reorganize Factory classes to follow relocation of Session object

IPython/parallel/apps/ipclusterapp.py

@@ -1,521 +1,521 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 The ipcluster application.
 """
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2008-2009  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
 #-----------------------------------------------------------------------------
 
 import errno
 import logging
 import os
 import re
 import signal
 
 from subprocess import check_call, CalledProcessError, PIPE
 import zmq
 from zmq.eventloop import ioloop
 
 from IPython.config.application import Application, boolean_flag
 from IPython.config.loader import Config
 from IPython.core.newapplication import BaseIPythonApplication, ProfileDir
 from IPython.utils.importstring import import_item
 from IPython.utils.traitlets import Int, Unicode, Bool, CFloat, Dict, List
 
 from IPython.parallel.apps.baseapp import (
     BaseParallelApplication,
     PIDFileError,
     base_flags, base_aliases
 )
 
 
 #-----------------------------------------------------------------------------
 # Module level variables
 #-----------------------------------------------------------------------------
 
 
 default_config_file_name = u'ipcluster_config.py'
 
 
 _description = """Start an IPython cluster for parallel computing.
 
 An IPython cluster consists of 1 controller and 1 or more engines.
 This command automates the startup of these processes using a wide
 range of startup methods (SSH, local processes, PBS, mpiexec,
 Windows HPC Server 2008). To start a cluster with 4 engines on your
 local host simply do 'ipcluster start n=4'. For more complex usage
 you will typically do 'ipcluster create profile=mycluster', then edit
 configuration files, followed by 'ipcluster start profile=mycluster n=4'.
 """
 
 
 # Exit codes for ipcluster
 
 # This will be the exit code if the ipcluster appears to be running because
 # a .pid file exists
 ALREADY_STARTED = 10
 
 
 # This will be the exit code if ipcluster stop is run, but there is not .pid
 # file to be found.
 ALREADY_STOPPED = 11
 
 # This will be the exit code if ipcluster engines is run, but there is not .pid
 # file to be found.
 NO_CLUSTER = 12
 
 
 #-----------------------------------------------------------------------------
 # Main application
 #-----------------------------------------------------------------------------
 start_help = """Start an IPython cluster for parallel computing
 
 Start an ipython cluster by its profile name or cluster
 directory. Cluster directories contain configuration, log and
 security related files and are named using the convention
 'cluster_<profile>' and should be creating using the 'start'
 subcommand of 'ipcluster'. If your cluster directory is in 
 the cwd or the ipython directory, you can simply refer to it
 using its profile name, 'ipcluster start n=4 profile=<profile>`,
 otherwise use the 'profile_dir' option.
 """
 stop_help = """Stop a running IPython cluster
 
 Stop a running ipython cluster by its profile name or cluster
 directory. Cluster directories are named using the convention
 'cluster_<profile>'. If your cluster directory is in 
 the cwd or the ipython directory, you can simply refer to it
 using its profile name, 'ipcluster stop profile=<profile>`, otherwise
 use the 'profile_dir' option.
 """
 engines_help = """Start engines connected to an existing IPython cluster
 
 Start one or more engines to connect to an existing Cluster
 by profile name or cluster directory.
 Cluster directories contain configuration, log and
 security related files and are named using the convention
 'cluster_<profile>' and should be creating using the 'start'
 subcommand of 'ipcluster'. If your cluster directory is in 
 the cwd or the ipython directory, you can simply refer to it
 using its profile name, 'ipcluster engines n=4 profile=<profile>`,
 otherwise use the 'profile_dir' option.
 """
 create_help = """Create an ipcluster profile by name
 
 Create an ipython cluster directory by its profile name or
 cluster directory path. Cluster directories contain
 configuration, log and security related files and are named
 using the convention 'cluster_<profile>'. By default they are
 located in your ipython directory. Once created, you will
 probably need to edit the configuration files in the cluster
 directory to configure your cluster. Most users will create a
 cluster directory by profile name,
 `ipcluster create profile=mycluster`, which will put the directory
 in `<ipython_dir>/cluster_mycluster`.
 """
 list_help = """List available cluster profiles
 
 List all available clusters, by cluster directory, that can
 be found in the current working directly or in the ipython
 directory. Cluster directories are named using the convention
 'cluster_<profile>'.
 """
 
 
 class IPClusterList(BaseIPythonApplication):
     name = u'ipcluster-list'
     description = list_help
     
     # empty aliases
     aliases=Dict()
     flags = Dict(base_flags)
     
     def _log_level_default(self):
         return 20
     
     def list_profile_dirs(self):
         # Find the search paths
         profile_dir_paths = os.environ.get('IPYTHON_PROFILE_PATH','')
         if profile_dir_paths:
             profile_dir_paths = profile_dir_paths.split(':')
         else:
             profile_dir_paths = []
         
         ipython_dir = self.ipython_dir
         
         paths = [os.getcwd(), ipython_dir] + profile_dir_paths
         paths = list(set(paths))
 
         self.log.info('Searching for cluster profiles in paths: %r' % paths)
         for path in paths:
             files = os.listdir(path)
             for f in files:
                 full_path = os.path.join(path, f)
                 if os.path.isdir(full_path) and f.startswith('profile_') and \
                     os.path.isfile(os.path.join(full_path, 'ipcontroller_config.py')):
                     profile = f.split('_')[-1]
                     start_cmd = 'ipcluster start profile=%s n=4' % profile
                     print start_cmd + " ==> " + full_path
     
     def start(self):
         self.list_profile_dirs()
 
 
 # `ipcluster create` will be deprecated when `ipython profile create` or equivalent exists
 
 create_flags = {}
 create_flags.update(base_flags)
 create_flags.update(boolean_flag('reset', 'IPClusterCreate.overwrite', 
                 "reset config files to defaults", "leave existing config files"))
 
 class IPClusterCreate(BaseParallelApplication):
     name = u'ipcluster-create'
     description = create_help
     auto_create = Bool(True)
     config_file_name = Unicode(default_config_file_name)
     
     flags = Dict(create_flags)
     
     aliases = Dict(dict(profile='BaseIPythonApplication.profile'))
     
     classes = [ProfileDir]
     
     
 stop_aliases = dict(
     signal='IPClusterStop.signal',
     profile='BaseIPythonApplication.profile',
     profile_dir='ProfileDir.location',
 )
 
 class IPClusterStop(BaseParallelApplication):
     name = u'ipcluster'
     description = stop_help
     config_file_name = Unicode(default_config_file_name)
     
     signal = Int(signal.SIGINT, config=True,
         help="signal to use for stopping processes.")
         
