use wrap_exception in controller, fix clear on kernel

IPython/zmq/parallel/controller.py

@@ -1,905 +1,933
 #!/usr/bin/env python
 """The IPython Controller with 0MQ
 This is the master object that handles connections from engines and clients,
 and monitors traffic through the various queues.
 """
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2010  The IPython Development Team
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 from __future__ import print_function
 
 from datetime import datetime
 import logging
 
 import zmq
 from zmq.eventloop import zmqstream, ioloop
 import uuid
 
 # internal:
 from IPython.zmq.log import logger # a Logger object
 from IPython.zmq.entry_point import bind_port
 
 from streamsession import Message, wrap_exception
 from entry_point import (make_base_argument_parser, select_random_ports, split_ports,
                         connect_logger, parse_url)
 
 #-----------------------------------------------------------------------------
 # Code
 #-----------------------------------------------------------------------------
 
 def _passer(*args, **kwargs):
     return
 
 class ReverseDict(dict):
     """simple double-keyed subset of dict methods."""
     
     def __init__(self, *args, **kwargs):
         dict.__init__(self, *args, **kwargs)
         self.reverse = dict()
         for key, value in self.iteritems():
             self.reverse[value] = key
     
     def __getitem__(self, key):
         try:
             return dict.__getitem__(self, key)
         except KeyError:
             return self.reverse[key]
     
     def __setitem__(self, key, value):
         if key in self.reverse:
             raise KeyError("Can't have key %r on both sides!"%key)
         dict.__setitem__(self, key, value)
         self.reverse[value] = key
     
     def pop(self, key):
         value = dict.pop(self, key)
         self.d1.pop(value)
         return value
     
 
 class EngineConnector(object):
     """A simple object for accessing the various zmq connections of an object.
     Attributes are:
     id (int): engine ID
     uuid (str): uuid (unused?)
     queue (str): identity of queue's XREQ socket
     registration (str): identity of registration XREQ socket
     heartbeat (str): identity of heartbeat XREQ socket
     """
     id=0
     queue=None
     control=None
     registration=None
     heartbeat=None
     pending=None
     
     def __init__(self, id, queue, registration, control, heartbeat=None):
         logger.info("engine::Engine Connected: %i"%id)
         self.id = id
         self.queue = queue
         self.registration = registration
         self.control = control
         self.heartbeat = heartbeat
         
 class Controller(object):
     """The IPython Controller with 0MQ connections
     
     Parameters
     ==========
     loop: zmq IOLoop instance
     session: StreamSession object
     <removed> context: zmq context for creating new connections (?)
     queue: ZMQStream for monitoring the command queue (SUB)
     registrar: ZMQStream for engine registration requests (XREP)
     heartbeat: HeartMonitor object checking the pulse of the engines
     clientele: ZMQStream for client connections (XREP)
                 not used for jobs, only query/control commands
     notifier: ZMQStream for broadcasting engine registration changes (PUB)
     db: connection to db for out of memory logging of commands
                 NotImplemented
     engine_addrs: dict of zmq connection information for engines to connect
                 to the queues.
     client_addrs: dict of zmq connection information for engines to connect
                 to the queues.
     """
     # internal data structures:
     ids=None # engine IDs
     keytable=None
     engines=None
     clients=None
     hearts=None
     pending=None
     results=None
     tasks=None
     completed=None
     mia=None
     incoming_registrations=None
     registration_timeout=None
     
     #objects from constructor:
     loop=None
     registrar=None
     clientelle=None
     queue=None
     heartbeat=None
     notifier=None
     db=None
     client_addr=None
     engine_addrs=None
     
     
     def __init__(self, loop, session, queue, registrar, heartbeat, clientele, notifier, db, engine_addrs, client_addrs):
         """
         # universal:
         loop: IOLoop for creating future connections
         session: streamsession for sending serialized data
         # engine:
         queue: ZMQStream for monitoring queue messages
         registrar: ZMQStream for engine registration
         heartbeat: HeartMonitor object for tracking engines
         # client:
         clientele: ZMQStream for client connections
         # extra:
         db: ZMQStream for db connection (NotImplemented)
         engine_addrs: zmq address/protocol dict for engine connections
         client_addrs: zmq address/protocol dict for client connections
         """
         self.ids = set()
         self.keytable={}
         self.incoming_registrations={}
         self.engines = {}
         self.by_ident = {}
         self.clients = {}
         self.hearts = {}
         self.mia = set()
         
         # self.sockets = {}
         self.loop = loop
         self.session = session
         self.registrar = registrar
         self.clientele = clientele
         self.queue = queue
         self.heartbeat = heartbeat
         self.notifier = notifier
         self.db = db
         
         # validate connection dicts:
         self.client_addrs = client_addrs
         assert isinstance(client_addrs['queue'], str)
         assert isinstance(client_addrs['control'], str)
         # self.hb_addrs = hb_addrs
         self.engine_addrs = engine_addrs
         assert isinstance(engine_addrs['queue'], str)
         assert isinstance(client_addrs['control'], str)
         assert len(engine_addrs['heartbeat']) == 2
         
         # register our callbacks
         self.registrar.on_recv(self.dispatch_register_request)
         self.clientele.on_recv(self.dispatch_client_msg)
         self.queue.on_recv(self.dispatch_queue_traffic)
         
         if heartbeat is not None:
             heartbeat.add_heart_failure_handler(self.handle_heart_failure)
             heartbeat.add_new_heart_handler(self.handle_new_heart)
         
         self.queue_handlers = { 'in' : self.save_queue_request,
                                 'out': self.save_queue_result,
                                 'intask': self.save_task_request,
                                 'outtask': self.save_task_result,
                                 'tracktask': self.save_task_destination,
                                 'incontrol': _passer,
                                 'outcontrol': _passer,
         }
         
         self.client_handlers = {'queue_request': self.queue_status,
                                 'result_request': self.get_results,
                                 'purge_request': self.purge_results,
                                 'load_request': self.check_load,
                                 'resubmit_request': self.resubmit_task,
                                 }
         
         self.registrar_handlers = {'registration_request' : self.register_engine,
                                 'unregistration_request' : self.unregister_engine,
                                 'connection_request': self.connection_request,
         }
         # 
         # this is the stuff that will move to DB:
         self.results = {} # completed results
         self.pending = {} # pending messages, keyed by msg_id
         self.queues = {} # pending msg_ids keyed by engine_id
         self.tasks = {} # pending msg_ids submitted as tasks, keyed by client_id
         self.completed = {} # completed msg_ids keyed by engine_id
         self.registration_timeout = max(5000, 2*self.heartbeat.period)
         
         logger.info("controller::created controller")
     
     def _new_id(self):
         """gemerate a new ID"""
         newid = 0
         incoming = [id[0] for id in self.incoming_registrations.itervalues()]
         # print newid, self.ids, self.incoming_registrations
         while newid in self.ids or newid in incoming:
             newid += 1
         return newid
     
