minor controller logging adjustments

* add debug message when heartbeats go out
* don't call unregister_engine more than once for a failed heart
* adjust logging calls, to avoid formatting unused messages

IPython/parallel/controller/heartmonitor.py

@@ -1,174 +1,174
 #!/usr/bin/env python
 """
 A multi-heart Heartbeat system using PUB and XREP sockets. pings are sent out on the PUB,
 and hearts are tracked based on their XREQ identities.
 
 Authors:
 
 * Min RK
 """
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2010-2011  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.
 #-----------------------------------------------------------------------------
 
 from __future__ import print_function
 import time
 import uuid
 
 import zmq
 from zmq.devices import ThreadDevice
 from zmq.eventloop import ioloop, zmqstream
 
 from IPython.config.configurable import LoggingConfigurable
 from IPython.utils.traitlets import Set, Instance, CFloat
 
 from IPython.parallel.util import asbytes
 
 class Heart(object):
     """A basic heart object for responding to a HeartMonitor.
     This is a simple wrapper with defaults for the most common
     Device model for responding to heartbeats.
 
     It simply builds a threadsafe zmq.FORWARDER Device, defaulting to using
     SUB/XREQ for in/out.
 
     You can specify the XREQ's IDENTITY via the optional heart_id argument."""
     device=None
     id=None
     def __init__(self, in_addr, out_addr, in_type=zmq.SUB, out_type=zmq.DEALER, heart_id=None):
         self.device = ThreadDevice(zmq.FORWARDER, in_type, out_type)
         self.device.daemon=True
         self.device.connect_in(in_addr)
         self.device.connect_out(out_addr)
         if in_type == zmq.SUB:
             self.device.setsockopt_in(zmq.SUBSCRIBE, b"")
         if heart_id is None:
             heart_id = uuid.uuid4().bytes
         self.device.setsockopt_out(zmq.IDENTITY, heart_id)
         self.id = heart_id
 
     def start(self):
         return self.device.start()
 
+
 class HeartMonitor(LoggingConfigurable):
     """A basic HeartMonitor class
     pingstream: a PUB stream
     pongstream: an XREP stream
     period: the period of the heartbeat in milliseconds"""
 
     period=CFloat(1000, config=True,
         help='The frequency at which the Hub pings the engines for heartbeats '
         '(in ms)',
     )
 
     pingstream=Instance('zmq.eventloop.zmqstream.ZMQStream')
     pongstream=Instance('zmq.eventloop.zmqstream.ZMQStream')
     loop = Instance('zmq.eventloop.ioloop.IOLoop')
     def _loop_default(self):
         return ioloop.IOLoop.instance()
 
     # not settable:
     hearts=Set()
     responses=Set()
     on_probation=Set()
     last_ping=CFloat(0)
     _new_handlers = Set()
     _failure_handlers = Set()
     lifetime = CFloat(0)
     tic = CFloat(0)
 
     def __init__(self, **kwargs):
         super(HeartMonitor, self).__init__(**kwargs)
 
         self.pongstream.on_recv(self.handle_pong)
 
     def start(self):
         self.tic = time.time()
         self.caller = ioloop.PeriodicCallback(self.beat, self.period, self.loop)
         self.caller.start()
 
     def add_new_heart_handler(self, handler):
         """add a new handler for new hearts"""
-        self.log.debug("heartbeat::new_heart_handler: %s"%handler)
+        self.log.debug("heartbeat::new_heart_handler: %s", handler)
         self._new_handlers.add(handler)
 
     def add_heart_failure_handler(self, handler):
         """add a new handler for heart failure"""
-        self.log.debug("heartbeat::new heart failure handler: %s"%handler)
+        self.log.debug("heartbeat::new heart failure handler: %s", handler)
         self._failure_handlers.add(handler)
 
     def beat(self):
         self.pongstream.flush()
         self.last_ping = self.lifetime
 
         toc = time.time()
         self.lifetime += toc-self.tic
         self.tic = toc
-        # self.log.debug("heartbeat::%s"%self.lifetime)
+        self.log.debug("heartbeat::sending %s", self.lifetime)
         goodhearts = self.hearts.intersection(self.responses)
         missed_beats = self.hearts.difference(goodhearts)
         heartfailures = self.on_probation.intersection(missed_beats)
         newhearts = self.responses.difference(goodhearts)
         map(self.handle_new_heart, newhearts)
         map(self.handle_heart_failure, heartfailures)
         self.on_probation = missed_beats.intersection(self.hearts)
         self.responses = set()
         # print self.on_probation, self.hearts
-        # self.log.debug("heartbeat::beat %.3f, %i beating hearts"%(self.lifetime, len(self.hearts)))
+        # self.log.debug("heartbeat::beat %.3f, %i beating hearts", self.lifetime, len(self.hearts))
         self.pingstream.send(asbytes(str(self.lifetime)))
 
     def handle_new_heart(self, heart):
         if self._new_handlers:
             for handler in self._new_handlers:
                 handler(heart)
         else:
-            self.log.info("heartbeat::yay, got new heart %s!"%heart)
+            self.log.info("heartbeat::yay, got new heart %s!", heart)
         self.hearts.add(heart)
 
     def handle_heart_failure(self, heart):
         if self._failure_handlers:
             for handler in self._failure_handlers:
                 try:
                     handler(heart)
                 except Exception as e:
-                    self.log.error("heartbeat::Bad Handler! %s"%handler, exc_info=True)
+                    self.log.error("heartbeat::Bad Handler! %s", handler, exc_info=True)
                     pass
         else:
-            self.log.info("heartbeat::Heart %s failed :("%heart)
+            self.log.info("heartbeat::Heart %s failed :(", heart)
         self.hearts.remove(heart)
 
 
     def handle_pong(self, msg):
         "a heart just beat"
         current = asbytes(str(self.lifetime))
         last = asbytes(str(self.last_ping))
         if msg[1] == current:
             delta = time.time()-self.tic
             # self.log.debug("heartbeat::heart %r took %.2f ms to respond"%(msg[0], 1000*delta))
             self.responses.add(msg[0])
         elif msg[1] == last:
             delta = time.time()-self.tic + (self.lifetime-self.last_ping)
-            self.log.warn("heartbeat::heart %r missed a beat, and took %.2f ms to respond"%(msg[0], 1000*delta))
+            self.log.warn("heartbeat::heart %r missed a beat, and took %.2f ms to respond", msg[0], 1000*delta)
             self.responses.add(msg[0])
         else:
-            self.log.warn("heartbeat::got bad heartbeat (possibly old?): %s (current=%.3f)"%
-            (msg[1],self.lifetime))
+            self.log.warn("heartbeat::got bad heartbeat (possibly old?): %s (current=%.3f)", msg[1], self.lifetime)
 
 
 if __name__ == '__main__':
     loop = ioloop.IOLoop.instance()
     context = zmq.Context()
     pub = context.socket(zmq.PUB)
     pub.bind('tcp://127.0.0.1:5555')
     xrep = context.socket(zmq.ROUTER)
     xrep.bind('tcp://127.0.0.1:5556')
 
     outstream = zmqstream.ZMQStream(pub, loop)
     instream = zmqstream.ZMQStream(xrep, loop)
 
     hb = HeartMonitor(loop, outstream, instream)
 
     loop.start()

IPython/parallel/controller/hub.py

@@ -1,1290 +1,1293
 """The IPython Controller Hub with 0MQ
 This is the master object that handles connections from engines and clients,
 and monitors traffic through the various queues.
 
 Authors:
 
 * Min RK
 """
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2010-2011  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
 
 import sys
 import time
 from datetime import datetime
 
 import zmq
 from zmq.eventloop import ioloop
 from zmq.eventloop.zmqstream import ZMQStream
 
 # internal:
 from IPython.utils.importstring import import_item
 from IPython.utils.traitlets import (
         HasTraits, Instance, Integer, Unicode, Dict, Set, Tuple, CBytes, DottedObjectName
         )
 
 from IPython.parallel import error, util
 from IPython.parallel.factory import RegistrationFactory
 
 from IPython.zmq.session import SessionFactory
 
 from .heartmonitor import HeartMonitor
 
 #-----------------------------------------------------------------------------
 # Code
 #-----------------------------------------------------------------------------
 
 def _passer(*args, **kwargs):
     return
 
 def _printer(*args, **kwargs):
     print (args)
     print (kwargs)
 
 def empty_record():
     """Return an empty dict with all record keys."""
     return {
         'msg_id' : None,
         'header' : None,
         'content': None,
         'buffers': None,
         'submitted': None,
         'client_uuid' : None,
         'engine_uuid' : None,
         'started': None,
         'completed': None,
         'resubmitted': None,
         'result_header' : None,
         'result_content' : None,
         'result_buffers' : None,
         'queue' : None,
         'pyin' : None,
         'pyout': None,
         'pyerr': None,
         'stdout': '',
         'stderr': '',
     }
 
 def init_record(msg):
     """Initialize a TaskRecord based on a request."""
     header = msg['header']
     return {
         'msg_id' : header['msg_id'],
         'header' : header,
         'content': msg['content'],
         'buffers': msg['buffers'],
         'submitted': header['date'],
         'client_uuid' : None,
         'engine_uuid' : None,
         'started': None,
         'completed': None,
         'resubmitted': None,
         'result_header' : None,
         'result_content' : None,
         'result_buffers' : None,
         'queue' : None,
         'pyin' : None,
         'pyout': None,
         'pyerr': None,
         'stdout': '',
         'stderr': '',
     }
 
 
 class EngineConnector(HasTraits):
     """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=Integer(0)
     queue=CBytes()
     control=CBytes()
     registration=CBytes()
     heartbeat=CBytes()
     pending=Set()
 
 class HubFactory(RegistrationFactory):
     """The Configurable for setting up a Hub."""
 