     aliases = Dict(stop_aliases)
     
     def start(self):
         """Start the app for the stop subcommand."""
         try:
             pid = self.get_pid_from_file()
         except PIDFileError:
             self.log.critical(
                 'Could not read pid file, cluster is probably not running.'
             )
             # Here I exit with a unusual exit status that other processes
             # can watch for to learn how I existed.
             self.remove_pid_file()
             self.exit(ALREADY_STOPPED)
         
         if not self.check_pid(pid):
             self.log.critical(
                 'Cluster [pid=%r] is not running.' % pid
             )
             self.remove_pid_file()
             # Here I exit with a unusual exit status that other processes
             # can watch for to learn how I existed.
             self.exit(ALREADY_STOPPED)
             
         elif os.name=='posix':
             sig = self.signal
             self.log.info(
                 "Stopping cluster [pid=%r] with [signal=%r]" % (pid, sig)
             )
             try:
                 os.kill(pid, sig)
             except OSError:
                 self.log.error("Stopping cluster failed, assuming already dead.",
                     exc_info=True)
                 self.remove_pid_file()
         elif os.name=='nt':
             try:
                 # kill the whole tree
                 p = check_call(['taskkill', '-pid', str(pid), '-t', '-f'], stdout=PIPE,stderr=PIPE)
             except (CalledProcessError, OSError):
                 self.log.error("Stopping cluster failed, assuming already dead.",
                     exc_info=True)
             self.remove_pid_file()
     
 engine_aliases = {}
 engine_aliases.update(base_aliases)
 engine_aliases.update(dict(
     n='IPClusterEngines.n',
     elauncher = 'IPClusterEngines.engine_launcher_class',
 ))
 class IPClusterEngines(BaseParallelApplication):
 
     name = u'ipcluster'
     description = engines_help
     usage = None
     config_file_name = Unicode(default_config_file_name)
     default_log_level = logging.INFO
     classes = List()
     def _classes_default(self):
         from IPython.parallel.apps import launcher
         launchers = launcher.all_launchers
         eslaunchers = [ l for l in launchers if 'EngineSet' in l.__name__]
         return [ProfileDir]+eslaunchers
     
     n = Int(2, config=True,
         help="The number of engines to start.")
 
     engine_launcher_class = Unicode('LocalEngineSetLauncher',
         config=True,
         help="The class for launching a set of Engines."
         )
     daemonize = Bool(False, config=True,
         help='Daemonize the ipcluster program. This implies --log-to-file')
 
     def _daemonize_changed(self, name, old, new):
         if new:
             self.log_to_file = True
 
     aliases = Dict(engine_aliases)
     # flags = Dict(flags)
     _stopping = False
 
     def initialize(self, argv=None):
         super(IPClusterEngines, self).initialize(argv)
         self.init_signal()
         self.init_launchers()
     
     def init_launchers(self):
         self.engine_launcher = self.build_launcher(self.engine_launcher_class)
         self.engine_launcher.on_stop(lambda r: self.loop.stop())
     
     def init_signal(self):
         # Setup signals
         signal.signal(signal.SIGINT, self.sigint_handler)
     
     def build_launcher(self, clsname):
         """import and instantiate a Launcher based on importstring"""
         if '.' not in clsname:
             # not a module, presume it's the raw name in apps.launcher
             clsname = 'IPython.parallel.apps.launcher.'+clsname
         # print repr(clsname)
         klass = import_item(clsname)
 
         launcher = klass(
-            work_dir=self.profile_dir.location, config=self.config, logname=self.log.name
+            work_dir=self.profile_dir.location, config=self.config, log=self.log
         )
         return launcher
     
     def start_engines(self):
         self.log.info("Starting %i engines"%self.n)
         self.engine_launcher.start(
             self.n,
             self.profile_dir.location
         )
 
     def stop_engines(self):
         self.log.info("Stopping Engines...")
         if self.engine_launcher.running:
             d = self.engine_launcher.stop()
             return d
         else:
             return None
 
     def stop_launchers(self, r=None):
         if not self._stopping:
             self._stopping = True
             self.log.error("IPython cluster: stopping")
             self.stop_engines()
             # Wait a few seconds to let things shut down.
             dc = ioloop.DelayedCallback(self.loop.stop, 4000, self.loop)
             dc.start()
 
     def sigint_handler(self, signum, frame):
         self.log.debug("SIGINT received, stopping launchers...")
         self.stop_launchers()
         
     def start_logging(self):
         # Remove old log files of the controller and engine
         if self.clean_logs:
             log_dir = self.profile_dir.log_dir
             for f in os.listdir(log_dir):
                 if re.match(r'ip(engine|controller)z-\d+\.(log|err|out)',f):
                     os.remove(os.path.join(log_dir, f))
         # This will remove old log files for ipcluster itself
         # super(IPBaseParallelApplication, self).start_logging()
 
     def start(self):
         """Start the app for the engines subcommand."""
         self.log.info("IPython cluster: started")
         # First see if the cluster is already running
         
         # Now log and daemonize
         self.log.info(
             'Starting engines with [daemon=%r]' % self.daemonize
         )
         # TODO: Get daemonize working on Windows or as a Windows Server.
         if self.daemonize:
             if os.name=='posix':
                 from twisted.scripts._twistd_unix import daemonize
                 daemonize()
 
         dc = ioloop.DelayedCallback(self.start_engines, 0, self.loop)
         dc.start()
         # Now write the new pid file AFTER our new forked pid is active.
         # self.write_pid_file()
         try:
             self.loop.start()
         except KeyboardInterrupt:
             pass
         except zmq.ZMQError as e:
             if e.errno == errno.EINTR:
                 pass
             else:
                 raise
 
 start_aliases = {}
 start_aliases.update(engine_aliases)
 start_aliases.update(dict(
     delay='IPClusterStart.delay',
     clean_logs='IPClusterStart.clean_logs',
 ))
 
 class IPClusterStart(IPClusterEngines):
 
     name = u'ipcluster'
     description = start_help
     default_log_level = logging.INFO
     auto_create = Bool(True, config=True,
         help="whether to create the profile_dir if it doesn't exist")
     classes = List()
     def _classes_default(self,):
         from IPython.parallel.apps import launcher
         return [ProfileDir]+launcher.all_launchers
     
     clean_logs = Bool(True, config=True, 
         help="whether to cleanup old logs before starting")
 
     delay = CFloat(1., config=True,
         help="delay (in s) between starting the controller and the engines")
 
     controller_launcher_class = Unicode('LocalControllerLauncher',
         config=True,
         help="The class for launching a Controller."
         )
     reset = Bool(False, config=True,
         help="Whether to reset config files as part of '--create'."
         )
     
     # flags = Dict(flags)
     aliases = Dict(start_aliases)
 
     def init_launchers(self):
         self.controller_launcher = self.build_launcher(self.controller_launcher_class)
         self.engine_launcher = self.build_launcher(self.engine_launcher_class)
         self.controller_launcher.on_stop(self.stop_launchers)
     
     def start_controller(self):
         self.controller_launcher.start(
             self.profile_dir.location
         )
         
     def stop_controller(self):
         # self.log.info("In stop_controller")
         if self.controller_launcher and self.controller_launcher.running:
             return self.controller_launcher.stop()
 
     def stop_launchers(self, r=None):
         if not self._stopping:
             self.stop_controller()
             super(IPClusterStart, self).stop_launchers()
 
     def start(self):
         """Start the app for the start subcommand."""
         # First see if the cluster is already running
         try:
             pid = self.get_pid_from_file()
         except PIDFileError:
             pass
         else:
             if self.check_pid(pid):
                 self.log.critical(
                     'Cluster is already running with [pid=%s]. '
                     'use "ipcluster stop" to stop the cluster.' % pid
                 )
                 # Here I exit with a unusual exit status that other processes
                 # can watch for to learn how I existed.
                 self.exit(ALREADY_STARTED)
             else:
                 self.remove_pid_file()
                 