     #-----------------------------------------------------------------------------
     # message validation
     #-----------------------------------------------------------------------------
     
     def _validate_targets(self, targets):
         """turn any valid targets argument into a list of integer ids"""
         if targets is None:
             # default to all
             targets = self.ids
             
         if isinstance(targets, (int,str,unicode)):
             # only one target specified
             targets = [targets]
         _targets = []
         for t in targets:
             # map raw identities to ids
             if isinstance(t, (str,unicode)):
                 t = self.by_ident.get(t, t)
             _targets.append(t)
         targets = _targets
         bad_targets = [ t for t in targets if t not in self.ids ]
         if bad_targets:
             raise IndexError("No Such Engine: %r"%bad_targets)
         if not targets:
             raise IndexError("No Engines Registered")
         return targets
     
     def _validate_client_msg(self, msg):
         """validates and unpacks headers of a message. Returns False if invalid,
         (ident, header, parent, content)"""
         client_id = msg[0]
         try:
             msg = self.session.unpack_message(msg[1:], content=True)
         except:
             logger.error("client::Invalid Message %s"%msg)
             return False
         
         msg_type = msg.get('msg_type', None)
         if msg_type is None:
             return False
         header = msg.get('header')
         # session doesn't handle split content for now:
         return client_id, msg
         
     
     #-----------------------------------------------------------------------------
     # dispatch methods (1 per stream)
     #-----------------------------------------------------------------------------
     
     def dispatch_register_request(self, msg):
         """"""
         logger.debug("registration::dispatch_register_request(%s)"%msg)
         idents,msg = self.session.feed_identities(msg)
         print (idents,msg, len(msg))
         try:
             msg = self.session.unpack_message(msg,content=True)
         except Exception as e:
             logger.error("registration::got bad registration message: %s"%msg)
             raise e
             return
         
         msg_type = msg['msg_type']
         content = msg['content']
         
         handler = self.registrar_handlers.get(msg_type, None)
         if handler is None:
             logger.error("registration::got bad registration message: %s"%msg)
         else:
             handler(idents, msg)
     
     def dispatch_queue_traffic(self, msg):
         """all ME and Task queue messages come through here"""
         logger.debug("queue traffic: %s"%msg[:2])
         switch = msg[0]
         idents, msg = self.session.feed_identities(msg[1:])
         handler = self.queue_handlers.get(switch, None)
         if handler is not None:
             handler(idents, msg)
         else:
             logger.error("Invalid message topic: %s"%switch)
         
     
     def dispatch_client_msg(self, msg):
         """Route messages from clients"""
         idents, msg = self.session.feed_identities(msg)
         client_id = idents[0]
         try:
             msg = self.session.unpack_message(msg, content=True)
         except:
             content = wrap_exception()
             logger.error("Bad Client Message: %s"%msg)
             self.session.send(self.clientele, "controller_error", ident=client_id, 
                     content=content)
             return
         
         # print client_id, header, parent, content
         #switch on message type:
         msg_type = msg['msg_type']
         logger.info("client:: client %s requested %s"%(client_id, msg_type))
         handler = self.client_handlers.get(msg_type, None)
         try:
             assert handler is not None, "Bad Message Type: %s"%msg_type
         except:
             content = wrap_exception()
             logger.error("Bad Message Type: %s"%msg_type)
             self.session.send(self.clientele, "controller_error", ident=client_id, 
                     content=content)
             return
         else:
             handler(client_id, msg)
             
     def dispatch_db(self, msg):
         """"""
         raise NotImplementedError
     
     #---------------------------------------------------------------------------
     # handler methods (1 per event)
     #---------------------------------------------------------------------------
     
     #----------------------- Heartbeat --------------------------------------
     
     def handle_new_heart(self, heart):
         """handler to attach to heartbeater.
         Called when a new heart starts to beat.
         Triggers completion of registration."""
         logger.debug("heartbeat::handle_new_heart(%r)"%heart)
         if heart not in self.incoming_registrations:
             logger.info("heartbeat::ignoring new heart: %r"%heart)
         else:
             self.finish_registration(heart)
         
     
     def handle_heart_failure(self, heart):
         """handler to attach to heartbeater.
         called when a previously registered heart fails to respond to beat request.
         triggers unregistration"""
         logger.debug("heartbeat::handle_heart_failure(%r)"%heart)
         eid = self.hearts.get(heart, None)
         queue = self.engines[eid].queue
         if eid is None:
             logger.info("heartbeat::ignoring heart failure %r"%heart)
         else:
             self.unregister_engine(heart, dict(content=dict(id=eid, queue=queue)))
     