     # port-pairs for monitoredqueues:
     hb = Tuple(Integer,Integer,config=True,
         help="""XREQ/SUB Port pair for Engine heartbeats""")
     def _hb_default(self):
         return tuple(util.select_random_ports(2))
 
     mux = Tuple(Integer,Integer,config=True,
         help="""Engine/Client Port pair for MUX queue""")
 
     def _mux_default(self):
         return tuple(util.select_random_ports(2))
 
     task = Tuple(Integer,Integer,config=True,
         help="""Engine/Client Port pair for Task queue""")
     def _task_default(self):
         return tuple(util.select_random_ports(2))
 
     control = Tuple(Integer,Integer,config=True,
         help="""Engine/Client Port pair for Control queue""")
 
     def _control_default(self):
         return tuple(util.select_random_ports(2))
 
     iopub = Tuple(Integer,Integer,config=True,
         help="""Engine/Client Port pair for IOPub relay""")
 
     def _iopub_default(self):
         return tuple(util.select_random_ports(2))
 
     # single ports:
     mon_port = Integer(config=True,
         help="""Monitor (SUB) port for queue traffic""")
 
     def _mon_port_default(self):
         return util.select_random_ports(1)[0]
 
     notifier_port = Integer(config=True,
         help="""PUB port for sending engine status notifications""")
 
     def _notifier_port_default(self):
         return util.select_random_ports(1)[0]
 
     engine_ip = Unicode('127.0.0.1', config=True,
         help="IP on which to listen for engine connections. [default: loopback]")
     engine_transport = Unicode('tcp', config=True,
         help="0MQ transport for engine connections. [default: tcp]")
 
     client_ip = Unicode('127.0.0.1', config=True,
         help="IP on which to listen for client connections. [default: loopback]")
     client_transport = Unicode('tcp', config=True,
         help="0MQ transport for client connections. [default : tcp]")
 
     monitor_ip = Unicode('127.0.0.1', config=True,
         help="IP on which to listen for monitor messages. [default: loopback]")
     monitor_transport = Unicode('tcp', config=True,
         help="0MQ transport for monitor messages. [default : tcp]")
 
     monitor_url = Unicode('')
 
     db_class = DottedObjectName('IPython.parallel.controller.dictdb.DictDB',
         config=True, help="""The class to use for the DB backend""")
 
     # not configurable
     db = Instance('IPython.parallel.controller.dictdb.BaseDB')
     heartmonitor = Instance('IPython.parallel.controller.heartmonitor.HeartMonitor')
 
     def _ip_changed(self, name, old, new):
         self.engine_ip = new
         self.client_ip = new
         self.monitor_ip = new
         self._update_monitor_url()
 
     def _update_monitor_url(self):
         self.monitor_url = "%s://%s:%i"%(self.monitor_transport, self.monitor_ip, self.mon_port)
 
     def _transport_changed(self, name, old, new):
         self.engine_transport = new
         self.client_transport = new
         self.monitor_transport = new
         self._update_monitor_url()
 
     def __init__(self, **kwargs):
         super(HubFactory, self).__init__(**kwargs)
         self._update_monitor_url()
 
 
     def construct(self):
         self.init_hub()
 
     def start(self):
         self.heartmonitor.start()
         self.log.info("Heartmonitor started")
 
     def init_hub(self):
         """construct"""
         client_iface = "%s://%s:"%(self.client_transport, self.client_ip) + "%i"
         engine_iface = "%s://%s:"%(self.engine_transport, self.engine_ip) + "%i"
 
         ctx = self.context
         loop = self.loop
 
         # Registrar socket
         q = ZMQStream(ctx.socket(zmq.ROUTER), loop)
         q.bind(client_iface % self.regport)
-        self.log.info("Hub listening on %s for registration."%(client_iface%self.regport))
+        self.log.info("Hub listening on %s for registration.", client_iface % self.regport)
         if self.client_ip != self.engine_ip:
             q.bind(engine_iface % self.regport)
-            self.log.info("Hub listening on %s for registration."%(engine_iface%self.regport))
+            self.log.info("Hub listening on %s for registration.", engine_iface % self.regport)
 
         ### Engine connections ###
 
         # heartbeat
         hpub = ctx.socket(zmq.PUB)
         hpub.bind(engine_iface % self.hb[0])
         hrep = ctx.socket(zmq.ROUTER)
         hrep.bind(engine_iface % self.hb[1])
         self.heartmonitor = HeartMonitor(loop=loop, config=self.config, log=self.log,
                                 pingstream=ZMQStream(hpub,loop),
                                 pongstream=ZMQStream(hrep,loop)
                             )
 
         ### Client connections ###
         # Notifier socket
         n = ZMQStream(ctx.socket(zmq.PUB), loop)
         n.bind(client_iface%self.notifier_port)
 
         ### build and launch the queues ###
 
         # monitor socket
         sub = ctx.socket(zmq.SUB)
         sub.setsockopt(zmq.SUBSCRIBE, b"")
         sub.bind(self.monitor_url)
         sub.bind('inproc://monitor')
         sub = ZMQStream(sub, loop)
 
         # connect the db
         self.log.info('Hub using DB backend: %r'%(self.db_class.split()[-1]))
         # cdir = self.config.Global.cluster_dir
         self.db = import_item(str(self.db_class))(session=self.session.session,
                                             config=self.config, log=self.log)
         time.sleep(.25)
         try:
             scheme = self.config.TaskScheduler.scheme_name
         except AttributeError:
             from .scheduler import TaskScheduler
             scheme = TaskScheduler.scheme_name.get_default_value()
         # build connection dicts
         self.engine_info = {
             'control' : engine_iface%self.control[1],
             'mux': engine_iface%self.mux[1],
             'heartbeat': (engine_iface%self.hb[0], engine_iface%self.hb[1]),
             'task' : engine_iface%self.task[1],
             'iopub' : engine_iface%self.iopub[1],
             # 'monitor' : engine_iface%self.mon_port,
             }
 
         self.client_info = {
             'control' : client_iface%self.control[0],
             'mux': client_iface%self.mux[0],
             'task' : (scheme, client_iface%self.task[0]),
             'iopub' : client_iface%self.iopub[0],
             'notification': client_iface%self.notifier_port
             }
-        self.log.debug("Hub engine addrs: %s"%self.engine_info)
-        self.log.debug("Hub client addrs: %s"%self.client_info)
+        self.log.debug("Hub engine addrs: %s", self.engine_info)
+        self.log.debug("Hub client addrs: %s", self.client_info)
 
         # resubmit stream
         r = ZMQStream(ctx.socket(zmq.DEALER), loop)
         url = util.disambiguate_url(self.client_info['task'][-1])
         r.setsockopt(zmq.IDENTITY, self.session.bsession)
         r.connect(url)
 
         self.hub = Hub(loop=loop, session=self.session, monitor=sub, heartmonitor=self.heartmonitor,
                 query=q, notifier=n, resubmit=r, db=self.db,
                 engine_info=self.engine_info, client_info=self.client_info,
                 log=self.log)
 
 
 class Hub(SessionFactory):
     """The IPython Controller Hub with 0MQ connections
 
     Parameters
     ==========
     loop: zmq IOLoop instance
     session: Session object
     <removed> context: zmq context for creating new connections (?)
     queue: ZMQStream for monitoring the command queue (SUB)
     query: ZMQStream for engine registration and client queries requests (XREP)
     heartbeat: HeartMonitor object checking the pulse of the engines
     notifier: ZMQStream for broadcasting engine registration changes (PUB)
     db: connection to db for out of memory logging of commands
                 NotImplemented
     engine_info: dict of zmq connection information for engines to connect
                 to the queues.
     client_info: dict of zmq connection information for engines to connect
                 to the queues.
     """
     # internal data structures:
     ids=Set() # engine IDs
     keytable=Dict()
     by_ident=Dict()
     engines=Dict()
     clients=Dict()
     hearts=Dict()
     pending=Set()
     queues=Dict()  # pending msg_ids keyed by engine_id
     tasks=Dict() # pending msg_ids submitted as tasks, keyed by client_id
     completed=Dict() # completed msg_ids keyed by engine_id
     all_completed=Set() # completed msg_ids keyed by engine_id
     dead_engines=Set() # completed msg_ids keyed by engine_id
     unassigned=Set() # set of task msg_ds not yet assigned a destination
     incoming_registrations=Dict()
     registration_timeout=Integer()
     _idcounter=Integer(0)
 