 
         # Now log and daemonize
         self.log.info(
             'Starting ipcluster with [daemon=%r]' % self.daemonize
         )
         # TODO: Get daemonize working on Windows or as a Windows Server.
         if self.daemonize:
             if os.name=='posix':
                 from twisted.scripts._twistd_unix import daemonize
                 daemonize()
 
         dc = ioloop.DelayedCallback(self.start_controller, 0, self.loop)
         dc.start()
         dc = ioloop.DelayedCallback(self.start_engines, 1000*self.delay, self.loop)
         dc.start()
         # Now write the new pid file AFTER our new forked pid is active.
         self.write_pid_file()
         try:
             self.loop.start()
         except KeyboardInterrupt:
             pass
         except zmq.ZMQError as e:
             if e.errno == errno.EINTR:
                 pass
             else:
                 raise
         finally:
             self.remove_pid_file()
 
 base='IPython.parallel.apps.ipclusterapp.IPCluster'
 
 class IPBaseParallelApplication(Application):
     name = u'ipcluster'
     description = _description
     
     subcommands = {'create' : (base+'Create', create_help),
                 'list' : (base+'List', list_help),
                 'start' : (base+'Start', start_help),
                 'stop' : (base+'Stop', stop_help),
                 'engines' : (base+'Engines', engines_help),
     }
     
     # no aliases or flags for parent App
     aliases = Dict()
     flags = Dict()
     
     def start(self):
         if self.subapp is None:
             print "No subcommand specified! Must specify one of: %s"%(self.subcommands.keys())
             print
             self.print_subcommands()
             self.exit(1)
         else:
             return self.subapp.start()
 
 def launch_new_instance():
     """Create and run the IPython cluster."""
     app = IPBaseParallelApplication.instance()
     app.initialize()
     app.start()
 
 
 if __name__ == '__main__':
     launch_new_instance()
 

IPython/parallel/apps/iploggerapp.py

@@ -1,96 +1,96 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 A simple IPython logger application
 """
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 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
 #-----------------------------------------------------------------------------
 
 import os
 import sys
 
 import zmq
 
 from IPython.core.newapplication import ProfileDir
 from IPython.utils.traitlets import Bool, Dict, Unicode
 
 from IPython.parallel.apps.baseapp import (
     BaseParallelApplication,
     base_aliases
 )
 from IPython.parallel.apps.logwatcher import LogWatcher
 
 #-----------------------------------------------------------------------------
 # Module level variables
 #-----------------------------------------------------------------------------
 
 #: The default config file name for this application
 default_config_file_name = u'iplogger_config.py'
 
 _description = """Start an IPython logger for parallel computing.
 
 IPython controllers and engines (and your own processes) can broadcast log messages
 by registering a `zmq.log.handlers.PUBHandler` with the `logging` module. The
 logger can be configured using command line options or using a cluster
 directory. Cluster directories contain config, log and security files and are
 usually located in your ipython directory and named as "cluster_<profile>".
 See the `profile` and `profile_dir` options for details.
 """
 
 
 #-----------------------------------------------------------------------------
 # Main application
 #-----------------------------------------------------------------------------
 aliases = {}
 aliases.update(base_aliases)
 aliases.update(dict(url='LogWatcher.url', topics='LogWatcher.topics'))
 
 class IPLoggerApp(BaseParallelApplication):
 
     name = u'iploggerz'
     description = _description
     config_file_name = Unicode(default_config_file_name)
     
     classes = [LogWatcher, ProfileDir]
     aliases = Dict(aliases)
 
     def initialize(self, argv=None):
         super(IPLoggerApp, self).initialize(argv)
         self.init_watcher()
     
     def init_watcher(self):
         try:
-            self.watcher = LogWatcher(config=self.config, logname=self.log.name)
+            self.watcher = LogWatcher(config=self.config, log=self.log)
         except:
             self.log.error("Couldn't start the LogWatcher", exc_info=True)
             self.exit(1)
         self.log.info("Listening for log messages on %r"%self.watcher.url)
         
 
     def start(self):
         self.watcher.start()
         try:
             self.watcher.loop.start()
         except KeyboardInterrupt:
             self.log.critical("Logging Interrupted, shutting down...\n")
 
 
 def launch_new_instance():
     """Create and run the IPython LogWatcher"""
     app = IPLoggerApp.instance()
     app.initialize()
     app.start()
 
 
 if __name__ == '__main__':
     launch_new_instance()
 

IPython/parallel/apps/launcher.py

@@ -1,1070 +1,1069 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 Facilities for launching IPython processes asynchronously.
 """
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2008-2009  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
 #-----------------------------------------------------------------------------
 
 import copy
 import logging
 import os
 import re
 import stat
 
 # signal imports, handling various platforms, versions
 
 from signal import SIGINT, SIGTERM
 try:
     from signal import SIGKILL
 except ImportError:
     # Windows
     SIGKILL=SIGTERM
 
 try:
     # Windows >= 2.7, 3.2
     from signal import CTRL_C_EVENT as SIGINT
 except ImportError:
     pass
 
 from subprocess import Popen, PIPE, STDOUT
 try:
     from subprocess import check_output
 except ImportError:
     # pre-2.7, define check_output with Popen
     def check_output(*args, **kwargs):
         kwargs.update(dict(stdout=PIPE))
         p = Popen(*args, **kwargs)
         out,err = p.communicate()
         return out
 
 from zmq.eventloop import ioloop
 
-# from IPython.config.configurable import Configurable
+from IPython.config.configurable import Configurable
 from IPython.utils.text import EvalFormatter
 from IPython.utils.traitlets import Any, Int, List, Unicode, Dict, Instance
 from IPython.utils.path import get_ipython_module_path
 from IPython.utils.process import find_cmd, pycmd2argv, FindCmdError
 
-from IPython.parallel.factory import LoggingFactory
-
 from .win32support import forward_read_events
 
 from .winhpcjob import IPControllerTask, IPEngineTask, IPControllerJob, IPEngineSetJob
 
 WINDOWS = os.name == 'nt'
 
 #-----------------------------------------------------------------------------
 # Paths to the kernel apps
 #-----------------------------------------------------------------------------
 
 
 ipcluster_cmd_argv = pycmd2argv(get_ipython_module_path(
     'IPython.parallel.apps.ipclusterapp'
 ))
 
 ipengine_cmd_argv = pycmd2argv(get_ipython_module_path(
     'IPython.parallel.apps.ipengineapp'
 ))
 
 ipcontroller_cmd_argv = pycmd2argv(get_ipython_module_path(
     'IPython.parallel.apps.ipcontrollerapp'
 ))
 
 #-----------------------------------------------------------------------------
 # Base launchers and errors
 #-----------------------------------------------------------------------------
 
 
 class LauncherError(Exception):
     pass
 
 
 class ProcessStateError(LauncherError):
     pass
 
 
 class UnknownStatus(LauncherError):
     pass
 
 
-class BaseLauncher(LoggingFactory):
+class BaseLauncher(Configurable):
     """An asbtraction for starting, stopping and signaling a process."""
 
     # In all of the launchers, the work_dir is where child processes will be
     # run. This will usually be the profile_dir, but may not be. any work_dir
     # passed into the __init__ method will override the config value.
     # This should not be used to set the work_dir for the actual engine
     # and controller. Instead, use their own config files or the
     # controller_args, engine_args attributes of the launchers to add
     # the work_dir option.
     work_dir = Unicode(u'.')
     loop = Instance('zmq.eventloop.ioloop.IOLoop')
+    log = Instance('logging.Logger', ('root',))
     
     start_data = Any()
     stop_data = Any()
     
     def _loop_default(self):
         return ioloop.IOLoop.instance()
 
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(BaseLauncher, self).__init__(work_dir=work_dir, config=config, **kwargs)
         self.state = 'before' # can be before, running, after
         self.stop_callbacks = []
         self.start_data = None
         self.stop_data = None
 
     @property
     def args(self):
         """A list of cmd and args that will be used to start the process.
 