     #----------------------- MUX Queue Traffic ------------------------------
     
     def save_queue_request(self, idents, msg):
         queue_id, client_id = idents[:2]
             
         try:
             msg = self.session.unpack_message(msg, content=False)
         except:
             logger.error("queue::client %r sent invalid message to %r: %s"%(client_id, queue_id, msg))
             return
         
         eid = self.by_ident.get(queue_id, None)
         if eid is None:
             logger.error("queue::target %r not registered"%queue_id)
             logger.debug("queue::    valid are: %s"%(self.by_ident.keys()))
             return
             
         header = msg['header']
         msg_id = header['msg_id']
         info = dict(submit=datetime.now(),
                     received=None,
                     engine=(eid, queue_id))
         self.pending[msg_id] = ( msg, info )
         self.queues[eid].append(msg_id)
     
     def save_queue_result(self, idents, msg):
         client_id, queue_id = idents[:2]
         
         try:
             msg = self.session.unpack_message(msg, content=False)
         except:
             logger.error("queue::engine %r sent invalid message to %r: %s"%(
                     queue_id,client_id, msg))
             return
         
         eid = self.by_ident.get(queue_id, None)
         if eid is None:
             logger.error("queue::unknown engine %r is sending a reply: "%queue_id)
             logger.debug("queue::       %s"%msg[2:])
             return
         
         parent = msg['parent_header']
         if not parent:
             return
         msg_id = parent['msg_id']
         self.results[msg_id] = msg
         if msg_id in self.pending:
             self.pending.pop(msg_id)
             self.queues[eid].remove(msg_id)
             self.completed[eid].append(msg_id)
         else:
             logger.debug("queue:: unknown msg finished %s"%msg_id)
             
     #--------------------- Task Queue Traffic ------------------------------
     
     def save_task_request(self, idents, msg):
         """Save the submission of a task."""
         client_id = idents[0]
         
         try:
             msg = self.session.unpack_message(msg, content=False)
         except:
             logger.error("task::client %r sent invalid task message: %s"%(
                     client_id, msg))
             return
         
         header = msg['header']
         msg_id = header['msg_id']
         self.mia.add(msg_id)
         info = dict(submit=datetime.now(),
                     received=None,
                     engine=None)
         self.pending[msg_id] = (msg, info)
         if not self.tasks.has_key(client_id):
             self.tasks[client_id] = []
         self.tasks[client_id].append(msg_id)
     
     def save_task_result(self, idents, msg):
         """save the result of a completed task."""
-        client_id, engine_uuid = idents[:2]
+        client_id = idents[0]
         try:
             msg = self.session.unpack_message(msg, content=False)
         except:
             logger.error("task::invalid task result message send to %r: %s"%(
                     client_id, msg))
             return
         
         parent = msg['parent_header']
-        eid = self.by_ident[engine_uuid]
         if not parent:
             # print msg
-            # logger.warn("")
+            logger.warn("Task %r had no parent!"%msg)
             return
         msg_id = parent['msg_id']
         self.results[msg_id] = msg
-        if msg_id in self.pending and msg_id in self.tasks[eid]:
+        
+        header = msg['header']
+        engine_uuid = header.get('engine', None)
+        eid = self.by_ident.get(engine_uuid, None)
+        
+        if msg_id in self.pending:
             self.pending.pop(msg_id)
             if msg_id in self.mia:
                 self.mia.remove(msg_id)
+            if eid is not None and msg_id in self.tasks[eid]:
             self.completed[eid].append(msg_id)
             self.tasks[eid].remove(msg_id)
         else:
             logger.debug("task::unknown task %s finished"%msg_id)
     
     def save_task_destination(self, idents, msg):
         try:
             msg = self.session.unpack_message(msg, content=True)
         except:
             logger.error("task::invalid task tracking message")
             return
         content = msg['content']
         print (content)
         msg_id = content['msg_id']
         engine_uuid = content['engine_id']
         eid = self.by_ident[engine_uuid]
         
         logger.info("task::task %s arrived on %s"%(msg_id, eid))
         if msg_id in self.mia:
             self.mia.remove(msg_id)
         else:
             logger.debug("task::task %s not listed as MIA?!"%(msg_id))
         
         self.tasks[eid].append(msg_id)
         self.pending[msg_id][1].update(received=datetime.now(),engine=(eid,engine_uuid))
     
     def mia_task_request(self, idents, msg):
         client_id = idents[0]
         content = dict(mia=self.mia,status='ok')
         self.session.send('mia_reply', content=content, idents=client_id)
         
         
             
     #-------------------------------------------------------------------------
     # Registration requests
     #-------------------------------------------------------------------------
         
     def connection_request(self, client_id, msg):
         """Reply with connection addresses for clients."""
         logger.info("client::client %s connected"%client_id)
         content = dict(status='ok')
         content.update(self.client_addrs)
         jsonable = {}
         for k,v in self.keytable.iteritems():
             jsonable[str(k)] = v
         content['engines'] = jsonable
         self.session.send(self.registrar, 'connection_reply', content, parent=msg, ident=client_id)
     
     def register_engine(self, reg, msg):
         """Register a new engine."""
         content = msg['content']
         try:
             queue = content['queue']
         except KeyError:
             logger.error("registration::queue not specified")
             return
         heart = content.get('heartbeat', None)
         """register a new engine, and create the socket(s) necessary"""
         eid = self._new_id()
         # print (eid, queue, reg, heart)
         
         logger.debug("registration::register_engine(%i, %r, %r, %r)"%(eid, queue, reg, heart))
         
         content = dict(id=eid,status='ok')
         content.update(self.engine_addrs)
         # check if requesting available IDs:
         if queue in self.by_ident:
-            content = {'status': 'error', 'reason': "queue_id %r in use"%queue}
+            try:
+                raise KeyError("queue_id %r in use"%queue)
+            except:
+                content = wrap_exception()
         elif heart in self.hearts: # need to check unique hearts?
-            content = {'status': 'error', 'reason': "heart_id %r in use"%heart}
+            try:
+                raise KeyError("heart_id %r in use"%heart)
+            except:
+                content = wrap_exception()
         else:
             for h, pack in self.incoming_registrations.iteritems():
                 if heart == h:
-                    content = {'status': 'error', 'reason': "heart_id %r in use"%heart}
+                    try:
+                        raise KeyError("heart_id %r in use"%heart)
+                    except:
+                        content = wrap_exception()
                     break
                 elif queue == pack[1]:
-                    content = {'status': 'error', 'reason': "queue_id %r in use"%queue}
+                    try:
+                        raise KeyError("queue_id %r in use"%queue)
+                    except:
+                        content = wrap_exception()
                     break
         
         msg = self.session.send(self.registrar, "registration_reply", 
                 content=content, 
                 ident=reg)
         