     # objects from constructor:
     query=Instance(ZMQStream)
     monitor=Instance(ZMQStream)
     notifier=Instance(ZMQStream)
     resubmit=Instance(ZMQStream)
     heartmonitor=Instance(HeartMonitor)
     db=Instance(object)
     client_info=Dict()
     engine_info=Dict()
 
 
     def __init__(self, **kwargs):
         """
         # universal:
         loop: IOLoop for creating future connections
         session: streamsession for sending serialized data
         # engine:
         queue: ZMQStream for monitoring queue messages
         query: ZMQStream for engine+client registration and client requests
         heartbeat: HeartMonitor object for tracking engines
         # extra:
         db: ZMQStream for db connection (NotImplemented)
         engine_info: zmq address/protocol dict for engine connections
         client_info: zmq address/protocol dict for client connections
         """
 
         super(Hub, self).__init__(**kwargs)
         self.registration_timeout = max(5000, 2*self.heartmonitor.period)
 
         # validate connection dicts:
         for k,v in self.client_info.iteritems():
             if k == 'task':
                 util.validate_url_container(v[1])
             else:
                 util.validate_url_container(v)
         # util.validate_url_container(self.client_info)
         util.validate_url_container(self.engine_info)
 
         # register our callbacks
         self.query.on_recv(self.dispatch_query)
         self.monitor.on_recv(self.dispatch_monitor_traffic)
 
         self.heartmonitor.add_heart_failure_handler(self.handle_heart_failure)
         self.heartmonitor.add_new_heart_handler(self.handle_new_heart)
 
         self.monitor_handlers = {b'in' : self.save_queue_request,
                                 b'out': self.save_queue_result,
                                 b'intask': self.save_task_request,
                                 b'outtask': self.save_task_result,
                                 b'tracktask': self.save_task_destination,
                                 b'incontrol': _passer,
                                 b'outcontrol': _passer,
                                 b'iopub': self.save_iopub_message,
         }
 
         self.query_handlers = {'queue_request': self.queue_status,
                                 'result_request': self.get_results,
                                 'history_request': self.get_history,
                                 'db_request': self.db_query,
                                 'purge_request': self.purge_results,
                                 'load_request': self.check_load,
                                 'resubmit_request': self.resubmit_task,
                                 'shutdown_request': self.shutdown_request,
                                 'registration_request' : self.register_engine,
                                 'unregistration_request' : self.unregister_engine,
                                 'connection_request': self.connection_request,
         }
 
         # ignore resubmit replies
         self.resubmit.on_recv(lambda msg: None, copy=False)
 
         self.log.info("hub::created hub")
 
     @property
     def _next_id(self):
         """gemerate a new ID.
 
         No longer reuse old ids, just count from 0."""
         newid = self._idcounter
         self._idcounter += 1
         return newid
         # 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
 
     #-----------------------------------------------------------------------------
     # dispatch methods (1 per stream)
     #-----------------------------------------------------------------------------
 
 
     def dispatch_monitor_traffic(self, msg):
         """all ME and Task queue messages come through here, as well as
         IOPub traffic."""
-        self.log.debug("monitor traffic: %r"%msg[:2])
+        self.log.debug("monitor traffic: %r", msg[:2])
         switch = msg[0]
         try:
             idents, msg = self.session.feed_identities(msg[1:])
         except ValueError:
             idents=[]
         if not idents:
-            self.log.error("Bad Monitor Message: %r"%msg)
+            self.log.error("Bad Monitor Message: %r", msg)
             return
         handler = self.monitor_handlers.get(switch, None)
         if handler is not None:
             handler(idents, msg)
         else:
-            self.log.error("Invalid monitor topic: %r"%switch)
+            self.log.error("Invalid monitor topic: %r", switch)
 
 
     def dispatch_query(self, msg):
         """Route registration requests and queries from clients."""
         try:
             idents, msg = self.session.feed_identities(msg)
         except ValueError:
             idents = []
         if not idents:
-            self.log.error("Bad Query Message: %r"%msg)
+            self.log.error("Bad Query Message: %r", msg)
             return
         client_id = idents[0]
         try:
             msg = self.session.unserialize(msg, content=True)
         except Exception:
             content = error.wrap_exception()
-            self.log.error("Bad Query Message: %r"%msg, exc_info=True)
+            self.log.error("Bad Query Message: %r", msg, exc_info=True)
             self.session.send(self.query, "hub_error", ident=client_id,
                     content=content)
             return
         # print client_id, header, parent, content
         #switch on message type:
         msg_type = msg['header']['msg_type']
-        self.log.info("client::client %r requested %r"%(client_id, msg_type))
+        self.log.info("client::client %r requested %r", client_id, msg_type)
         handler = self.query_handlers.get(msg_type, None)
         try:
             assert handler is not None, "Bad Message Type: %r"%msg_type
         except:
             content = error.wrap_exception()
-            self.log.error("Bad Message Type: %r"%msg_type, exc_info=True)
+            self.log.error("Bad Message Type: %r", msg_type, exc_info=True)
             self.session.send(self.query, "hub_error", ident=client_id,
                     content=content)
             return
 
         else:
             handler(idents, 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."""
-        self.log.debug("heartbeat::handle_new_heart(%r)"%heart)
+        self.log.debug("heartbeat::handle_new_heart(%r)", heart)
         if heart not in self.incoming_registrations:
-            self.log.info("heartbeat::ignoring new heart: %r"%heart)
+            self.log.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"""
-        self.log.debug("heartbeat::handle_heart_failure(%r)"%heart)
+        self.log.debug("heartbeat::handle_heart_failure(%r)", heart)
         eid = self.hearts.get(heart, None)
         queue = self.engines[eid].queue
-        if eid is None:
-            self.log.info("heartbeat::ignoring heart failure %r"%heart)
+        if eid is None or self.keytable[eid] in self.dead_engines:
+            self.log.info("heartbeat::ignoring heart failure %r (not an engine or already dead)", heart)
         else:
             self.unregister_engine(heart, dict(content=dict(id=eid, queue=queue)))
 
     #----------------------- MUX Queue Traffic ------------------------------
 
     def save_queue_request(self, idents, msg):
         if len(idents) < 2:
-            self.log.error("invalid identity prefix: %r"%idents)
+            self.log.error("invalid identity prefix: %r", idents)
             return
         queue_id, client_id = idents[:2]
         try:
             msg = self.session.unserialize(msg)
         except Exception:
-            self.log.error("queue::client %r sent invalid message to %r: %r"%(client_id, queue_id, msg), exc_info=True)
+            self.log.error("queue::client %r sent invalid message to %r: %r", client_id, queue_id, msg, exc_info=True)
             return
 
         eid = self.by_ident.get(queue_id, None)
         if eid is None:
-            self.log.error("queue::target %r not registered"%queue_id)
-            self.log.debug("queue::    valid are: %r"%(self.by_ident.keys()))
+            self.log.error("queue::target %r not registered", queue_id)
+            self.log.debug("queue::    valid are: %r", self.by_ident.keys())
             return
         record = init_record(msg)
         msg_id = record['msg_id']
+        self.log.info("queue::client %r submitted request %r to %s", client_id, msg_id, eid)
         # Unicode in records
         record['engine_uuid'] = queue_id.decode('ascii')
         record['client_uuid'] = client_id.decode('ascii')
         record['queue'] = 'mux'
 
         try:
             # it's posible iopub arrived first:
             existing = self.db.get_record(msg_id)
             for key,evalue in existing.iteritems():
                 rvalue = record.get(key, None)
                 if evalue and rvalue and evalue != rvalue:
-                    self.log.warn("conflicting initial state for record: %r:%r <%r> %r"%(msg_id, rvalue, key, evalue))
+                    self.log.warn("conflicting initial state for record: %r:%r <%r> %r", msg_id, rvalue, key, evalue)
                 elif evalue and not rvalue:
                     record[key] = evalue
             try:
                 self.db.update_record(msg_id, record)
             except Exception:
-                self.log.error("DB Error updating record %r"%msg_id, exc_info=True)
+                self.log.error("DB Error updating record %r", msg_id, exc_info=True)
         except KeyError:
             try:
                 self.db.add_record(msg_id, record)
             except Exception:
-                self.log.error("DB Error adding record %r"%msg_id, exc_info=True)
+                self.log.error("DB Error adding record %r", msg_id, exc_info=True)
 
 
         self.pending.add(msg_id)
         self.queues[eid].append(msg_id)
 
     def save_queue_result(self, idents, msg):
         if len(idents) < 2:
-            self.log.error("invalid identity prefix: %r"%idents)
+            self.log.error("invalid identity prefix: %r", idents)
             return
 
         client_id, queue_id = idents[:2]
         try:
             msg = self.session.unserialize(msg)
         except Exception:
-            self.log.error("queue::engine %r sent invalid message to %r: %r"%(
-                    queue_id,client_id, msg), exc_info=True)
+            self.log.error("queue::engine %r sent invalid message to %r: %r",
+                    queue_id, client_id, msg, exc_info=True)
             return
 
         eid = self.by_ident.get(queue_id, None)
         if eid is None:
-            self.log.error("queue::unknown engine %r is sending a reply: "%queue_id)
+            self.log.error("queue::unknown engine %r is sending a reply: ", queue_id)
             return
 
         parent = msg['parent_header']
         if not parent:
             return
         msg_id = parent['msg_id']
         if msg_id in self.pending:
             self.pending.remove(msg_id)
             self.all_completed.add(msg_id)
             self.queues[eid].remove(msg_id)
             self.completed[eid].append(msg_id)
+            self.log.info("queue::request %r completed on %s", msg_id, eid)
         elif msg_id not in self.all_completed:
             # it could be a result from a dead engine that died before delivering the
             # result
-            self.log.warn("queue:: unknown msg finished %r"%msg_id)
+            self.log.warn("queue:: unknown msg finished %r", msg_id)
             return
         # update record anyway, because the unregistration could have been premature
         rheader = msg['header']
         completed = rheader['date']
         started = rheader.get('started', None)
         result = {
             'result_header' : rheader,
             'result_content': msg['content'],
             'started' : started,
             'completed' : completed
         }
 
         result['result_buffers'] = msg['buffers']
         try:
             self.db.update_record(msg_id, result)
         except Exception:
-            self.log.error("DB Error updating record %r"%msg_id, exc_info=True)
+            self.log.error("DB Error updating record %r", msg_id, exc_info=True)
 