         This is what is passed to :func:`spawnProcess` and the first element
         will be the process name.
         """
         return self.find_args()
 
     def find_args(self):
         """The ``.args`` property calls this to find the args list.
 
         Subcommand should implement this to construct the cmd and args.
         """
         raise NotImplementedError('find_args must be implemented in a subclass')
 
     @property
     def arg_str(self):
         """The string form of the program arguments."""
         return ' '.join(self.args)
 
     @property
     def running(self):
         """Am I running."""
         if self.state == 'running':
             return True
         else:
             return False
 
     def start(self):
         """Start the process.
 
         This must return a deferred that fires with information about the
         process starting (like a pid, job id, etc.).
         """
         raise NotImplementedError('start must be implemented in a subclass')
 
     def stop(self):
         """Stop the process and notify observers of stopping.
 
         This must return a deferred that fires with information about the
         processing stopping, like errors that occur while the process is
         attempting to be shut down. This deferred won't fire when the process
         actually stops. To observe the actual process stopping, see
         :func:`observe_stop`.
         """
         raise NotImplementedError('stop must be implemented in a subclass')
 
     def on_stop(self, f):
         """Get a deferred that will fire when the process stops.
 
         The deferred will fire with data that contains information about
         the exit status of the process.
         """
         if self.state=='after':
             return f(self.stop_data)
         else:
             self.stop_callbacks.append(f)
 
     def notify_start(self, data):
         """Call this to trigger startup actions.
 
         This logs the process startup and sets the state to 'running'.  It is
         a pass-through so it can be used as a callback.
         """
 
         self.log.info('Process %r started: %r' % (self.args[0], data))
         self.start_data = data
         self.state = 'running'
         return data
 
     def notify_stop(self, data):
         """Call this to trigger process stop actions.
 
         This logs the process stopping and sets the state to 'after'. Call
         this to trigger all the deferreds from :func:`observe_stop`."""
 
         self.log.info('Process %r stopped: %r' % (self.args[0], data))
         self.stop_data = data
         self.state = 'after'
         for i in range(len(self.stop_callbacks)):
             d = self.stop_callbacks.pop()
             d(data)
         return data
 
     def signal(self, sig):
         """Signal the process.
 
         Return a semi-meaningless deferred after signaling the process.
 
         Parameters
         ----------
         sig : str or int
             'KILL', 'INT', etc., or any signal number
         """
         raise NotImplementedError('signal must be implemented in a subclass')
 
 
 #-----------------------------------------------------------------------------
 # Local process launchers
 #-----------------------------------------------------------------------------
 
 
 class LocalProcessLauncher(BaseLauncher):
     """Start and stop an external process in an asynchronous manner.
 
     This will launch the external process with a working directory of
     ``self.work_dir``.
     """
 
     # This is used to to construct self.args, which is passed to 
     # spawnProcess.
     cmd_and_args = List([])
     poll_frequency = Int(100) # in ms
 
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(LocalProcessLauncher, self).__init__(
             work_dir=work_dir, config=config, **kwargs
         )
         self.process = None
         self.start_deferred = None
         self.poller = None
 
     def find_args(self):
         return self.cmd_and_args
 
     def start(self):
         if self.state == 'before':
             self.process = Popen(self.args,
                 stdout=PIPE,stderr=PIPE,stdin=PIPE,
                 env=os.environ,
                 cwd=self.work_dir
             )
             if WINDOWS:
                 self.stdout = forward_read_events(self.process.stdout)
                 self.stderr = forward_read_events(self.process.stderr)
             else:
                 self.stdout = self.process.stdout.fileno()
                 self.stderr = self.process.stderr.fileno()
             self.loop.add_handler(self.stdout, self.handle_stdout, self.loop.READ)
             self.loop.add_handler(self.stderr, self.handle_stderr, self.loop.READ)
             self.poller = ioloop.PeriodicCallback(self.poll, self.poll_frequency, self.loop)
             self.poller.start()
             self.notify_start(self.process.pid)
         else:
             s = 'The process was already started and has state: %r' % self.state
             raise ProcessStateError(s)
 
     def stop(self):
         return self.interrupt_then_kill()
 
     def signal(self, sig):
         if self.state == 'running':
             if WINDOWS and sig != SIGINT:
                 # use Windows tree-kill for better child cleanup
                 check_output(['taskkill', '-pid', str(self.process.pid), '-t', '-f'])
             else:
                 self.process.send_signal(sig)
 
     def interrupt_then_kill(self, delay=2.0):
         """Send INT, wait a delay and then send KILL."""
         try:
             self.signal(SIGINT)
         except Exception:
             self.log.debug("interrupt failed")
             pass
         self.killer  = ioloop.DelayedCallback(lambda : self.signal(SIGKILL), delay*1000, self.loop)
         self.killer.start()
 
     # callbacks, etc:
     
     def handle_stdout(self, fd, events):
         if WINDOWS:
             line = self.stdout.recv()
         else:
             line = self.process.stdout.readline()
         # a stopped process will be readable but return empty strings
         if line:
             self.log.info(line[:-1])
         else:
             self.poll()
     
     def handle_stderr(self, fd, events):
         if WINDOWS:
             line = self.stderr.recv()
         else:
             line = self.process.stderr.readline()
         # a stopped process will be readable but return empty strings
         if line:
             self.log.error(line[:-1])
         else:
             self.poll()
     
     def poll(self):
         status = self.process.poll()
         if status is not None:
             self.poller.stop()
             self.loop.remove_handler(self.stdout)
             self.loop.remove_handler(self.stderr)
             self.notify_stop(dict(exit_code=status, pid=self.process.pid))
         return status
 
 class LocalControllerLauncher(LocalProcessLauncher):
     """Launch a controller as a regular external process."""
 
     controller_cmd = List(ipcontroller_cmd_argv, config=True,
         help="""Popen command to launch ipcontroller.""")
     # Command line arguments to ipcontroller.
     controller_args = List(['--log-to-file','log_level=%i'%logging.INFO], config=True,
         help="""command-line args to pass to ipcontroller""")
 
     def find_args(self):
         return self.controller_cmd + self.controller_args
 
     def start(self, profile_dir):
         """Start the controller by profile_dir."""
         self.controller_args.extend(['profile_dir=%s'%profile_dir])
         self.profile_dir = unicode(profile_dir)
         self.log.info("Starting LocalControllerLauncher: %r" % self.args)
         return super(LocalControllerLauncher, self).start()
 
 
 class LocalEngineLauncher(LocalProcessLauncher):
     """Launch a single engine as a regular externall process."""
 
     engine_cmd = List(ipengine_cmd_argv, config=True,
         help="""command to launch the Engine.""")
     # Command line arguments for ipengine.
     engine_args = List(['--log-to-file','log_level=%i'%logging.INFO], config=True,
         help="command-line arguments to pass to ipengine"
     )
 
     def find_args(self):
         return self.engine_cmd + self.engine_args
 
     def start(self, profile_dir):
         """Start the engine by profile_dir."""
         self.engine_args.extend(['profile_dir=%s'%profile_dir])
         self.profile_dir = unicode(profile_dir)
         return super(LocalEngineLauncher, self).start()
 
 
 class LocalEngineSetLauncher(BaseLauncher):
     """Launch a set of engines as regular external processes."""
 
