purposes. self._requestDict = requestDict if self._verbose: requestTime = time.localtime(requestTime)[:6] env = requestDict.get('environ') uri = env and requestURI(env) or '-' print '%5d %4d-%02d-%02d %02d:%02d:%02d %s' % ( (requestID,) + requestTime + (uri,)) def endRequest(self, error=None): """Track end of a raw request. Subclasses can use and override this method. """ if self._verbose: requestDict = self._requestDict requestID = requestDict['requestID'] duration = round((time.time() - requestDict['time'])*1000) env = requestDict.get('environ') if not error: error = env and requestURI(env) or '-' print '%5d %14.0f msec %s\n' % ( requestID, duration, error) class MonitorHandler(Handler): """Monitor server status. Monitor is a minimal service that accepts a simple protocol, and returns a value indicating the status of the server. The protocol passes a marshalled dict, much like the Adapter interface, which looks like ``{'format': 'CMD'}``, where CMD is a command (``STATUS`` or ``QUIT``). Responds with a simple string, either the number of requests we've received (for ``STATUS``) or ``OK`` for ``QUIT`` (which also stops the server). """ # @@ 2003-03 ib: we should have a RESTART command, and # perhaps better status indicators (number of threads, etc). protocolName = 'monitor' settingPrefix = 'Monitor' def handleRequest(self): requestDict = self.receiveDict() if not requestDict: return requestDict['environ'] = { 'REQUEST_URI': '*%s %s*' % (self.settingPrefix, requestDict['format'])} self.startRequest(requestDict) conn = self._sock if requestDict['format'] == "STATUS": conn.send(str(self._server._requestID)) elif requestDict['format'] == 'QUIT': conn.send("OK") conn.close() self._server.shutDown() class TASStreamOut(ASStreamOut): """Response stream for ThreadedAppServer. The `TASStreamOut` class streams to a given socket, so that when `flush` is called and the buffer is ready to be written, it sends the data from the buffer out on the socket. This is the response stream used for requests generated by ThreadedAppServer. """ _ignoreErrnos = [] # silently ignore these errors: for e in 'EPIPE', 'ECONNABORTED', 'ECONNRESET': try: _ignoreErrnos.append(getattr(errno, e)) except AttributeError: pass def __init__(self, sock, autoCommit=False, bufferSize=8192): """Create stream. We get an extra `sock` argument, which is the socket which we'll stream output to (if we're streaming). """ ASStreamOut.__init__(self, autoCommit, bufferSize) self._socket = sock def flush(self): """Flush stream. Calls `ASStreamOut.ASStreamOut.flush`, and if that returns True (indicating the buffer is full enough) then we send data from the buffer out on the socket. """ result = ASStreamOut.flush(self) if result: # a true return value means we can send reslen = len(self._buffer) sent = 0 bufferSize = self._bufferSize while sent < reslen: try: sent += self._socket.send( self._buffer[sent:sent+bufferSize]) except socket.error, e: if debug or e[0] not in self._ignoreErrnos: print "StreamOut Error:", e self._closed = True raise ConnectionAbortedError self.pop(sent) class AdapterHandler(Handler): """Adapter handler. Handles the Adapter protocol (as used in mod_webkit, wkcgi, WebKit.cgi, HTTPAdapter, etc). This protocol passes a marshalled dictionary which contains the keys ``format`` and ``environ``. ``format`` is currently always the string ``CGI``, and ``environ`` is a dictionary of string: string, with values like those passed in the environment to a CGI request (QUERY_STRING, HTTP_HOST, etc). The handler adds one more key, ``input``, which contains a file object based off the socket, which contains the body of the request (the POST data, for instance). It's left to Application to handle that data. """ protocolName = 'adapter' settingPrefix = 'Adapter' def handleRequest(self): """Handle request. Creates the request dictionary, and creates a `TASStreamOut` object for the response, then calls `Application.dispatchRawRequest`, which does the rest of the work (here we just clean up after). """ requestDict = self.receiveDict() if not requestDict: return self.startRequest(requestDict) requestDict['input'] = self.makeInput() streamOut = TASStreamOut(self._sock, bufferSize=self._server._responseBufferSize) transaction = self._server._app.dispatchRawRequest(requestDict, streamOut) try: streamOut.close() aborted = False except ConnectionAbortedError: aborted = True try: self._sock.shutdown(1) self._sock.close() except Exception: pass self.endRequest(aborted and '*connection aborted*') transaction._application = None transaction.die() del transaction def makeInput(self): """Create a file-like object from the socket.""" return self._sock.makefile("rb", self._server._requestBufferSize) class SCGIHandler(AdapterHandler): """SCGI handler. Modified Adapter handler speaking the SCGI protocol. """ protocolName = 'scgi' settingPrefix = 'SCGI' def receiveDict(self): """Receive a dictionary from the socket. Utility function to receive the SCGI headers from the socket. Returns None if the request was empty. """ chunk = '' while 1: c = self._sock.recv(1) if not c and not chunk: self._sock.close() return None if c == ':': break else: chunk += c if len(chunk) > 12: break try: if len(chunk) > 12 or not chunk.isdigit(): raise ValueError, 'Malformed SCGI netstring' dictLength = long(chunk) except ValueError, msg: if chunk[:3] == 'GET': # Common error: client is speaking HTTP. while msg and len(chunk) < 8192: block = self._sock.recv(1) if not block: break chunk += block if chunk.endswith('\r\r') or chunk.endswith('\n\n') \ or chunk.endswith('\r\n\r\n'): msg = None if msg: print "ERROR:", msg else: print "ERROR: HTTP GET from SCGI adapter port." self._sock.sendall('''\ HTTP/1.0 505 HTTP Version Not Supported\r Content-type: text/plain\r \r Error: Invalid AppServer protocol.\r Sorry, I don't speak HTTP. You must connect via an SCGI adapter.\r See the Troubleshooting section of the WebKit Install Guide.\r''') self._sock.close() print " You can only connect to", self._serverAddress[1], \ "via an adapter like mod_scgi or pyscgi." return None chunk = '' missing = dictLength while missing > 0: block = self._sock.recv(missing) if not block: self._sock.close() raise NotEnoughDataError, 'received only %d of %d bytes' \ ' when receiving netstring' % (len(chunk), dictLength) chunk += block missing -= len(block) if self._sock.recv(1) != ',': self._sock.close() raise ProtocolError, 'Missing SCGI netstring terminator' items = chunk.split('\0')[:-1] environ = {} try: for i in range(0, len(items), 2): environ[items[i]] = items[i+1] except IndexError: raise ProtocolError, 'Malformed SCGI headers' return { 'format': 'CGI', 'time': time.time(), 'environ': environ } # Determines whether the main look should run in another thread. # On Win NT/2K/XP, we run the mainloop in a different thread because # it's not safe for Ctrl-C to be caught while manipulating the queues. # It's not safe on Linux either, but there, it appears that Ctrl-C will # trigger an exception in ANY thread, so this fix doesn't help. def runMainLoopInThread(): return os.name == 'nt' # Set to False in DebugAppServer so Python debuggers can trap exceptions: doesRunHandleExceptions = True class RestartAppServerError(Exception): """Raised by DebugAppServer when needed.""" pass _chdir = os.chdir def chdir(path, force=False): """Execute os.chdir() with safety provision.""" assert force, \ "You cannot reliably use os.chdir() in a threaded environment.\n" \ + 16*" " + "Set force=True if you want to do it anway (using a lock)." _chdir(path) ## Script usage ## def run(workDir=None): """Start the server (`ThreadedAppServer`). `workDir` is the server-side path for the server, which may not be the ``Webware/WebKit`` directory (though by default it is). After setting up the ThreadedAppServer we call `ThreadedAppServer.mainloop` to start the server main loop. It also catches exceptions as a last resort. """ global server server = None global exitStatus exitStatus = 0 os.chdir = chdir # inhibit use of os.chdir() runAgain = True while runAgain: # looping in support of RestartAppServerError try: try: runAgain = False server = ThreadedAppServer(workDir) if runMainLoopInThread(): # catch the exception raised by sys.exit so # that we can re-call