 
     #--------------------- Task Queue Traffic ------------------------------
 
     def save_task_request(self, idents, msg):
         """Save the submission of a task."""
         client_id = idents[0]
 
         try:
             msg = self.session.unserialize(msg)
         except Exception:
-            self.log.error("task::client %r sent invalid task message: %r"%(
-                    client_id, msg), exc_info=True)
+            self.log.error("task::client %r sent invalid task message: %r",
+                    client_id, msg, exc_info=True)
             return
         record = init_record(msg)
 
         record['client_uuid'] = client_id.decode('ascii')
         record['queue'] = 'task'
         header = msg['header']
         msg_id = header['msg_id']
         self.pending.add(msg_id)
         self.unassigned.add(msg_id)
         try:
             # it's posible iopub arrived first:
             existing = self.db.get_record(msg_id)
             if existing['resubmitted']:
                 for key in ('submitted', 'client_uuid', 'buffers'):
                     # don't clobber these keys on resubmit
                     # submitted and client_uuid should be different
                     # and buffers might be big, and shouldn't have changed
                     record.pop(key)
                     # still check content,header which should not change
                     # but are not expensive to compare as buffers
 
             for key,evalue in existing.iteritems():
                 if key.endswith('buffers'):
                     # don't compare buffers
                     continue
                 rvalue = record.get(key, None)
                 if evalue and rvalue and evalue != rvalue:
-                    self.log.warn("conflicting initial state for record: %r:%r <%r> %r"%(msg_id, rvalue, key, evalue))
+                    self.log.warn("conflicting initial state for record: %r:%r <%r> %r", msg_id, rvalue, key, evalue)
                 elif evalue and not rvalue:
                     record[key] = evalue
             try:
                 self.db.update_record(msg_id, record)
             except Exception:
-                self.log.error("DB Error updating record %r"%msg_id, exc_info=True)
+                self.log.error("DB Error updating record %r", msg_id, exc_info=True)
         except KeyError:
             try:
                 self.db.add_record(msg_id, record)
             except Exception:
-                self.log.error("DB Error adding record %r"%msg_id, exc_info=True)
+                self.log.error("DB Error adding record %r", msg_id, exc_info=True)
         except Exception:
-            self.log.error("DB Error saving task request %r"%msg_id, exc_info=True)
+            self.log.error("DB Error saving task request %r", msg_id, exc_info=True)
 
     def save_task_result(self, idents, msg):
         """save the result of a completed task."""
         client_id = idents[0]
         try:
             msg = self.session.unserialize(msg)
         except Exception:
-            self.log.error("task::invalid task result message send to %r: %r"%(
-                    client_id, msg), exc_info=True)
+            self.log.error("task::invalid task result message send to %r: %r",
+                    client_id, msg, exc_info=True)
             return
 
         parent = msg['parent_header']
         if not parent:
             # print msg
-            self.log.warn("Task %r had no parent!"%msg)
+            self.log.warn("Task %r had no parent!", msg)
             return
         msg_id = parent['msg_id']
         if msg_id in self.unassigned:
             self.unassigned.remove(msg_id)
 
         header = msg['header']
         engine_uuid = header.get('engine', None)
         eid = self.by_ident.get(engine_uuid, None)
 
         if msg_id in self.pending:
+            self.log.info("task::task %r finished on %s", msg_id, eid)
             self.pending.remove(msg_id)
             self.all_completed.add(msg_id)
             if eid is not None:
                 self.completed[eid].append(msg_id)
                 if msg_id in self.tasks[eid]:
                     self.tasks[eid].remove(msg_id)
             completed = header['date']
             started = header.get('started', None)
             result = {
                 'result_header' : header,
                 'result_content': msg['content'],
                 'started' : started,
                 'completed' : completed,
                 'engine_uuid': engine_uuid
             }
 
             result['result_buffers'] = msg['buffers']
             try:
                 self.db.update_record(msg_id, result)
             except Exception:
-                self.log.error("DB Error saving task request %r"%msg_id, exc_info=True)
+                self.log.error("DB Error saving task request %r", msg_id, exc_info=True)
 
         else:
-            self.log.debug("task::unknown task %r finished"%msg_id)
+            self.log.debug("task::unknown task %r finished", msg_id)
 
     def save_task_destination(self, idents, msg):
         try:
             msg = self.session.unserialize(msg, content=True)
         except Exception:
             self.log.error("task::invalid task tracking message", exc_info=True)
             return
         content = msg['content']
         # print (content)
         msg_id = content['msg_id']
         engine_uuid = content['engine_id']
         eid = self.by_ident[util.asbytes(engine_uuid)]
 
-        self.log.info("task::task %r arrived on %r"%(msg_id, eid))
+        self.log.info("task::task %r arrived on %r", msg_id, eid)
         if msg_id in self.unassigned:
             self.unassigned.remove(msg_id)
         # else:
         #     self.log.debug("task::task %r 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))
         try:
             self.db.update_record(msg_id, dict(engine_uuid=engine_uuid))
         except Exception:
-            self.log.error("DB Error saving task destination %r"%msg_id, exc_info=True)
+            self.log.error("DB Error saving task destination %r", msg_id, exc_info=True)
 
 
     def mia_task_request(self, idents, msg):
         raise NotImplementedError
         client_id = idents[0]
         # content = dict(mia=self.mia,status='ok')
         # self.session.send('mia_reply', content=content, idents=client_id)
 
 
     #--------------------- IOPub Traffic ------------------------------
 
     def save_iopub_message(self, topics, msg):
         """save an iopub message into the db"""
         # print (topics)
         try:
             msg = self.session.unserialize(msg, content=True)
         except Exception:
             self.log.error("iopub::invalid IOPub message", exc_info=True)
             return
 
         parent = msg['parent_header']
         if not parent:
-            self.log.error("iopub::invalid IOPub message: %r"%msg)
+            self.log.error("iopub::invalid IOPub message: %r", msg)
             return
         msg_id = parent['msg_id']
         msg_type = msg['header']['msg_type']
         content = msg['content']
 
         # ensure msg_id is in db
         try:
             rec = self.db.get_record(msg_id)
         except KeyError:
             rec = empty_record()
             rec['msg_id'] = msg_id
             self.db.add_record(msg_id, rec)
         # stream
         d = {}
         if msg_type == 'stream':
             name = content['name']
             s = rec[name] or ''
             d[name] = s + content['data']
 
         elif msg_type == 'pyerr':
             d['pyerr'] = content
         elif msg_type == 'pyin':
             d['pyin'] = content['code']
         else:
             d[msg_type] = content.get('data', '')
 
         try:
             self.db.update_record(msg_id, d)
         except Exception:
-            self.log.error("DB Error saving iopub message %r"%msg_id, exc_info=True)
+            self.log.error("DB Error saving iopub message %r", msg_id, exc_info=True)
 
 
 
     #-------------------------------------------------------------------------
     # Registration requests
     #-------------------------------------------------------------------------
 
     def connection_request(self, client_id, msg):
         """Reply with connection addresses for clients."""
-        self.log.info("client::client %r connected"%client_id)
+        self.log.info("client::client %r connected", client_id)
         content = dict(status='ok')
         content.update(self.client_info)
         jsonable = {}
         for k,v in self.keytable.iteritems():
             if v not in self.dead_engines:
                 jsonable[str(k)] = v.decode('ascii')
         content['engines'] = jsonable
         self.session.send(self.query, 'connection_reply', content, parent=msg, ident=client_id)
 
     def register_engine(self, reg, msg):
         """Register a new engine."""
         content = msg['content']
         try:
             queue = util.asbytes(content['queue'])
         except KeyError:
             self.log.error("registration::queue not specified", exc_info=True)
             return
         heart = content.get('heartbeat', None)
         if heart:
             heart = util.asbytes(heart)
         """register a new engine, and create the socket(s) necessary"""
         eid = self._next_id
         # print (eid, queue, reg, heart)
 