     # Command line arguments for ipengine.
     engine_args = List(
         ['--log-to-file','log_level=%i'%logging.INFO], config=True,
         help="command-line arguments to pass to ipengine"
     )
     # launcher class
     launcher_class = LocalEngineLauncher
     
     launchers = Dict()
     stop_data = Dict()
     
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(LocalEngineSetLauncher, self).__init__(
             work_dir=work_dir, config=config, **kwargs
         )
         self.stop_data = {}
 
     def start(self, n, profile_dir):
         """Start n engines by profile or profile_dir."""
         self.profile_dir = unicode(profile_dir)
         dlist = []
         for i in range(n):
-            el = self.launcher_class(work_dir=self.work_dir, config=self.config, logname=self.log.name)
+            el = self.launcher_class(work_dir=self.work_dir, config=self.config, log=self.log)
             # Copy the engine args over to each engine launcher.
             el.engine_args = copy.deepcopy(self.engine_args)
             el.on_stop(self._notice_engine_stopped)
             d = el.start(profile_dir)
             if i==0:
                 self.log.info("Starting LocalEngineSetLauncher: %r" % el.args)
             self.launchers[i] = el
             dlist.append(d)
         self.notify_start(dlist)
         # The consumeErrors here could be dangerous
         # dfinal = gatherBoth(dlist, consumeErrors=True)
         # dfinal.addCallback(self.notify_start)
         return dlist
 
     def find_args(self):
         return ['engine set']
 
     def signal(self, sig):
         dlist = []
         for el in self.launchers.itervalues():
             d = el.signal(sig)
             dlist.append(d)
         # dfinal = gatherBoth(dlist, consumeErrors=True)
         return dlist
 
     def interrupt_then_kill(self, delay=1.0):
         dlist = []
         for el in self.launchers.itervalues():
             d = el.interrupt_then_kill(delay)
             dlist.append(d)
         # dfinal = gatherBoth(dlist, consumeErrors=True)
         return dlist
 
     def stop(self):
         return self.interrupt_then_kill()
     
     def _notice_engine_stopped(self, data):
         pid = data['pid']
         for idx,el in self.launchers.iteritems():
             if el.process.pid == pid:
                 break
         self.launchers.pop(idx)
         self.stop_data[idx] = data
         if not self.launchers:
             self.notify_stop(self.stop_data)
 
 
 #-----------------------------------------------------------------------------
 # MPIExec launchers
 #-----------------------------------------------------------------------------
 
 
 class MPIExecLauncher(LocalProcessLauncher):
     """Launch an external process using mpiexec."""
 
     mpi_cmd = List(['mpiexec'], config=True,
         help="The mpiexec command to use in starting the process."
     )
     mpi_args = List([], config=True,
         help="The command line arguments to pass to mpiexec."
     )
     program = List(['date'], config=True,
         help="The program to start via mpiexec.")
     program_args = List([], config=True,
         help="The command line argument to the program."
     )
     n = Int(1)
 
     def find_args(self):
         """Build self.args using all the fields."""
         return self.mpi_cmd + ['-n', str(self.n)] + self.mpi_args + \
                self.program + self.program_args
 
     def start(self, n):
         """Start n instances of the program using mpiexec."""
         self.n = n
         return super(MPIExecLauncher, self).start()
 
 
 class MPIExecControllerLauncher(MPIExecLauncher):
     """Launch a controller using mpiexec."""
 
     controller_cmd = List(ipcontroller_cmd_argv, config=True,
         help="Popen command to launch the Contropper"
     )
     controller_args = List(['--log-to-file','log_level=%i'%logging.INFO], config=True,
         help="Command line arguments to pass to ipcontroller."
     )
     n = Int(1)
 
     def start(self, profile_dir):
         """Start the controller by profile_dir."""
         self.controller_args.extend(['profile_dir=%s'%profile_dir])
         self.profile_dir = unicode(profile_dir)
         self.log.info("Starting MPIExecControllerLauncher: %r" % self.args)
         return super(MPIExecControllerLauncher, self).start(1)
 
     def find_args(self):
         return self.mpi_cmd + ['-n', self.n] + self.mpi_args + \
                self.controller_cmd + self.controller_args
 
 
 class MPIExecEngineSetLauncher(MPIExecLauncher):
 
     program = List(ipengine_cmd_argv, config=True,
         help="Popen command for ipengine"
     )
     program_args = List(
         ['--log-to-file','log_level=%i'%logging.INFO], config=True,
         help="Command line arguments for ipengine."
     )
     n = Int(1)
 
     def start(self, n, profile_dir):
         """Start n engines by profile or profile_dir."""
         self.program_args.extend(['profile_dir=%s'%profile_dir])
         self.profile_dir = unicode(profile_dir)
         self.n = n
         self.log.info('Starting MPIExecEngineSetLauncher: %r' % self.args)
         return super(MPIExecEngineSetLauncher, self).start(n)
 
 #-----------------------------------------------------------------------------
 # SSH launchers
 #-----------------------------------------------------------------------------
 
 # TODO: Get SSH Launcher working again.
 
 class SSHLauncher(LocalProcessLauncher):
     """A minimal launcher for ssh.
 
     To be useful this will probably have to be extended to use the ``sshx``
     idea for environment variables.  There could be other things this needs
     as well.
     """
 
     ssh_cmd = List(['ssh'], config=True,
         help="command for starting ssh")
     ssh_args = List(['-tt'], config=True,
         help="args to pass to ssh")
     program = List(['date'], config=True,
         help="Program to launch via ssh")
     program_args = List([], config=True,
         help="args to pass to remote program")
     hostname = Unicode('', config=True,
         help="hostname on which to launch the program")
     user = Unicode('', config=True,
         help="username for ssh")
     location = Unicode('', config=True,
         help="user@hostname location for ssh in one setting")
 
     def _hostname_changed(self, name, old, new):
         if self.user:
             self.location = u'%s@%s' % (self.user, new)
         else:
             self.location = new
 
     def _user_changed(self, name, old, new):
         self.location = u'%s@%s' % (new, self.hostname)
 
     def find_args(self):
         return self.ssh_cmd + self.ssh_args + [self.location] + \
                self.program + self.program_args
 
     def start(self, profile_dir, hostname=None, user=None):
         self.profile_dir = unicode(profile_dir)
         if hostname is not None:
             self.hostname = hostname
         if user is not None:
             self.user = user
         
         return super(SSHLauncher, self).start()
     
     def signal(self, sig):
         if self.state == 'running':
             # send escaped ssh connection-closer
             self.process.stdin.write('~.')
             self.process.stdin.flush()
         
 
 
 class SSHControllerLauncher(SSHLauncher):
 
     program = List(ipcontroller_cmd_argv, config=True,
         help="remote ipcontroller command.")
     program_args = List(['--reuse-files', '--log-to-file','log_level=%i'%logging.INFO], config=True,
         help="Command line arguments to ipcontroller.")
 
 
 class SSHEngineLauncher(SSHLauncher):
     program = List(ipengine_cmd_argv, config=True,
         help="remote ipengine command.")
     # Command line arguments for ipengine.
     program_args = List(
         ['--log-to-file','log_level=%i'%logging.INFO], config=True,
         help="Command line arguments to ipengine."
     )
     
 class SSHEngineSetLauncher(LocalEngineSetLauncher):
     launcher_class = SSHEngineLauncher
     engines = Dict(config=True,
         help="""dict of engines to launch.  This is a dict by hostname of ints,
         corresponding to the number of engines to start on that host.""")
     