         if content['status'] == 'ok':
             if heart in self.heartbeat.hearts:
                 # already beating
                 self.incoming_registrations[heart] = (eid,queue,reg,None)
                 self.finish_registration(heart)
             else:
                 purge = lambda : self._purge_stalled_registration(heart)
                 dc = ioloop.DelayedCallback(purge, self.registration_timeout, self.loop)
                 dc.start()
                 self.incoming_registrations[heart] = (eid,queue,reg,dc)
         else:
-            logger.error("registration::registration %i failed: %s"%(eid, content['reason']))
+            logger.error("registration::registration %i failed: %s"%(eid, content['evalue']))
         return eid
     
     def unregister_engine(self, ident, msg):
         """Unregister an engine that explicitly requested to leave."""
         try:
             eid = msg['content']['id']
         except:
             logger.error("registration::bad engine id for unregistration: %s"%ident)
             return
         logger.info("registration::unregister_engine(%s)"%eid)
         content=dict(id=eid, queue=self.engines[eid].queue)
         self.ids.remove(eid)
         self.keytable.pop(eid)
         ec = self.engines.pop(eid)
         self.hearts.pop(ec.heartbeat)
         self.by_ident.pop(ec.queue)
         self.completed.pop(eid)
         for msg_id in self.queues.pop(eid):
             msg = self.pending.pop(msg_id)
             ############## TODO: HANDLE IT ################
         
         if self.notifier:
             self.session.send(self.notifier, "unregistration_notification", content=content)
     
     def finish_registration(self, heart):
         """Second half of engine registration, called after our HeartMonitor
         has received a beat from the Engine's Heart."""
         try: 
             (eid,queue,reg,purge) = self.incoming_registrations.pop(heart)
         except KeyError:
             logger.error("registration::tried to finish nonexistant registration")
             return
         logger.info("registration::finished registering engine %i:%r"%(eid,queue))
         if purge is not None:
             purge.stop()
         control = queue
         self.ids.add(eid)
         self.keytable[eid] = queue
         self.engines[eid] = EngineConnector(eid, queue, reg, control, heart)
         self.by_ident[queue] = eid
         self.queues[eid] = list()
         self.tasks[eid] = list()
         self.completed[eid] = list()
         self.hearts[heart] = eid
         content = dict(id=eid, queue=self.engines[eid].queue)
         if self.notifier:
             self.session.send(self.notifier, "registration_notification", content=content)
     
     def _purge_stalled_registration(self, heart):
         if heart in self.incoming_registrations:
             eid = self.incoming_registrations.pop(heart)[0]
             logger.info("registration::purging stalled registration: %i"%eid)
         else:
             pass
             
     #-------------------------------------------------------------------------
     # Client Requests
     #-------------------------------------------------------------------------
     
     def check_load(self, client_id, msg):
         content = msg['content']
         try:
             targets = content['targets']
             targets = self._validate_targets(targets)
         except:
             content = wrap_exception()
             self.session.send(self.clientele, "controller_error", 
                     content=content, ident=client_id)
             return
         
         content = dict(status='ok')
         # loads = {}
         for t in targets:
             content[bytes(t)] = len(self.queues[t])+len(self.tasks[t])
         self.session.send(self.clientele, "load_reply", content=content, ident=client_id)
             
     
     def queue_status(self, client_id, msg):
         """Return the Queue status of one or more targets.
         if verbose: return the msg_ids
         else: return len of each type.
         keys: queue (pending MUX jobs)
             tasks (pending Task jobs)
             completed (finished jobs from both queues)"""
         content = msg['content']
         targets = content['targets']
         try:
             targets = self._validate_targets(targets)
         except:
             content = wrap_exception()
             self.session.send(self.clientele, "controller_error", 
                     content=content, ident=client_id)
             return
         verbose = content.get('verbose', False)
         content = dict(status='ok')
         for t in targets:
             queue = self.queues[t]
             completed = self.completed[t]
             tasks = self.tasks[t]
             if not verbose:
                 queue = len(queue)
                 completed = len(completed)
                 tasks = len(tasks)
             content[bytes(t)] = {'queue': queue, 'completed': completed , 'tasks': tasks}
             # pending
         self.session.send(self.clientele, "queue_reply", content=content, ident=client_id)
     
     def purge_results(self, client_id, msg):
         """Purge results from memory. This method is more valuable before we move
         to a DB based message storage mechanism."""
         content = msg['content']
         msg_ids = content.get('msg_ids', [])
         reply = dict(status='ok')
         if msg_ids == 'all':
             self.results = {}
         else:
             for msg_id in msg_ids:
                 if msg_id in self.results:
                     self.results.pop(msg_id)
                 else:
                     if msg_id in self.pending:
-                        reply = dict(status='error', reason="msg pending: %r"%msg_id)
+                        try:
+                            raise IndexError("msg pending: %r"%msg_id)
+                        except:
+                            reply = wrap_exception()
                     else:
-                        reply = dict(status='error', reason="No such msg: %r"%msg_id)
+                        try:
+                            raise IndexError("No such msg: %r"%msg_id)
+                        except:
+                            reply = wrap_exception()
                     break
             eids = content.get('engine_ids', [])
             for eid in eids:
                 if eid not in self.engines:
-                    reply = dict(status='error', reason="No such engine: %i"%eid)
+                    try:
+                        raise IndexError("No such engine: %i"%eid)
+                    except:
+                        reply = wrap_exception()
                     break
                 msg_ids = self.completed.pop(eid)
                 for msg_id in msg_ids:
                     self.results.pop(msg_id)
         
         self.sesison.send(self.clientele, 'purge_reply', content=reply, ident=client_id)
     
     def resubmit_task(self, client_id, msg, buffers):
         """Resubmit a task."""
         raise NotImplementedError
     
     def get_results(self, client_id, msg):
         """Get the result of 1 or more messages."""
         content = msg['content']
         msg_ids = set(content['msg_ids'])
         statusonly = content.get('status_only', False)
         pending = []
         completed = []
         content = dict(status='ok')
         content['pending'] = pending
         content['completed'] = completed
         for msg_id in msg_ids:
             if msg_id in self.pending:
                 pending.append(msg_id)
             elif msg_id in self.results:
                 completed.append(msg_id)
                 if not statusonly:
                     content[msg_id] = self.results[msg_id]['content']
             else:
-                content = dict(status='error')
-                content['reason'] = 'no such message: '+msg_id
+                try:
+                    raise KeyError('No such message: '+msg_id)
+                except:
+                    content = wrap_exception()
                 break
         self.session.send(self.clientele, "result_reply", content=content, 
                                             parent=msg, ident=client_id)
 