-        self.log.debug("registration::register_engine(%i, %r, %r, %r)"%(eid, queue, reg, heart))
+        self.log.debug("registration::register_engine(%i, %r, %r, %r)", eid, queue, reg, heart)
 
         content = dict(id=eid,status='ok')
         content.update(self.engine_info)
         # check if requesting available IDs:
         if queue in self.by_ident:
             try:
                 raise KeyError("queue_id %r in use"%queue)
             except:
                 content = error.wrap_exception()
-                self.log.error("queue_id %r in use"%queue, exc_info=True)
+                self.log.error("queue_id %r in use", queue, exc_info=True)
         elif heart in self.hearts: # need to check unique hearts?
             try:
                 raise KeyError("heart_id %r in use"%heart)
             except:
-                self.log.error("heart_id %r in use"%heart, exc_info=True)
+                self.log.error("heart_id %r in use", heart, exc_info=True)
                 content = error.wrap_exception()
         else:
             for h, pack in self.incoming_registrations.iteritems():
                 if heart == h:
                     try:
                         raise KeyError("heart_id %r in use"%heart)
                     except:
-                        self.log.error("heart_id %r in use"%heart, exc_info=True)
+                        self.log.error("heart_id %r in use", heart, exc_info=True)
                         content = error.wrap_exception()
                     break
                 elif queue == pack[1]:
                     try:
                         raise KeyError("queue_id %r in use"%queue)
                     except:
-                        self.log.error("queue_id %r in use"%queue, exc_info=True)
+                        self.log.error("queue_id %r in use", queue, exc_info=True)
                         content = error.wrap_exception()
                     break
 
         msg = self.session.send(self.query, "registration_reply",
                 content=content,
                 ident=reg)
 
         if content['status'] == 'ok':
             if heart in self.heartmonitor.hearts:
                 # already beating
                 self.incoming_registrations[heart] = (eid,queue,reg[0],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[0],dc)
         else:
-            self.log.error("registration::registration %i failed: %r"%(eid, content['evalue']))
+            self.log.error("registration::registration %i failed: %r", 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:
-            self.log.error("registration::bad engine id for unregistration: %r"%ident, exc_info=True)
+            self.log.error("registration::bad engine id for unregistration: %r", ident, exc_info=True)
             return
-        self.log.info("registration::unregister_engine(%r)"%eid)
+        self.log.info("registration::unregister_engine(%r)", eid)
         # print (eid)
         uuid = self.keytable[eid]
         content=dict(id=eid, queue=uuid.decode('ascii'))
         self.dead_engines.add(uuid)
         # self.ids.remove(eid)
         # uuid = self.keytable.pop(eid)
         #
         # ec = self.engines.pop(eid)
         # self.hearts.pop(ec.heartbeat)
         # self.by_ident.pop(ec.queue)
         # self.completed.pop(eid)
         handleit = lambda : self._handle_stranded_msgs(eid, uuid)
         dc = ioloop.DelayedCallback(handleit, self.registration_timeout, self.loop)
         dc.start()
         ############## TODO: HANDLE IT ################
 
         if self.notifier:
             self.session.send(self.notifier, "unregistration_notification", content=content)
 
     def _handle_stranded_msgs(self, eid, uuid):
         """Handle messages known to be on an engine when the engine unregisters.
 
         It is possible that this will fire prematurely - that is, an engine will
         go down after completing a result, and the client will be notified
         that the result failed and later receive the actual result.
         """
 
         outstanding = self.queues[eid]
 
         for msg_id in outstanding:
             self.pending.remove(msg_id)
             self.all_completed.add(msg_id)
             try:
                 raise error.EngineError("Engine %r died while running task %r"%(eid, msg_id))
             except:
                 content = error.wrap_exception()
             # build a fake header:
             header = {}
             header['engine'] = uuid
             header['date'] = datetime.now()
             rec = dict(result_content=content, result_header=header, result_buffers=[])
             rec['completed'] = header['date']
             rec['engine_uuid'] = uuid
             try:
                 self.db.update_record(msg_id, rec)
             except Exception:
-                self.log.error("DB Error handling stranded msg %r"%msg_id, exc_info=True)
+                self.log.error("DB Error handling stranded msg %r", msg_id, exc_info=True)
 
 
     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:
             self.log.error("registration::tried to finish nonexistant registration", exc_info=True)
             return
-        self.log.info("registration::finished registering engine %i:%r"%(eid,queue))
+        self.log.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(id=eid, queue=queue, registration=reg,
                                     control=control, heartbeat=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.decode('ascii'))
         if self.notifier:
             self.session.send(self.notifier, "registration_notification", content=content)
-        self.log.info("engine::Engine Connected: %i"%eid)
+        self.log.info("engine::Engine Connected: %i", eid)
 
     def _purge_stalled_registration(self, heart):
         if heart in self.incoming_registrations:
             eid = self.incoming_registrations.pop(heart)[0]
-            self.log.info("registration::purging stalled registration: %i"%eid)
+            self.log.info("registration::purging stalled registration: %i", eid)
         else:
             pass
 
     #-------------------------------------------------------------------------
     # Client Requests
     #-------------------------------------------------------------------------
 
     def shutdown_request(self, client_id, msg):
         """handle shutdown request."""
         self.session.send(self.query, 'shutdown_reply', content={'status': 'ok'}, ident=client_id)
         # also notify other clients of shutdown
         self.session.send(self.notifier, 'shutdown_notice', content={'status': 'ok'})
         dc = ioloop.DelayedCallback(lambda : self._shutdown(), 1000, self.loop)
         dc.start()
 
     def _shutdown(self):
         self.log.info("hub::hub shutting down.")
         time.sleep(0.1)
         sys.exit(0)
 
 
     def check_load(self, client_id, msg):
         content = msg['content']
         try:
             targets = content['targets']
             targets = self._validate_targets(targets)
         except:
             content = error.wrap_exception()
             self.session.send(self.query, "hub_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.query, "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 = error.wrap_exception()
             self.session.send(self.query, "hub_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[str(t)] = {'queue': queue, 'completed': completed , 'tasks': tasks}
         content['unassigned'] = list(self.unassigned) if verbose else len(self.unassigned)
         # print (content)
         self.session.send(self.query, "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']
         self.log.info("Dropping records with %s", content)
         msg_ids = content.get('msg_ids', [])
         reply = dict(status='ok')
         if msg_ids == 'all':
             try:
                 self.db.drop_matching_records(dict(completed={'$ne':None}))
             except Exception:
                 reply = error.wrap_exception()
         else:
             pending = filter(lambda m: m in self.pending, msg_ids)
             if pending:
                 try:
                     raise IndexError("msg pending: %r"%pending[0])
                 except:
                     reply = error.wrap_exception()
             else:
                 try:
                     self.db.drop_matching_records(dict(msg_id={'$in':msg_ids}))
                 except Exception:
                     reply = error.wrap_exception()
 
             if reply['status'] == 'ok':
                 eids = content.get('engine_ids', [])
                 for eid in eids:
                     if eid not in self.engines:
                         try:
                             raise IndexError("No such engine: %i"%eid)
                         except:
                             reply = error.wrap_exception()
                         break
                     uid = self.engines[eid].queue
                     try:
                         self.db.drop_matching_records(dict(engine_uuid=uid, completed={'$ne':None}))
                     except Exception:
                         reply = error.wrap_exception()
                         break
 
         self.session.send(self.query, 'purge_reply', content=reply, ident=client_id)
 
     def resubmit_task(self, client_id, msg):
         """Resubmit one or more tasks."""
         def finish(reply):
             self.session.send(self.query, 'resubmit_reply', content=reply, ident=client_id)
 
         content = msg['content']
         msg_ids = content['msg_ids']
         reply = dict(status='ok')
         try:
             records = self.db.find_records({'msg_id' : {'$in' : msg_ids}}, keys=[
                 'header', 'content', 'buffers'])
         except Exception:
             self.log.error('db::db error finding tasks to resubmit', exc_info=True)
             return finish(error.wrap_exception())
 
         # validate msg_ids
         found_ids = [ rec['msg_id'] for rec in records ]
         invalid_ids = filter(lambda m: m in self.pending, found_ids)
         if len(records) > len(msg_ids):
             try:
                 raise RuntimeError("DB appears to be in an inconsistent state."
                     "More matching records were found than should exist")
             except Exception:
                 return finish(error.wrap_exception())
         elif len(records) < len(msg_ids):
             missing = [ m for m in msg_ids if m not in found_ids ]
             try:
                 raise KeyError("No such msg(s): %r"%missing)
             except KeyError:
                 return finish(error.wrap_exception())
         elif invalid_ids:
             msg_id = invalid_ids[0]
             try:
-                raise ValueError("Task %r appears to be inflight"%(msg_id))
+                raise ValueError("Task %r appears to be inflight" % msg_id)
             except Exception:
                 return finish(error.wrap_exception())
 