     def start(self, n, profile_dir):
         """Start engines by profile or profile_dir.
         `n` is ignored, and the `engines` config property is used instead.
         """
         
         self.profile_dir = unicode(profile_dir)
         dlist = []
         for host, n in self.engines.iteritems():
             if isinstance(n, (tuple, list)):
                 n, args = n
             else:
                 args = copy.deepcopy(self.engine_args)
             
             if '@' in host:
                 user,host = host.split('@',1)
             else:
                 user=None
             for i in range(n):
-                el = self.launcher_class(work_dir=self.work_dir, config=self.config, logname=self.log.name)
+                el = self.launcher_class(work_dir=self.work_dir, config=self.config, log=self.log)
                 
                 # Copy the engine args over to each engine launcher.
                 i
                 el.program_args = args
                 el.on_stop(self._notice_engine_stopped)
                 d = el.start(profile_dir, user=user, hostname=host)
                 if i==0:
                     self.log.info("Starting SSHEngineSetLauncher: %r" % el.args)
                 self.launchers[host+str(i)] = el
                 dlist.append(d)
         self.notify_start(dlist)
         return dlist
     
 
 
 #-----------------------------------------------------------------------------
 # Windows HPC Server 2008 scheduler launchers
 #-----------------------------------------------------------------------------
 
 
 # This is only used on Windows.
 def find_job_cmd():
     if WINDOWS:
         try:
             return find_cmd('job')
         except (FindCmdError, ImportError):
             # ImportError will be raised if win32api is not installed
             return 'job'
     else:
         return 'job'
 
 
 class WindowsHPCLauncher(BaseLauncher):
 
     job_id_regexp = Unicode(r'\d+', config=True,
         help="""A regular expression used to get the job id from the output of the
         submit_command. """
         )
     job_file_name = Unicode(u'ipython_job.xml', config=True,
         help="The filename of the instantiated job script.")
     # The full path to the instantiated job script. This gets made dynamically
     # by combining the work_dir with the job_file_name.
     job_file = Unicode(u'')
     scheduler = Unicode('', config=True,
         help="The hostname of the scheduler to submit the job to.")
     job_cmd = Unicode(find_job_cmd(), config=True,
         help="The command for submitting jobs.")
 
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(WindowsHPCLauncher, self).__init__(
             work_dir=work_dir, config=config, **kwargs
         )
 
     @property
     def job_file(self):
         return os.path.join(self.work_dir, self.job_file_name)
 
     def write_job_file(self, n):
         raise NotImplementedError("Implement write_job_file in a subclass.")
 
     def find_args(self):
         return [u'job.exe']
         
     def parse_job_id(self, output):
         """Take the output of the submit command and return the job id."""
         m = re.search(self.job_id_regexp, output)
         if m is not None:
             job_id = m.group()
         else:
             raise LauncherError("Job id couldn't be determined: %s" % output)
         self.job_id = job_id
         self.log.info('Job started with job id: %r' % job_id)
         return job_id
 
     def start(self, n):
         """Start n copies of the process using the Win HPC job scheduler."""
         self.write_job_file(n)
         args = [
             'submit',
             '/jobfile:%s' % self.job_file,
             '/scheduler:%s' % self.scheduler
         ]
         self.log.info("Starting Win HPC Job: %s" % (self.job_cmd + ' ' + ' '.join(args),))
         # Twisted will raise DeprecationWarnings if we try to pass unicode to this
         output = check_output([self.job_cmd]+args,
             env=os.environ,
             cwd=self.work_dir,
             stderr=STDOUT
         )
         job_id = self.parse_job_id(output)
         self.notify_start(job_id)
         return job_id
 
     def stop(self):
         args = [
             'cancel',
             self.job_id,
             '/scheduler:%s' % self.scheduler
         ]
         self.log.info("Stopping Win HPC Job: %s" % (self.job_cmd + ' ' + ' '.join(args),))
         try:
             output = check_output([self.job_cmd]+args,
                 env=os.environ,
                 cwd=self.work_dir,
                 stderr=STDOUT
             )
         except:
             output = 'The job already appears to be stoppped: %r' % self.job_id
         self.notify_stop(dict(job_id=self.job_id, output=output))  # Pass the output of the kill cmd
         return output
 
 
 class WindowsHPCControllerLauncher(WindowsHPCLauncher):
 
     job_file_name = Unicode(u'ipcontroller_job.xml', config=True,
         help="WinHPC xml job file.")
     extra_args = List([], config=False,
         help="extra args to pass to ipcontroller")
 
     def write_job_file(self, n):
         job = IPControllerJob(config=self.config)
 
         t = IPControllerTask(config=self.config)
         # The tasks work directory is *not* the actual work directory of 
         # the controller. It is used as the base path for the stdout/stderr
         # files that the scheduler redirects to.
         t.work_directory = self.profile_dir
         # Add the profile_dir and from self.start().
         t.controller_args.extend(self.extra_args)
         job.add_task(t)
 
         self.log.info("Writing job description file: %s" % self.job_file)
         job.write(self.job_file)
 
     @property
     def job_file(self):
         return os.path.join(self.profile_dir, self.job_file_name)
 
     def start(self, profile_dir):
         """Start the controller by profile_dir."""
         self.extra_args = ['profile_dir=%s'%profile_dir]
         self.profile_dir = unicode(profile_dir)
         return super(WindowsHPCControllerLauncher, self).start(1)
 
 
 class WindowsHPCEngineSetLauncher(WindowsHPCLauncher):
 
     job_file_name = Unicode(u'ipengineset_job.xml', config=True,
         help="jobfile for ipengines job")
     extra_args = List([], config=False,
         help="extra args to pas to ipengine")
 
     def write_job_file(self, n):
         job = IPEngineSetJob(config=self.config)
 
         for i in range(n):
             t = IPEngineTask(config=self.config)
             # The tasks work directory is *not* the actual work directory of 
             # the engine. It is used as the base path for the stdout/stderr
             # files that the scheduler redirects to.
             t.work_directory = self.profile_dir
             # Add the profile_dir and from self.start().
             t.engine_args.extend(self.extra_args)
             job.add_task(t)
 
         self.log.info("Writing job description file: %s" % self.job_file)
         job.write(self.job_file)
 
     @property
     def job_file(self):
         return os.path.join(self.profile_dir, self.job_file_name)
 
     def start(self, n, profile_dir):
         """Start the controller by profile_dir."""
         self.extra_args = ['profile_dir=%s'%profile_dir]
         self.profile_dir = unicode(profile_dir)
         return super(WindowsHPCEngineSetLauncher, self).start(n)
 
 
 #-----------------------------------------------------------------------------
 # Batch (PBS) system launchers
 #-----------------------------------------------------------------------------
 
 class BatchSystemLauncher(BaseLauncher):
     """Launch an external process using a batch system.
 
     This class is designed to work with UNIX batch systems like PBS, LSF,
     GridEngine, etc.  The overall model is that there are different commands
     like qsub, qdel, etc. that handle the starting and stopping of the process.
 