 
 #-------------------------------------------------------------------------
 # Entry Point
 #-------------------------------------------------------------------------
 
 def make_argument_parser():
     """Make an argument parser"""
     parser = make_base_argument_parser()
     
     parser.add_argument('--client', type=int, metavar='PORT', default=0,
                         help='set the XREP port for clients [default: random]')
     parser.add_argument('--notice', type=int, metavar='PORT', default=0,
                         help='set the PUB socket for registration notification [default: random]')
     parser.add_argument('--hb', type=str, metavar='PORTS',
                         help='set the 2 ports for heartbeats [default: random]')
     parser.add_argument('--ping', type=int, default=3000,
                         help='set the heartbeat period in ms [default: 3000]')
     parser.add_argument('--monitor', type=int, metavar='PORT', default=0,
                         help='set the SUB port for queue monitoring [default: random]')
     parser.add_argument('--mux', type=str, metavar='PORTS',
                         help='set the XREP ports for the MUX queue [default: random]')
     parser.add_argument('--task', type=str, metavar='PORTS',
                         help='set the XREP/XREQ ports for the task queue [default: random]')
     parser.add_argument('--control', type=str, metavar='PORTS',
                         help='set the XREP ports for the control queue [default: random]')
     parser.add_argument('--scheduler', type=str, default='pure',
                         choices = ['pure', 'lru', 'plainrandom', 'weighted', 'twobin','leastload'],
                         help='select the task scheduler  [default: pure ZMQ]')
     
     return parser
     
 def main():
     import time
     from multiprocessing import Process
     
     from zmq.eventloop.zmqstream import ZMQStream
     from zmq.devices import ProcessMonitoredQueue
     from zmq.log import handlers
     
     import streamsession as session
     import heartmonitor
     from scheduler import launch_scheduler
     
     parser = make_argument_parser()
     
     args = parser.parse_args()
     parse_url(args)
     
     iface="%s://%s"%(args.transport,args.ip)+':%i'
     
     random_ports = 0
     if args.hb:
         hb = split_ports(args.hb, 2)
     else:
         hb = select_random_ports(2)
     if args.mux:
         mux = split_ports(args.mux, 2)
     else:
         mux = None
         random_ports += 2
     if args.task:
         task = split_ports(args.task, 2)
     else:
         task = None
         random_ports += 2
     if args.control:
         control = split_ports(args.control, 2)
     else:
         control = None
         random_ports += 2
     
     ctx = zmq.Context()
     loop = ioloop.IOLoop.instance()
     
     # setup logging
     connect_logger(ctx, iface%args.logport, root="controller", loglevel=args.loglevel)
     
     # Registrar socket
     reg = ZMQStream(ctx.socket(zmq.XREP), loop)
     regport = bind_port(reg, args.ip, args.regport)
     
     ### Engine connections ###
     
     # heartbeat
     hpub = ctx.socket(zmq.PUB)
     bind_port(hpub, args.ip, hb[0])
     hrep = ctx.socket(zmq.XREP)
     bind_port(hrep, args.ip, hb[1])
     
     hmon = heartmonitor.HeartMonitor(loop, ZMQStream(hpub,loop), ZMQStream(hrep,loop),args.ping)
     hmon.start()
     
     ### Client connections ###
     # Clientele socket
     c = ZMQStream(ctx.socket(zmq.XREP), loop)
     cport = bind_port(c, args.ip, args.client)
     # Notifier socket
     n = ZMQStream(ctx.socket(zmq.PUB), loop)
     nport = bind_port(n, args.ip, args.notice)
     
     thesession = session.StreamSession(username=args.ident or "controller")
     
     ### build and launch the queues ###
     
     # monitor socket
     sub = ctx.socket(zmq.SUB)
     sub.setsockopt(zmq.SUBSCRIBE, "")
     monport = bind_port(sub, args.ip, args.monitor)
     sub = ZMQStream(sub, loop)
     
     ports = select_random_ports(random_ports)
     # Multiplexer Queue (in a Process)
     if not mux:
         mux = (ports.pop(),ports.pop())
     q = ProcessMonitoredQueue(zmq.XREP, zmq.XREP, zmq.PUB, 'in', 'out')
     q.bind_in(iface%mux[0])
     q.bind_out(iface%mux[1])
     q.connect_mon(iface%monport)
     q.daemon=True
     q.start()
     
     # Control Queue (in a Process)
     if not control:
         control = (ports.pop(),ports.pop())
     q = ProcessMonitoredQueue(zmq.XREP, zmq.XREP, zmq.PUB, 'incontrol', 'outcontrol')
     q.bind_in(iface%control[0])
     q.bind_out(iface%control[1])
     q.connect_mon(iface%monport)
     q.daemon=True
     q.start()
     
     # Task Queue (in a Process)
     if not task:
         task = (ports.pop(),ports.pop())
     if args.scheduler == 'pure':
         q = ProcessMonitoredQueue(zmq.XREP, zmq.XREQ, zmq.PUB, 'intask', 'outtask')
         q.bind_in(iface%task[0])
         q.bind_out(iface%task[1])
         q.connect_mon(iface%monport)
         q.daemon=True
         q.start()
     else:
         sargs = (iface%task[0],iface%task[1],iface%monport,iface%nport,args.scheduler)
         print (sargs)
         p = Process(target=launch_scheduler, args=sargs)
         p.daemon=True
         p.start()
     
     time.sleep(.25)
     