         # clear the existing records
         now = datetime.now()
         rec = empty_record()
         map(rec.pop, ['msg_id', 'header', 'content', 'buffers', 'submitted'])
         rec['resubmitted'] = now
         rec['queue'] = 'task'
         rec['client_uuid'] = client_id[0]
         try:
             for msg_id in msg_ids:
                 self.all_completed.discard(msg_id)
                 self.db.update_record(msg_id, rec)
         except Exception:
             self.log.error('db::db error upating record', exc_info=True)
             reply = error.wrap_exception()
         else:
             # send the messages
             for rec in records:
                 header = rec['header']
                 # include resubmitted in header to prevent digest collision
                 header['resubmitted'] = now
                 msg = self.session.msg(header['msg_type'])
                 msg['content'] = rec['content']
                 msg['header'] = header
                 msg['header']['msg_id'] = rec['msg_id']
                 self.session.send(self.resubmit, msg, buffers=rec['buffers'])
 
         finish(dict(status='ok'))
 
 
     def _extract_record(self, rec):
         """decompose a TaskRecord dict into subsection of reply for get_result"""
         io_dict = {}
         for key in 'pyin pyout pyerr stdout stderr'.split():
                 io_dict[key] = rec[key]
         content = { 'result_content': rec['result_content'],
                             'header': rec['header'],
                             'result_header' : rec['result_header'],
                             'io' : io_dict,
                           }
         if rec['result_buffers']:
             buffers = map(bytes, rec['result_buffers'])
         else:
             buffers = []
 
         return content, buffers
 
     def get_results(self, client_id, msg):
         """Get the result of 1 or more messages."""
         content = msg['content']
         msg_ids = sorted(set(content['msg_ids']))
         statusonly = content.get('status_only', False)
         pending = []
         completed = []
         content = dict(status='ok')
         content['pending'] = pending
         content['completed'] = completed
         buffers = []
         if not statusonly:
             try:
                 matches = self.db.find_records(dict(msg_id={'$in':msg_ids}))
                 # turn match list into dict, for faster lookup
                 records = {}
                 for rec in matches:
                     records[rec['msg_id']] = rec
             except Exception:
                 content = error.wrap_exception()
                 self.session.send(self.query, "result_reply", content=content,
                                                     parent=msg, ident=client_id)
                 return
         else:
             records = {}
         for msg_id in msg_ids:
             if msg_id in self.pending:
                 pending.append(msg_id)
             elif msg_id in self.all_completed:
                 completed.append(msg_id)
                 if not statusonly:
                     c,bufs = self._extract_record(records[msg_id])
                     content[msg_id] = c
                     buffers.extend(bufs)
             elif msg_id in records:
                 if rec['completed']:
                     completed.append(msg_id)
                     c,bufs = self._extract_record(records[msg_id])
                     content[msg_id] = c
                     buffers.extend(bufs)
                 else:
                     pending.append(msg_id)
             else:
                 try:
                     raise KeyError('No such message: '+msg_id)
                 except:
                     content = error.wrap_exception()
                 break
         self.session.send(self.query, "result_reply", content=content,
                                             parent=msg, ident=client_id,
                                             buffers=buffers)
 
     def get_history(self, client_id, msg):
         """Get a list of all msg_ids in our DB records"""
         try:
             msg_ids = self.db.get_history()
         except Exception as e:
             content = error.wrap_exception()
         else:
             content = dict(status='ok', history=msg_ids)
 
         self.session.send(self.query, "history_reply", content=content,
                                             parent=msg, ident=client_id)
 
     def db_query(self, client_id, msg):
         """Perform a raw query on the task record database."""
         content = msg['content']
         query = content.get('query', {})
         keys = content.get('keys', None)
         buffers = []
         empty = list()
         try:
             records = self.db.find_records(query, keys)
         except Exception as e:
             content = error.wrap_exception()
         else:
             # extract buffers from reply content:
             if keys is not None:
                 buffer_lens = [] if 'buffers' in keys else None
                 result_buffer_lens = [] if 'result_buffers' in keys else None
             else:
                 buffer_lens = []
                 result_buffer_lens = []
 
             for rec in records:
                 # buffers may be None, so double check
                 if buffer_lens is not None:
                     b = rec.pop('buffers', empty) or empty
                     buffer_lens.append(len(b))
                     buffers.extend(b)
                 if result_buffer_lens is not None:
                     rb = rec.pop('result_buffers', empty) or empty
                     result_buffer_lens.append(len(rb))
                     buffers.extend(rb)
             content = dict(status='ok', records=records, buffer_lens=buffer_lens,
                                     result_buffer_lens=result_buffer_lens)
         # self.log.debug (content)
         self.session.send(self.query, "db_reply", content=content,
                                             parent=msg, ident=client_id,
                                             buffers=buffers)
 

IPython/parallel/controller/scheduler.py

@@ -1,716 +1,716
 """The Python scheduler for rich scheduling.
 
 The Pure ZMQ scheduler does not allow routing schemes other than LRU,
 nor does it check msg_id DAG dependencies. For those, a slightly slower
 Python Scheduler exists.
 
 Authors:
 
 * Min RK
 """
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2010-2011  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
 
 import logging
 import sys
 
 from datetime import datetime, timedelta
 from random import randint, random
 from types import FunctionType
 
 try:
     import numpy
 except ImportError:
     numpy = None
 
 import zmq
 from zmq.eventloop import ioloop, zmqstream
 
 # local imports
 from IPython.external.decorator import decorator
 from IPython.config.application import Application
 from IPython.config.loader import Config
 from IPython.utils.traitlets import Instance, Dict, List, Set, Integer, Enum, CBytes
 
 from IPython.parallel import error
 from IPython.parallel.factory import SessionFactory
 from IPython.parallel.util import connect_logger, local_logger, asbytes
 
 from .dependency import Dependency
 
 @decorator
 def logged(f,self,*args,**kwargs):
     # print ("#--------------------")
     self.log.debug("scheduler::%s(*%s,**%s)", f.func_name, args, kwargs)
     # print ("#--")
     return f(self,*args, **kwargs)
 
 #----------------------------------------------------------------------
 # Chooser functions
 #----------------------------------------------------------------------
 
 def plainrandom(loads):
     """Plain random pick."""
     n = len(loads)
     return randint(0,n-1)
 
 def lru(loads):
     """Always pick the front of the line.
 
     The content of `loads` is ignored.
 
     Assumes LRU ordering of loads, with oldest first.
     """
     return 0
 
 def twobin(loads):
     """Pick two at random, use the LRU of the two.
 
     The content of loads is ignored.
 
     Assumes LRU ordering of loads, with oldest first.
     """
     n = len(loads)
     a = randint(0,n-1)
     b = randint(0,n-1)
     return min(a,b)
 
 def weighted(loads):
     """Pick two at random using inverse load as weight.
 
     Return the less loaded of the two.
     """
     # weight 0 a million times more than 1:
     weights = 1./(1e-6+numpy.array(loads))
     sums = weights.cumsum()
     t = sums[-1]
     x = random()*t
     y = random()*t
     idx = 0
     idy = 0
     while sums[idx] < x:
         idx += 1
     while sums[idy] < y:
         idy += 1
     if weights[idy] > weights[idx]:
         return idy
     else:
         return idx
 
 def leastload(loads):
     """Always choose the lowest load.
 
     If the lowest load occurs more than once, the first
     occurance will be used.  If loads has LRU ordering, this means
     the LRU of those with the lowest load is chosen.
     """
     return loads.index(min(loads))
 
 #---------------------------------------------------------------------
 # Classes
 #---------------------------------------------------------------------
 # store empty default dependency:
 MET = Dependency([])
 
 class TaskScheduler(SessionFactory):
     """Python TaskScheduler object.
 
     This is the simplest object that supports msg_id based
     DAG dependencies. *Only* task msg_ids are checked, not
     msg_ids of jobs submitted via the MUX queue.
 
     """
 
     hwm = Integer(0, config=True, shortname='hwm',
         help="""specify the High Water Mark (HWM) for the downstream
         socket in the Task scheduler. This is the maximum number
         of allowed outstanding tasks on each engine."""
     )
     scheme_name = Enum(('leastload', 'pure', 'lru', 'plainrandom', 'weighted', 'twobin'),
         'leastload', config=True, shortname='scheme', allow_none=False,
         help="""select the task scheduler scheme  [default: Python LRU]
         Options are: 'pure', 'lru', 'plainrandom', 'weighted', 'twobin','leastload'"""
     )
     def _scheme_name_changed(self, old, new):
         self.log.debug("Using scheme %r"%new)
         self.scheme = globals()[new]
 
     # input arguments:
     scheme = Instance(FunctionType) # function for determining the destination
     def _scheme_default(self):
         return leastload
     client_stream = Instance(zmqstream.ZMQStream) # client-facing stream
     engine_stream = Instance(zmqstream.ZMQStream) # engine-facing stream
     notifier_stream = Instance(zmqstream.ZMQStream) # hub-facing sub stream
     mon_stream = Instance(zmqstream.ZMQStream) # hub-facing pub stream
 
     # internals:
     graph = Dict() # dict by msg_id of [ msg_ids that depend on key ]
     retries = Dict() # dict by msg_id of retries remaining (non-neg ints)
     # waiting = List() # list of msg_ids ready to run, but haven't due to HWM
     depending = Dict() # dict by msg_id of (msg_id, raw_msg, after, follow)
     pending = Dict() # dict by engine_uuid of submitted tasks
     completed = Dict() # dict by engine_uuid of completed tasks
     failed = Dict() # dict by engine_uuid of failed tasks
     destinations = Dict() # dict by msg_id of engine_uuids where jobs ran (reverse of completed+failed)
     clients = Dict() # dict by msg_id for who submitted the task
     targets = List() # list of target IDENTs
     loads = List() # list of engine loads
     # full = Set() # set of IDENTs that have HWM outstanding tasks
     all_completed = Set() # set of all completed tasks
     all_failed = Set() # set of all failed tasks
     all_done = Set() # set of all finished tasks=union(completed,failed)
     all_ids = Set() # set of all submitted task IDs
     blacklist = Dict() # dict by msg_id of locations where a job has encountered UnmetDependency
     auditor = Instance('zmq.eventloop.ioloop.PeriodicCallback')
 
     ident = CBytes() # ZMQ identity. This should just be self.session.session
                      # but ensure Bytes
     def _ident_default(self):
         return self.session.bsession
 
     def start(self):
         self.engine_stream.on_recv(self.dispatch_result, copy=False)
         self._notification_handlers = dict(
             registration_notification = self._register_engine,
             unregistration_notification = self._unregister_engine
         )
         self.notifier_stream.on_recv(self.dispatch_notification)
         self.auditor = ioloop.PeriodicCallback(self.audit_timeouts, 2e3, self.loop) # 1 Hz
         self.auditor.start()
         self.log.info("Scheduler started [%s]"%self.scheme_name)
 
     def resume_receiving(self):
         """Resume accepting jobs."""
         self.client_stream.on_recv(self.dispatch_submission, copy=False)
 
     def stop_receiving(self):
         """Stop accepting jobs while there are no engines.
         Leave them in the ZMQ queue."""
         self.client_stream.on_recv(None)
 