     This class also has the notion of a batch script. The ``batch_template``
     attribute can be set to a string that is a template for the batch script.
     This template is instantiated using string formatting. Thus the template can
     use {n} fot the number of instances. Subclasses can add additional variables
     to the template dict.
     """
 
     # Subclasses must fill these in.  See PBSEngineSet
     submit_command = List([''], config=True,
         help="The name of the command line program used to submit jobs.")
     delete_command = List([''], config=True,
         help="The name of the command line program used to delete jobs.")
     job_id_regexp = Unicode('', config=True,
         help="""A regular expression used to get the job id from the output of the
         submit_command.""")
     batch_template = Unicode('', config=True,
         help="The string that is the batch script template itself.")
     batch_template_file = Unicode(u'', config=True,
         help="The file that contains the batch template.")
     batch_file_name = Unicode(u'batch_script', config=True,
         help="The filename of the instantiated batch script.")
     queue = Unicode(u'', config=True,
         help="The PBS Queue.")
     
     # not configurable, override in subclasses
     # PBS Job Array regex
     job_array_regexp = Unicode('')
     job_array_template = Unicode('')
     # PBS Queue regex
     queue_regexp = Unicode('')
     queue_template = Unicode('')
     # The default batch template, override in subclasses
     default_template = Unicode('')
     # The full path to the instantiated batch script.
     batch_file = Unicode(u'')
     # the format dict used with batch_template:
     context = Dict()
     # the Formatter instance for rendering the templates:
     formatter = Instance(EvalFormatter, (), {})
 
     
     def find_args(self):
         return self.submit_command + [self.batch_file]
     
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(BatchSystemLauncher, self).__init__(
             work_dir=work_dir, config=config, **kwargs
         )
         self.batch_file = os.path.join(self.work_dir, self.batch_file_name)
 
     def parse_job_id(self, output):
         """Take the output of the submit command and return the job id."""
         m = re.search(self.job_id_regexp, output)
         if m is not None:
             job_id = m.group()
         else:
             raise LauncherError("Job id couldn't be determined: %s" % output)
         self.job_id = job_id
         self.log.info('Job submitted with job id: %r' % job_id)
         return job_id
 
     def write_batch_script(self, n):
         """Instantiate and write the batch script to the work_dir."""
         self.context['n'] = n
         self.context['queue'] = self.queue
         # first priority is batch_template if set
         if self.batch_template_file and not self.batch_template:
             # second priority is batch_template_file
             with open(self.batch_template_file) as f:
                 self.batch_template = f.read()
         if not self.batch_template:
             # third (last) priority is default_template
             self.batch_template = self.default_template
         
         regex = re.compile(self.job_array_regexp)
         # print regex.search(self.batch_template)
         if not regex.search(self.batch_template):
             self.log.info("adding job array settings to batch script")
             firstline, rest = self.batch_template.split('\n',1)
             self.batch_template = u'\n'.join([firstline, self.job_array_template, rest])
         
         regex = re.compile(self.queue_regexp)
         # print regex.search(self.batch_template)
         if self.queue and not regex.search(self.batch_template):
             self.log.info("adding PBS queue settings to batch script")
             firstline, rest = self.batch_template.split('\n',1)
             self.batch_template = u'\n'.join([firstline, self.queue_template, rest])
             
         script_as_string = self.formatter.format(self.batch_template, **self.context)
         self.log.info('Writing instantiated batch script: %s' % self.batch_file)
     
         with open(self.batch_file, 'w') as f:
             f.write(script_as_string)
         os.chmod(self.batch_file, stat.S_IRUSR | stat.S_IWUSR | stat.S_IXUSR)
 
     def start(self, n, profile_dir):
         """Start n copies of the process using a batch system."""
         # Here we save profile_dir in the context so they
         # can be used in the batch script template as {profile_dir}
         self.context['profile_dir'] = profile_dir
         self.profile_dir = unicode(profile_dir)
         self.write_batch_script(n)
         output = check_output(self.args, env=os.environ)
         
         job_id = self.parse_job_id(output)
         self.notify_start(job_id)
         return job_id
 
     def stop(self):
         output = check_output(self.delete_command+[self.job_id], env=os.environ)
         self.notify_stop(dict(job_id=self.job_id, output=output)) # Pass the output of the kill cmd
         return output
 
 
 class PBSLauncher(BatchSystemLauncher):
     """A BatchSystemLauncher subclass for PBS."""
 
     submit_command = List(['qsub'], config=True,
         help="The PBS submit command ['qsub']")
     delete_command = List(['qdel'], config=True,
         help="The PBS delete command ['qsub']")
     job_id_regexp = Unicode(r'\d+', config=True,
         help="Regular expresion for identifying the job ID [r'\d+']")
     
     batch_file = Unicode(u'')
     job_array_regexp = Unicode('#PBS\W+-t\W+[\w\d\-\$]+')
     job_array_template = Unicode('#PBS -t 1-{n}')
     queue_regexp = Unicode('#PBS\W+-q\W+\$?\w+')
     queue_template = Unicode('#PBS -q {queue}')
 
 
 class PBSControllerLauncher(PBSLauncher):
     """Launch a controller using PBS."""
 
     batch_file_name = Unicode(u'pbs_controller', config=True,
         help="batch file name for the controller job.")
     default_template= Unicode("""#!/bin/sh
 #PBS -V
 #PBS -N ipcontroller
 %s --log-to-file profile_dir={profile_dir}
 """%(' '.join(ipcontroller_cmd_argv)))
 
     def start(self, profile_dir):
         """Start the controller by profile or profile_dir."""
         self.log.info("Starting PBSControllerLauncher: %r" % self.args)
         return super(PBSControllerLauncher, self).start(1, profile_dir)
 
 
 class PBSEngineSetLauncher(PBSLauncher):
     """Launch Engines using PBS"""
     batch_file_name = Unicode(u'pbs_engines', config=True,
         help="batch file name for the engine(s) job.")
     default_template= Unicode(u"""#!/bin/sh
 #PBS -V
 #PBS -N ipengine
 %s profile_dir={profile_dir}
 """%(' '.join(ipengine_cmd_argv)))
 
     def start(self, n, profile_dir):
         """Start n engines by profile or profile_dir."""
         self.log.info('Starting %i engines with PBSEngineSetLauncher: %r' % (n, self.args))
         return super(PBSEngineSetLauncher, self).start(n, profile_dir)
 
 #SGE is very similar to PBS
 
 class SGELauncher(PBSLauncher):
     """Sun GridEngine is a PBS clone with slightly different syntax"""
     job_array_regexp = Unicode('#\$\W+\-t')
     job_array_template = Unicode('#$ -t 1-{n}')
     queue_regexp = Unicode('#\$\W+-q\W+\$?\w+')
     queue_template = Unicode('#$ -q $queue')
 
 class SGEControllerLauncher(SGELauncher):
     """Launch a controller using SGE."""
 
     batch_file_name = Unicode(u'sge_controller', config=True,
         help="batch file name for the ipontroller job.")
     default_template= Unicode(u"""#$ -V
 #$ -S /bin/sh
 #$ -N ipcontroller
 %s --log-to-file profile_dir={profile_dir}
 """%(' '.join(ipcontroller_cmd_argv)))
 
     def start(self, profile_dir):
         """Start the controller by profile or profile_dir."""
         self.log.info("Starting PBSControllerLauncher: %r" % self.args)
         return super(SGEControllerLauncher, self).start(1, profile_dir)
 
 class SGEEngineSetLauncher(SGELauncher):
     """Launch Engines with SGE"""
     batch_file_name = Unicode(u'sge_engines', config=True,
         help="batch file name for the engine(s) job.")
     default_template = Unicode("""#$ -V
 #$ -S /bin/sh
 #$ -N ipengine
 %s profile_dir={profile_dir}
 """%(' '.join(ipengine_cmd_argv)))
 
     def start(self, n, profile_dir):
         """Start n engines by profile or profile_dir."""
         self.log.info('Starting %i engines with SGEEngineSetLauncher: %r' % (n, self.args))
         return super(SGEEngineSetLauncher, self).start(n, profile_dir)
 