     # build connection dicts
     engine_addrs = {
         'control' : iface%control[1],
         'queue': iface%mux[1],
         'heartbeat': (iface%hb[0], iface%hb[1]),
         'task' : iface%task[1],
         'monitor' : iface%monport,
         }
     
     client_addrs = {
         'control' : iface%control[0],
         'query': iface%cport,
         'queue': iface%mux[0],
         'task' : iface%task[0],
         'notification': iface%nport
         }
     con = Controller(loop, thesession, sub, reg, hmon, c, n, None, engine_addrs, client_addrs)
     dc = ioloop.DelayedCallback(lambda : print("Controller started..."), 100, loop)
     loop.start()
 
 if __name__ == '__main__':
     main()

IPython/zmq/parallel/streamkernel.py

@@ -1,498 +1,505
 #!/usr/bin/env python
 """
 Kernel adapted from kernel.py to use ZMQ Streams
 """
 
 from __future__ import print_function
 import __builtin__
 import os
 import sys
 import time
 import traceback
 from signal import SIGTERM, SIGKILL
 from pprint import pprint
 
 from code import CommandCompiler
 
 import zmq
 from zmq.eventloop import ioloop, zmqstream
 
 from IPython.zmq.completer import KernelCompleter
 
 from streamsession import StreamSession, Message, extract_header, serialize_object,\
                 unpack_apply_message
 from dependency import UnmetDependency
 
 def printer(*args):
     pprint(args)
 
 class OutStream(object):
     """A file like object that publishes the stream to a 0MQ PUB socket."""
 
     def __init__(self, session, pub_socket, name, max_buffer=200):
         self.session = session
         self.pub_socket = pub_socket
         self.name = name
         self._buffer = []
         self._buffer_len = 0
         self.max_buffer = max_buffer
         self.parent_header = {}
 
     def set_parent(self, parent):
         self.parent_header = extract_header(parent)
 
     def close(self):
         self.pub_socket = None
 
     def flush(self):
         if self.pub_socket is None:
             raise ValueError(u'I/O operation on closed file')
         else:
             if self._buffer:
                 data = ''.join(self._buffer)
                 content = {u'name':self.name, u'data':data}
                 # msg = self.session.msg(u'stream', content=content,
                 #                        parent=self.parent_header)
                 msg = self.session.send(self.pub_socket, u'stream', content=content, parent=self.parent_header)
                 # print>>sys.__stdout__, Message(msg)
                 # self.pub_socket.send_json(msg)
                 self._buffer_len = 0
                 self._buffer = []
 
     def isattr(self):
         return False
 
     def next(self):
         raise IOError('Read not supported on a write only stream.')
 
     def read(self, size=None):
         raise IOError('Read not supported on a write only stream.')
 
     readline=read
 
     def write(self, s):
         if self.pub_socket is None:
             raise ValueError('I/O operation on closed file')
         else:
             self._buffer.append(s)
             self._buffer_len += len(s)
             self._maybe_send()
 
     def _maybe_send(self):
         if '\n' in self._buffer[-1]:
             self.flush()
         if self._buffer_len > self.max_buffer:
             self.flush()
 
     def writelines(self, sequence):
         if self.pub_socket is None:
             raise ValueError('I/O operation on closed file')
         else:
             for s in sequence:
                 self.write(s)
 
 
 class DisplayHook(object):
 
     def __init__(self, session, pub_socket):
         self.session = session
         self.pub_socket = pub_socket
         self.parent_header = {}
 
     def __call__(self, obj):
         if obj is None:
             return
 
         __builtin__._ = obj
         # msg = self.session.msg(u'pyout', {u'data':repr(obj)},
         #                        parent=self.parent_header)
         # self.pub_socket.send_json(msg)
         self.session.send(self.pub_socket, u'pyout', content={u'data':repr(obj)}, parent=self.parent_header)
 
     def set_parent(self, parent):
         self.parent_header = extract_header(parent)
 
 
 class RawInput(object):
 
     def __init__(self, session, socket):
         self.session = session
         self.socket = socket
 
     def __call__(self, prompt=None):
         msg = self.session.msg(u'raw_input')
         self.socket.send_json(msg)
         while True:
             try:
                 reply = self.socket.recv_json(zmq.NOBLOCK)
             except zmq.ZMQError as e:
                 if e.errno == zmq.EAGAIN:
                     pass
                 else:
                     raise
             else:
                 break
         return reply[u'content'][u'data']
 
 
 class Kernel(object):
 
     def __init__(self, session, control_stream, reply_stream, pub_stream, 
                                             task_stream=None, client=None):
         self.session = session
         self.control_stream = control_stream
         # self.control_socket = control_stream.socket
         self.reply_stream = reply_stream
+        self.identity = self.reply_stream.getsockopt(zmq.IDENTITY)
         self.task_stream = task_stream
         self.pub_stream = pub_stream
         self.client = client
         self.user_ns = {}
         self.history = []
         self.compiler = CommandCompiler()
         self.completer = KernelCompleter(self.user_ns)
         self.aborted = set()
         
         # Build dict of handlers for message types
         self.queue_handlers = {}
         self.control_handlers = {}
         for msg_type in ['execute_request', 'complete_request', 'apply_request']:
             self.queue_handlers[msg_type] = getattr(self, msg_type)
         
-        for msg_type in ['kill_request', 'abort_request']+self.queue_handlers.keys():
+        for msg_type in ['kill_request', 'abort_request', 'clear_request']+self.queue_handlers.keys():
             self.control_handlers[msg_type] = getattr(self, msg_type)
 
     #-------------------- control handlers -----------------------------
     def abort_queues(self):
         for stream in (self.task_stream, self.reply_stream):
             if stream:
                 self.abort_queue(stream)
     
     def abort_queue(self, stream):
         while True:
             try:
                 msg = self.session.recv(stream, zmq.NOBLOCK,content=True)
             except zmq.ZMQError as e:
                 if e.errno == zmq.EAGAIN:
                     break
                 else:
                     return
             else:
                 if msg is None:
                     return
                 else:
                     idents,msg = msg
                 