     #-----------------------------------------------------------------------
     # [Un]Registration Handling
     #-----------------------------------------------------------------------
 
     def dispatch_notification(self, msg):
         """dispatch register/unregister events."""
         try:
             idents,msg = self.session.feed_identities(msg)
         except ValueError:
             self.log.warn("task::Invalid Message: %r",msg)
             return
         try:
             msg = self.session.unserialize(msg)
         except ValueError:
             self.log.warn("task::Unauthorized message from: %r"%idents)
             return
 
         msg_type = msg['header']['msg_type']
 
         handler = self._notification_handlers.get(msg_type, None)
         if handler is None:
             self.log.error("Unhandled message type: %r"%msg_type)
         else:
             try:
                 handler(asbytes(msg['content']['queue']))
             except Exception:
-                self.log.error("task::Invalid notification msg: %r",msg)
+                self.log.error("task::Invalid notification msg: %r", msg, exc_info=True)
 
     def _register_engine(self, uid):
         """New engine with ident `uid` became available."""
         # head of the line:
         self.targets.insert(0,uid)
         self.loads.insert(0,0)
 
         # initialize sets
         self.completed[uid] = set()
         self.failed[uid] = set()
         self.pending[uid] = {}
         if len(self.targets) == 1:
             self.resume_receiving()
         # rescan the graph:
         self.update_graph(None)
 
     def _unregister_engine(self, uid):
         """Existing engine with ident `uid` became unavailable."""
         if len(self.targets) == 1:
             # this was our only engine
             self.stop_receiving()
 
         # handle any potentially finished tasks:
         self.engine_stream.flush()
 
         # don't pop destinations, because they might be used later
         # map(self.destinations.pop, self.completed.pop(uid))
         # map(self.destinations.pop, self.failed.pop(uid))
 
         # prevent this engine from receiving work
         idx = self.targets.index(uid)
         self.targets.pop(idx)
         self.loads.pop(idx)
 
         # wait 5 seconds before cleaning up pending jobs, since the results might
         # still be incoming
         if self.pending[uid]:
             dc = ioloop.DelayedCallback(lambda : self.handle_stranded_tasks(uid), 5000, self.loop)
             dc.start()
         else:
             self.completed.pop(uid)
             self.failed.pop(uid)
 
 
     def handle_stranded_tasks(self, engine):
         """Deal with jobs resident in an engine that died."""
         lost = self.pending[engine]
         for msg_id in lost.keys():
             if msg_id not in self.pending[engine]:
                 # prevent double-handling of messages
                 continue
 
             raw_msg = lost[msg_id][0]
             idents,msg = self.session.feed_identities(raw_msg, copy=False)
             parent = self.session.unpack(msg[1].bytes)
             idents = [engine, idents[0]]
 
             # build fake error reply
             try:
                 raise error.EngineError("Engine %r died while running task %r"%(engine, msg_id))
             except:
                 content = error.wrap_exception()
             msg = self.session.msg('apply_reply', content, parent=parent, subheader={'status':'error'})
             raw_reply = map(zmq.Message, self.session.serialize(msg, ident=idents))
             # and dispatch it
             self.dispatch_result(raw_reply)
 
         # finally scrub completed/failed lists
         self.completed.pop(engine)
         self.failed.pop(engine)
 
 
     #-----------------------------------------------------------------------
     # Job Submission
     #-----------------------------------------------------------------------
     def dispatch_submission(self, raw_msg):
         """Dispatch job submission to appropriate handlers."""
         # ensure targets up to date:
         self.notifier_stream.flush()
         try:
             idents, msg = self.session.feed_identities(raw_msg, copy=False)
             msg = self.session.unserialize(msg, content=False, copy=False)
         except Exception:
             self.log.error("task::Invaid task msg: %r"%raw_msg, exc_info=True)
             return
 
 
         # send to monitor
         self.mon_stream.send_multipart([b'intask']+raw_msg, copy=False)
 
         header = msg['header']
         msg_id = header['msg_id']
         self.all_ids.add(msg_id)
 
         # get targets as a set of bytes objects
         # from a list of unicode objects
         targets = header.get('targets', [])
         targets = map(asbytes, targets)
         targets = set(targets)
 
         retries = header.get('retries', 0)
         self.retries[msg_id] = retries
 
         # time dependencies
         after = header.get('after', None)
         if after:
             after = Dependency(after)
             if after.all:
                 if after.success:
                     after = Dependency(after.difference(self.all_completed),
                                 success=after.success,
                                 failure=after.failure,
                                 all=after.all,
                     )
                 if after.failure:
                     after = Dependency(after.difference(self.all_failed),
                                 success=after.success,
                                 failure=after.failure,
                                 all=after.all,
                     )
             if after.check(self.all_completed, self.all_failed):
                 # recast as empty set, if `after` already met,
                 # to prevent unnecessary set comparisons
                 after = MET
         else:
             after = MET
 
         # location dependencies
         follow = Dependency(header.get('follow', []))
 
         # turn timeouts into datetime objects:
         timeout = header.get('timeout', None)
         if timeout:
             # cast to float, because jsonlib returns floats as decimal.Decimal,
             # which timedelta does not accept
             timeout = datetime.now() + timedelta(0,float(timeout),0)
 
         args = [raw_msg, targets, after, follow, timeout]
 
         # validate and reduce dependencies:
         for dep in after,follow:
             if not dep: # empty dependency
                 continue
             # check valid:
             if msg_id in dep or dep.difference(self.all_ids):
                 self.depending[msg_id] = args
                 return self.fail_unreachable(msg_id, error.InvalidDependency)
             # check if unreachable:
             if dep.unreachable(self.all_completed, self.all_failed):
                 self.depending[msg_id] = args
                 return self.fail_unreachable(msg_id)
 
         if after.check(self.all_completed, self.all_failed):
             # time deps already met, try to run
             if not self.maybe_run(msg_id, *args):
                 # can't run yet
                 if msg_id not in self.all_failed:
                     # could have failed as unreachable
                     self.save_unmet(msg_id, *args)
         else:
             self.save_unmet(msg_id, *args)
 
     def audit_timeouts(self):
         """Audit all waiting tasks for expired timeouts."""
         now = datetime.now()
         for msg_id in self.depending.keys():
             # must recheck, in case one failure cascaded to another:
             if msg_id in self.depending:
                 raw,after,targets,follow,timeout = self.depending[msg_id]
                 if timeout and timeout < now:
                     self.fail_unreachable(msg_id, error.TaskTimeout)
 
     def fail_unreachable(self, msg_id, why=error.ImpossibleDependency):
         """a task has become unreachable, send a reply with an ImpossibleDependency
         error."""
         if msg_id not in self.depending:
             self.log.error("msg %r already failed!", msg_id)
             return
         raw_msg,targets,after,follow,timeout = self.depending.pop(msg_id)
         for mid in follow.union(after):
             if mid in self.graph:
                 self.graph[mid].remove(msg_id)
 