it in the main thread. def _windowsmainloop(): global exitStatus try: server.mainloop() except SystemExit, e: exitStatus = e[0] # Run the server thread t = Thread(target=_windowsmainloop) t.start() try: while server._running > 1: try: time.sleep(1) # wait for interrupt except Exception: if server._running < 3: raise # shutdown finally: t.join() else: server.mainloop() sys.exit(exitStatus) except RestartAppServerError: print print "Restarting AppServer:" sys.stdout.flush() sys.stderr.flush() runAgain = True except SystemExit, e: print print "Exiting AppServer%s." % ( e[0] == 3 and ' for reload' or '') exitStatus = e[0] except KeyboardInterrupt: print print "Exiting AppServer due to keyboard interrupt." exitStatus = 0 except Exception, e: if isinstance(e, IOError) and e[0] == errno.EINTR: print print "Exiting AppServer due to interrupt signal." exitStatus = 0 else: if doesRunHandleExceptions: if not server and isinstance(e, ProcessRunning): print "Error:", str(e) else: print traceback.print_exc() print print "Exiting AppServer due to above exception." exitStatus = 1 else: raise finally: sys.stdout.flush() sys.stderr.flush() if server and server._running: server.initiateShutdown() server._closeThread.join() AppServerModule.globalAppServer = None sys.stdout.flush() sys.stderr.flush() os.chdir = _chdir # allow use of os.chdir() again return exitStatus # Signal handlers def shutDown(signum, frame): """Signal handler for shutting down the server.""" print print "App server has been signaled to shutdown." if server and server._running > 2: print "Shutting down at", asclocaltime() sys.stdout.flush() server._running = 2 if signum == SIGINT: raise KeyboardInterrupt elif signum == SIGHUP: sys.exit(3) # force reload else: sys.exit(0) # normal exit else: print "No running app server was found." try: # Use the threadframe module for dumping thread stack frames: # http://www.majid.info/mylos/stories/2004/06/10/threadframe.html import threadframe def threadDump(signum, frame): """Signal handler for dumping thread stack frames to stdout.""" print print "App server has been signaled to attempt a thread dump." print print "Thread stack frame dump at", asclocaltime() sys.stdout.flush() frames = threadframe.dict() items = frames.items() items.sort() print print "-" * 79 print for threadID, frame in items: print "Thread ID: %d (reference count = %d)" % ( threadID, sys.getrefcount(frame)) print ''.join(traceback.format_list(traceback.extract_stack(frame))) items.sort() print "-" * 79 sys.stdout.flush() except ImportError: # threadframe module not available threadDump = None import signal # Shutdown signals try: SIGHUP = signal.SIGHUP signal.signal(SIGHUP, shutDown) except AttributeError: SIGHUP = None try: SIGTERM = signal.SIGTERM signal.signal(SIGTERM, shutDown) except AttributeError: SIGTERM = None try: # this is Ctrl-C on Windows SIGINT = signal.SIGINT signal.signal(SIGINT, shutDown) except AttributeError: SIGINT = None if threadDump: # Signals for creating a thread dump try: SIGQUIT = signal.SIGQUIT signal.signal(SIGQUIT, threadDump) except AttributeError: SIGQUIT = None try: # this is Ctrl-Break on Windows (not Cygwin) SIGBREAK = signal.SIGBREAK signal.signal(SIGBREAK, threadDump) except AttributeError: SIGBREAK = None import re settingRE = re.compile(r'^(?:--)?([a-zA-Z][a-zA-Z0-9]*\.[a-zA-Z][a-zA-Z0-9]*)=') from MiscUtils import Configurable usage = re.search('\n.* arguments:\n\n(.*\n)*?\n', __doc__).group(0) def main(args): """Command line interface. Run by `Launch`, this is the main entrance and command-line interface for ThreadedAppServer. """ function = run daemon = False workDir = None for a in args[:]: if settingRE.match(a): match = settingRE.match(a) name = match.group(1) value = a[match.end():] Configurable.addCommandLineSetting(name, value) elif a == "stop": function = AppServerModule.stop elif a == "daemon": daemon = True elif a == "start": pass elif a[:8] == "workdir=": workDir = a[8:] else: print usage return if daemon: if os.name == "posix": pid = os.fork() if pid: sys.exit() else: print "Daemon mode not available on your OS." return function(workDir=workDir)