                 # assert self.reply_socketly_socket.rcvmore(), "Unexpected missing message part."
                 # msg = self.reply_socket.recv_json()
             print ("Aborting:", file=sys.__stdout__)
             print (Message(msg), file=sys.__stdout__)
             msg_type = msg['msg_type']
             reply_type = msg_type.split('_')[0] + '_reply'
             # reply_msg = self.session.msg(reply_type, {'status' : 'aborted'}, msg)
             # self.reply_socket.send(ident,zmq.SNDMORE)
             # self.reply_socket.send_json(reply_msg)
             reply_msg = self.session.send(stream, reply_type, 
                         content={'status' : 'aborted'}, parent=msg, ident=idents)[0]
             print(Message(reply_msg), file=sys.__stdout__)
             # We need to wait a bit for requests to come in. This can probably
             # be set shorter for true asynchronous clients.
             time.sleep(0.05)
     
     def abort_request(self, stream, ident, parent):
         """abort a specifig msg by id"""
         msg_ids = parent['content'].get('msg_ids', None)
         if isinstance(msg_ids, basestring):
             msg_ids = [msg_ids]
         if not msg_ids:
             self.abort_queues()
         for mid in msg_ids:
             self.aborted.add(str(mid))
         
         content = dict(status='ok')
         reply_msg = self.session.send(stream, 'abort_reply', content=content, 
                 parent=parent, ident=ident)[0]
         print(Message(reply_msg), file=sys.__stdout__)
     
     def kill_request(self, stream, idents, parent):
-        """kill ourselves.  This should really be handled in an external process"""
+        """kill ourself.  This should really be handled in an external process"""
         self.abort_queues()
         msg = self.session.send(stream, 'kill_reply', ident=idents, parent=parent, 
                 content = dict(status='ok'))
         # we can know that a message is done if we *don't* use streams, but 
         # use a socket directly with MessageTracker
         time.sleep(.5)
         os.kill(os.getpid(), SIGTERM)
         time.sleep(1)
         os.kill(os.getpid(), SIGKILL)
     
+    def clear_request(self, stream, idents, parent):
+        """Clear our namespace."""
+        self.user_ns = {}
+        msg = self.session.send(stream, 'clear_reply', ident=idents, parent=parent, 
+                content = dict(status='ok'))
+    
     def dispatch_control(self, msg):
         idents,msg = self.session.feed_identities(msg, copy=False)
         msg = self.session.unpack_message(msg, content=True, copy=False)
         
         header = msg['header']
         msg_id = header['msg_id']
         
         handler = self.control_handlers.get(msg['msg_type'], None)
         if handler is None:
             print ("UNKNOWN CONTROL MESSAGE TYPE:", msg, file=sys.__stderr__)
         else:
             handler(self.control_stream, idents, msg)
     
 
     #-------------------- queue helpers ------------------------------
     
     def check_dependencies(self, dependencies):
         if not dependencies:
             return True
         if len(dependencies) == 2 and dependencies[0] in 'any all'.split():
             anyorall = dependencies[0]
             dependencies = dependencies[1]
         else:
             anyorall = 'all'
         results = self.client.get_results(dependencies,status_only=True)
         if results['status'] != 'ok':
             return False
         
         if anyorall == 'any':
             if not results['completed']:
                 return False
         else:
             if results['pending']:
                 return False
         
         return True
     
     def check_aborted(self, msg_id):
         return msg_id in self.aborted
     
     #-------------------- queue handlers -----------------------------
     
     def execute_request(self, stream, ident, parent):
         try:
             code = parent[u'content'][u'code']
         except:
             print("Got bad msg: ", file=sys.__stderr__)
             print(Message(parent), file=sys.__stderr__)
             return
         # pyin_msg = self.session.msg(u'pyin',{u'code':code}, parent=parent)
         # self.pub_stream.send(pyin_msg)
         self.session.send(self.pub_stream, u'pyin', {u'code':code},parent=parent)
         try:
             comp_code = self.compiler(code, '<zmq-kernel>')
             # allow for not overriding displayhook
             if hasattr(sys.displayhook, 'set_parent'):
                 sys.displayhook.set_parent(parent)
             exec comp_code in self.user_ns, self.user_ns
         except:
             # result = u'error'
             etype, evalue, tb = sys.exc_info()
             tb = traceback.format_exception(etype, evalue, tb)
             exc_content = {
                 u'status' : u'error',
                 u'traceback' : tb,
                 u'etype' : unicode(etype),
                 u'evalue' : unicode(evalue)
             }
             # exc_msg = self.session.msg(u'pyerr', exc_content, parent)
             self.session.send(self.pub_stream, u'pyerr', exc_content, parent=parent)
             reply_content = exc_content
         else:
             reply_content = {'status' : 'ok'}
         # reply_msg = self.session.msg(u'execute_reply', reply_content, parent)
         # self.reply_socket.send(ident, zmq.SNDMORE)
         # self.reply_socket.send_json(reply_msg)
         reply_msg = self.session.send(stream, u'execute_reply', reply_content, parent=parent, ident=ident)
         print(Message(reply_msg), file=sys.__stdout__)
         if reply_msg['content']['status'] == u'error':
             self.abort_queues()
 
     def complete_request(self, stream, ident, parent):
         matches = {'matches' : self.complete(parent),
                    'status' : 'ok'}
         completion_msg = self.session.send(stream, 'complete_reply',
                                            matches, parent, ident)
         # print >> sys.__stdout__, completion_msg
 
     def complete(self, msg):
         return self.completer.complete(msg.content.line, msg.content.text)
     
     def apply_request(self, stream, ident, parent):
         print (parent)
         try:
             content = parent[u'content']
             bufs = parent[u'buffers']
             msg_id = parent['header']['msg_id']
             bound = content.get('bound', False)
         except:
             print("Got bad msg: ", file=sys.__stderr__)
             print(Message(parent), file=sys.__stderr__)
             return
         # pyin_msg = self.session.msg(u'pyin',{u'code':code}, parent=parent)
         # self.pub_stream.send(pyin_msg)
         # self.session.send(self.pub_stream, u'pyin', {u'code':code},parent=parent)
-        sub = {'dependencies_met' : True}
+        sub = {'dependencies_met' : True, 'engine' : self.identity}
         try:
             # allow for not overriding displayhook
             if hasattr(sys.displayhook, 'set_parent'):
                 sys.displayhook.set_parent(parent)
             # exec "f(*args,**kwargs)" in self.user_ns, self.user_ns
             if bound:
                 working = self.user_ns
                 suffix = str(msg_id).replace("-","")
                 prefix = "_"
                 