         # FIXME: unpacking a message I've already unpacked, but didn't save:
         idents,msg = self.session.feed_identities(raw_msg, copy=False)
         header = self.session.unpack(msg[1].bytes)
 
         try:
             raise why()
         except:
             content = error.wrap_exception()
 
         self.all_done.add(msg_id)
         self.all_failed.add(msg_id)
 
         msg = self.session.send(self.client_stream, 'apply_reply', content,
                                                 parent=header, ident=idents)
         self.session.send(self.mon_stream, msg, ident=[b'outtask']+idents)
 
         self.update_graph(msg_id, success=False)
 
     def maybe_run(self, msg_id, raw_msg, targets, after, follow, timeout):
         """check location dependencies, and run if they are met."""
         blacklist = self.blacklist.setdefault(msg_id, set())
         if follow or targets or blacklist or self.hwm:
             # we need a can_run filter
             def can_run(idx):
                 # check hwm
                 if self.hwm and self.loads[idx] == self.hwm:
                     return False
                 target = self.targets[idx]
                 # check blacklist
                 if target in blacklist:
                     return False
                 # check targets
                 if targets and target not in targets:
                     return False
                 # check follow
                 return follow.check(self.completed[target], self.failed[target])
 
             indices = filter(can_run, range(len(self.targets)))
 
             if not indices:
                 # couldn't run
                 if follow.all:
                     # check follow for impossibility
                     dests = set()
                     relevant = set()
                     if follow.success:
                         relevant = self.all_completed
                     if follow.failure:
                         relevant = relevant.union(self.all_failed)
                     for m in follow.intersection(relevant):
                         dests.add(self.destinations[m])
                     if len(dests) > 1:
                         self.depending[msg_id] = (raw_msg, targets, after, follow, timeout)
                         self.fail_unreachable(msg_id)
                         return False
                 if targets:
                     # check blacklist+targets for impossibility
                     targets.difference_update(blacklist)
                     if not targets or not targets.intersection(self.targets):
                         self.depending[msg_id] = (raw_msg, targets, after, follow, timeout)
                         self.fail_unreachable(msg_id)
                         return False
                 return False
         else:
             indices = None
 
         self.submit_task(msg_id, raw_msg, targets, follow, timeout, indices)
         return True
 
     def save_unmet(self, msg_id, raw_msg, targets, after, follow, timeout):
         """Save a message for later submission when its dependencies are met."""
         self.depending[msg_id] = [raw_msg,targets,after,follow,timeout]
         # track the ids in follow or after, but not those already finished
         for dep_id in after.union(follow).difference(self.all_done):
             if dep_id not in self.graph:
                 self.graph[dep_id] = set()
             self.graph[dep_id].add(msg_id)
 
     def submit_task(self, msg_id, raw_msg, targets, follow, timeout, indices=None):
         """Submit a task to any of a subset of our targets."""
         if indices:
             loads = [self.loads[i] for i in indices]
         else:
             loads = self.loads
         idx = self.scheme(loads)
         if indices:
             idx = indices[idx]
         target = self.targets[idx]
         # print (target, map(str, msg[:3]))
         # send job to the engine
         self.engine_stream.send(target, flags=zmq.SNDMORE, copy=False)
         self.engine_stream.send_multipart(raw_msg, copy=False)
         # update load
         self.add_job(idx)
         self.pending[target][msg_id] = (raw_msg, targets, MET, follow, timeout)
         # notify Hub
         content = dict(msg_id=msg_id, engine_id=target.decode('ascii'))
         self.session.send(self.mon_stream, 'task_destination', content=content,
                         ident=[b'tracktask',self.ident])
 
 
     #-----------------------------------------------------------------------
     # Result Handling
     #-----------------------------------------------------------------------
     def dispatch_result(self, raw_msg):
         """dispatch method for result replies"""
         try:
             idents,msg = self.session.feed_identities(raw_msg, copy=False)
             msg = self.session.unserialize(msg, content=False, copy=False)
             engine = idents[0]
             try:
                 idx = self.targets.index(engine)
             except ValueError:
                 pass # skip load-update for dead engines
             else:
                 self.finish_job(idx)
         except Exception:
             self.log.error("task::Invaid result: %r", raw_msg, exc_info=True)
             return
 
         header = msg['header']
         parent = msg['parent_header']
         if header.get('dependencies_met', True):
             success = (header['status'] == 'ok')
             msg_id = parent['msg_id']
             retries = self.retries[msg_id]
             if not success and retries > 0:
                 # failed
                 self.retries[msg_id] = retries - 1
                 self.handle_unmet_dependency(idents, parent)
             else:
                 del self.retries[msg_id]
                 # relay to client and update graph
                 self.handle_result(idents, parent, raw_msg, success)
                 # send to Hub monitor
                 self.mon_stream.send_multipart([b'outtask']+raw_msg, copy=False)
         else:
             self.handle_unmet_dependency(idents, parent)
 
     def handle_result(self, idents, parent, raw_msg, success=True):
         """handle a real task result, either success or failure"""
         # first, relay result to client
         engine = idents[0]
         client = idents[1]
         # swap_ids for XREP-XREP mirror
         raw_msg[:2] = [client,engine]
         # print (map(str, raw_msg[:4]))
         self.client_stream.send_multipart(raw_msg, copy=False)
         # now, update our data structures
         msg_id = parent['msg_id']
         self.blacklist.pop(msg_id, None)
         self.pending[engine].pop(msg_id)
         if success:
             self.completed[engine].add(msg_id)
             self.all_completed.add(msg_id)
         else:
             self.failed[engine].add(msg_id)
             self.all_failed.add(msg_id)
         self.all_done.add(msg_id)
         self.destinations[msg_id] = engine
 
         self.update_graph(msg_id, success)
 
     def handle_unmet_dependency(self, idents, parent):
         """handle an unmet dependency"""
         engine = idents[0]
         msg_id = parent['msg_id']
 
         if msg_id not in self.blacklist:
             self.blacklist[msg_id] = set()
         self.blacklist[msg_id].add(engine)
 
         args = self.pending[engine].pop(msg_id)
         raw,targets,after,follow,timeout = args
 
         if self.blacklist[msg_id] == targets:
             self.depending[msg_id] = args
             self.fail_unreachable(msg_id)
         elif not self.maybe_run(msg_id, *args):
             # resubmit failed
             if msg_id not in self.all_failed:
                 # put it back in our dependency tree
                 self.save_unmet(msg_id, *args)
 
         if self.hwm:
             try:
                 idx = self.targets.index(engine)
             except ValueError:
                 pass # skip load-update for dead engines
             else:
                 if self.loads[idx] == self.hwm-1:
                     self.update_graph(None)
 
 
 
     def update_graph(self, dep_id=None, success=True):
         """dep_id just finished. Update our dependency
         graph and submit any jobs that just became runable.
 
         Called with dep_id=None to update entire graph for hwm, but without finishing
         a task.
         """
         # print ("\n\n***********")
         # pprint (dep_id)
         # pprint (self.graph)
         # pprint (self.depending)
         # pprint (self.all_completed)
         # pprint (self.all_failed)
         # print ("\n\n***********\n\n")
         # update any jobs that depended on the dependency
         jobs = self.graph.pop(dep_id, [])
 
         # recheck *all* jobs if
         # a) we have HWM and an engine just become no longer full
         # or b) dep_id was given as None
         if dep_id is None or self.hwm and any( [ load==self.hwm-1 for load in self.loads ]):
             jobs = self.depending.keys()
 
         for msg_id in jobs:
             raw_msg, targets, after, follow, timeout = self.depending[msg_id]
 
             if after.unreachable(self.all_completed, self.all_failed)\
                     or follow.unreachable(self.all_completed, self.all_failed):
                 self.fail_unreachable(msg_id)
 
             elif after.check(self.all_completed, self.all_failed): # time deps met, maybe run
                 if self.maybe_run(msg_id, raw_msg, targets, MET, follow, timeout):
 
                     self.depending.pop(msg_id)
                     for mid in follow.union(after):
                         if mid in self.graph:
                             self.graph[mid].remove(msg_id)
 
     #----------------------------------------------------------------------
     # methods to be overridden by subclasses
     #----------------------------------------------------------------------
 
     def add_job(self, idx):
         """Called after self.targets[idx] just got the job with header.
         Override with subclasses.  The default ordering is simple LRU.
         The default loads are the number of outstanding jobs."""
         self.loads[idx] += 1
         for lis in (self.targets, self.loads):
             lis.append(lis.pop(idx))
 
 
     def finish_job(self, idx):
         """Called after self.targets[idx] just finished a job.
         Override with subclasses."""
         self.loads[idx] -= 1
 
 
 
 def launch_scheduler(in_addr, out_addr, mon_addr, not_addr, config=None,
                         logname='root', log_url=None, loglevel=logging.DEBUG,
                         identity=b'task', in_thread=False):
 
     ZMQStream = zmqstream.ZMQStream
 
     if config:
         # unwrap dict back into Config
         config = Config(config)
 
     if in_thread:
         # use instance() to get the same Context/Loop as our parent
         ctx = zmq.Context.instance()
         loop = ioloop.IOLoop.instance()
     else:
         # in a process, don't use instance()
         # for safety with multiprocessing
         ctx = zmq.Context()
         loop = ioloop.IOLoop()
     ins = ZMQStream(ctx.socket(zmq.ROUTER),loop)
     ins.setsockopt(zmq.IDENTITY, identity)
     ins.bind(in_addr)
 
     outs = ZMQStream(ctx.socket(zmq.ROUTER),loop)
     outs.setsockopt(zmq.IDENTITY, identity)
     outs.bind(out_addr)
     mons = zmqstream.ZMQStream(ctx.socket(zmq.PUB),loop)
     mons.connect(mon_addr)
     nots = zmqstream.ZMQStream(ctx.socket(zmq.SUB),loop)
     nots.setsockopt(zmq.SUBSCRIBE, b'')
     nots.connect(not_addr)
 
     # setup logging.
     if in_thread:
         log = Application.instance().log
     else:
         if log_url:
             log = connect_logger(logname, ctx, log_url, root="scheduler", loglevel=loglevel)
         else:
             log = local_logger(logname, loglevel)
 
     scheduler = TaskScheduler(client_stream=ins, engine_stream=outs,
                             mon_stream=mons, notifier_stream=nots,
                             loop=loop, log=log,
                             config=config)
     scheduler.start()
     if not in_thread:
         try:
             loop.start()
         except KeyboardInterrupt:
             print ("interrupted, exiting...", file=sys.__stderr__)