             else:
                 working = dict()
                 suffix = prefix = "_" # prevent keyword collisions with lambda
             f,args,kwargs = unpack_apply_message(bufs, working, copy=False)
             # if f.fun
             fname = prefix+f.func_name.strip('<>')+suffix
             argname = prefix+"args"+suffix
             kwargname = prefix+"kwargs"+suffix
             resultname = prefix+"result"+suffix
             
             ns = { fname : f, argname : args, kwargname : kwargs }
             # print ns
             working.update(ns)
             code = "%s=%s(*%s,**%s)"%(resultname, fname, argname, kwargname)
             exec code in working, working
             result = working.get(resultname)
             # clear the namespace
             if bound:
                 for key in ns.iterkeys():
                     self.user_ns.pop(key)
             else:
                 del working
             
             packed_result,buf = serialize_object(result)
             result_buf = [packed_result]+buf
         except:
             result = u'error'
             etype, evalue, tb = sys.exc_info()
             tb = traceback.format_exception(etype, evalue, tb)
             exc_content = {
                 u'status' : u'error',
                 u'traceback' : tb,
                 u'etype' : unicode(etype),
                 u'evalue' : unicode(evalue)
             }
             # exc_msg = self.session.msg(u'pyerr', exc_content, parent)
             self.session.send(self.pub_stream, u'pyerr', exc_content, parent=parent)
             reply_content = exc_content
             result_buf = []
             
             if etype is UnmetDependency:
-                sub = {'dependencies_met' : False}
+                sub['dependencies_met'] = False
         else:
             reply_content = {'status' : 'ok'}
         # reply_msg = self.session.msg(u'execute_reply', reply_content, parent)
         # self.reply_socket.send(ident, zmq.SNDMORE)
         # self.reply_socket.send_json(reply_msg)
         reply_msg = self.session.send(stream, u'apply_reply', reply_content, 
                     parent=parent, ident=ident,buffers=result_buf, subheader=sub)
         print(Message(reply_msg), file=sys.__stdout__)
         # if reply_msg['content']['status'] == u'error':
         #     self.abort_queues()
     
     def dispatch_queue(self, stream, msg):
         self.control_stream.flush()
         idents,msg = self.session.feed_identities(msg, copy=False)
         msg = self.session.unpack_message(msg, content=True, copy=False)
         
         header = msg['header']
         msg_id = header['msg_id']
         if self.check_aborted(msg_id):
             self.aborted.remove(msg_id)
             # is it safe to assume a msg_id will not be resubmitted?
             reply_type = msg['msg_type'].split('_')[0] + '_reply'
             reply_msg = self.session.send(stream, reply_type, 
                         content={'status' : 'aborted'}, parent=msg, ident=idents)
             return
         handler = self.queue_handlers.get(msg['msg_type'], None)
         if handler is None:
             print ("UNKNOWN MESSAGE TYPE:", msg, file=sys.__stderr__)
         else:
             handler(stream, idents, msg)
     
     def start(self):
         #### stream mode:
         if self.control_stream:
             self.control_stream.on_recv(self.dispatch_control, copy=False)
             self.control_stream.on_err(printer)
         if self.reply_stream:
             self.reply_stream.on_recv(lambda msg: 
                     self.dispatch_queue(self.reply_stream, msg), copy=False)
             self.reply_stream.on_err(printer)
         if self.task_stream:
             self.task_stream.on_recv(lambda msg: 
                     self.dispatch_queue(self.task_stream, msg), copy=False)
             self.task_stream.on_err(printer)
         
         #### while True mode:
         # while True:
         #     idle = True
         #     try:
         #         msg = self.reply_stream.socket.recv_multipart(
         #                     zmq.NOBLOCK, copy=False)
         #     except zmq.ZMQError, e:
         #         if e.errno != zmq.EAGAIN:
         #             raise e
         #     else:
         #         idle=False
         #         self.dispatch_queue(self.reply_stream, msg)
         #             
         #     if not self.task_stream.empty():
         #         idle=False
         #         msg = self.task_stream.recv_multipart()
         #         self.dispatch_queue(self.task_stream, msg)
         #     if idle:
         #         # don't busywait
         #         time.sleep(1e-3)
 
 
 def main():
     raise Exception("Don't run me anymore")
     loop = ioloop.IOLoop.instance()
     c = zmq.Context()
 
     ip = '127.0.0.1'
     port_base = 5575
     connection = ('tcp://%s' % ip) + ':%i'
     rep_conn = connection % port_base
     pub_conn = connection % (port_base+1)
 
     print("Starting the kernel...", file=sys.__stdout__)
     # print >>sys.__stdout__, "XREQ Channel:", rep_conn
     # print >>sys.__stdout__, "PUB Channel:", pub_conn
 
     session = StreamSession(username=u'kernel')
 
     reply_socket = c.socket(zmq.XREQ)
     reply_socket.connect(rep_conn)
 
     pub_socket = c.socket(zmq.PUB)
     pub_socket.connect(pub_conn)
 
     stdout = OutStream(session, pub_socket, u'stdout')
     stderr = OutStream(session, pub_socket, u'stderr')
     sys.stdout = stdout
     sys.stderr = stderr
 
     display_hook = DisplayHook(session, pub_socket)
     sys.displayhook = display_hook
     reply_stream = zmqstream.ZMQStream(reply_socket,loop)
     pub_stream = zmqstream.ZMQStream(pub_socket,loop)
     kernel = Kernel(session, reply_stream, pub_stream)
 
     # For debugging convenience, put sleep and a string in the namespace, so we
     # have them every time we start.
     kernel.user_ns['sleep'] = time.sleep
     kernel.user_ns['s'] = 'Test string'
     
     print ("Use Ctrl-\\ (NOT Ctrl-C!) to terminate.", file=sys.__stdout__)
     kernel.start()
     loop.start()
 
 
 if __name__ == '__main__':
     main()