 
 #-----------------------------------------------------------------------------
 # A launcher for ipcluster itself!
 #-----------------------------------------------------------------------------
 
 
 class IPClusterLauncher(LocalProcessLauncher):
     """Launch the ipcluster program in an external process."""
 
     ipcluster_cmd = List(ipcluster_cmd_argv, config=True,
         help="Popen command for ipcluster")
     ipcluster_args = List(
         ['--clean-logs', '--log-to-file', 'log_level=%i'%logging.INFO], config=True,
         help="Command line arguments to pass to ipcluster.")
     ipcluster_subcommand = Unicode('start')
     ipcluster_n = Int(2)
 
     def find_args(self):
         return self.ipcluster_cmd + ['--'+self.ipcluster_subcommand] + \
             ['n=%i'%self.ipcluster_n] + self.ipcluster_args
 
     def start(self):
         self.log.info("Starting ipcluster: %r" % self.args)
         return super(IPClusterLauncher, self).start()
 
 #-----------------------------------------------------------------------------
 # Collections of launchers
 #-----------------------------------------------------------------------------
 
 local_launchers = [
     LocalControllerLauncher,
     LocalEngineLauncher,
     LocalEngineSetLauncher,
 ]
 mpi_launchers = [
     MPIExecLauncher,
     MPIExecControllerLauncher,
     MPIExecEngineSetLauncher,
 ]
 ssh_launchers = [
     SSHLauncher,
     SSHControllerLauncher,
     SSHEngineLauncher,
     SSHEngineSetLauncher,
 ]
 winhpc_launchers = [
     WindowsHPCLauncher,
     WindowsHPCControllerLauncher,
     WindowsHPCEngineSetLauncher,
 ]
 pbs_launchers = [
     PBSLauncher,
     PBSControllerLauncher,
     PBSEngineSetLauncher,
 ]
 sge_launchers = [
     SGELauncher,
     SGEControllerLauncher,
     SGEEngineSetLauncher,
 ]
 all_launchers = local_launchers + mpi_launchers + ssh_launchers + winhpc_launchers\
                 + pbs_launchers + sge_launchers

IPython/parallel/apps/logwatcher.py

@@ -1,108 +1,110 b''
 #!/usr/bin/env python
 """A simple logger object that consolidates messages incoming from ipcluster processes."""
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 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
 #-----------------------------------------------------------------------------
 
 
 import logging
 import sys
 
 import zmq
 from zmq.eventloop import ioloop, zmqstream
 
+from IPython.config.configurable import Configurable
 from IPython.utils.traitlets import Int, Unicode, Instance, List
 
-from IPython.parallel.factory import LoggingFactory
-
 #-----------------------------------------------------------------------------
 # Classes
 #-----------------------------------------------------------------------------
 
 
-class LogWatcher(LoggingFactory):
+class LogWatcher(Configurable):
     """A simple class that receives messages on a SUB socket, as published
     by subclasses of `zmq.log.handlers.PUBHandler`, and logs them itself.
     
     This can subscribe to multiple topics, but defaults to all topics.
     """
+    
+    log = Instance('logging.Logger', ('root',))
+    
     # configurables
     topics = List([''], config=True,
         help="The ZMQ topics to subscribe to. Default is to subscribe to all messages")
     url = Unicode('tcp://127.0.0.1:20202', config=True,
         help="ZMQ url on which to listen for log messages")
     
     # internals
     stream = Instance('zmq.eventloop.zmqstream.ZMQStream')
     
     context = Instance(zmq.Context)
     def _context_default(self):
         return zmq.Context.instance()
     
     loop = Instance(zmq.eventloop.ioloop.IOLoop)
     def _loop_default(self):
         return ioloop.IOLoop.instance()
     
     def __init__(self, **kwargs):
         super(LogWatcher, self).__init__(**kwargs)
         s = self.context.socket(zmq.SUB)
         s.bind(self.url)
         self.stream = zmqstream.ZMQStream(s, self.loop)
         self.subscribe()
         self.on_trait_change(self.subscribe, 'topics')
     
     def start(self):
         self.stream.on_recv(self.log_message)
     
     def stop(self):
         self.stream.stop_on_recv()
     
     def subscribe(self):
         """Update our SUB socket's subscriptions."""
         self.stream.setsockopt(zmq.UNSUBSCRIBE, '')
         if '' in self.topics:
             self.log.debug("Subscribing to: everything")
             self.stream.setsockopt(zmq.SUBSCRIBE, '')
         else:
             for topic in self.topics:
                 self.log.debug("Subscribing to: %r"%(topic))
                 self.stream.setsockopt(zmq.SUBSCRIBE, topic)
     
     def _extract_level(self, topic_str):
         """Turn 'engine.0.INFO.extra' into (logging.INFO, 'engine.0.extra')"""
         topics = topic_str.split('.')
         for idx,t in enumerate(topics):
             level = getattr(logging, t, None)
             if level is not None:
                 break
         
         if level is None:
             level = logging.INFO
         else:
             topics.pop(idx)
         
         return level, '.'.join(topics)
             
             
     def log_message(self, raw):
         """receive and parse a message, then log it."""
         if len(raw) != 2 or '.' not in raw[0]:
             self.log.error("Invalid log message: %s"%raw)
             return
         else:
             topic, msg = raw
             # don't newline, since log messages always newline:
             topic,level_name = topic.rsplit('.',1)
             level,topic = self._extract_level(topic)
             if msg[-1] == '\n':
                 msg = msg[:-1]
             self.log.log(level, "[%s] %s" % (topic, msg))
 

IPython/parallel/controller/heartmonitor.py

@@ -1,165 +1,166 b''
 #!/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.
 """
 #-----------------------------------------------------------------------------
 #  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 Configurable
 from IPython.utils.traitlets import Set, Instance, CFloat
-from IPython.parallel.factory import LoggingFactory
 
 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.XREQ, 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, "")
         if heart_id is None:
             heart_id = str(uuid.uuid4())
         self.device.setsockopt_out(zmq.IDENTITY, heart_id)
         self.id = heart_id
     
     def start(self):
         return self.device.start()
         
-class HeartMonitor(LoggingFactory):
+class HeartMonitor(Configurable):
     """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) [default: 100]',
     )
     
+    log = Instance('logging.Logger', ('root',))
     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.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._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._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)
         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.pingstream.send(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.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)
                     pass
         else:
             self.log.info("heartbeat::Heart %s failed :("%heart)
         self.hearts.remove(heart)
         
     
     def handle_pong(self, msg):
         "a heart just beat"
         if msg[1] == str(self.lifetime):
             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] == str(self.last_ping):
             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.responses.add(msg[0])
         else:
             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.XREP)
     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

@@ -1,1274 +1,1277 b''
 #!/usr/bin/env python
 """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.
 """
 #-----------------------------------------------------------------------------
 #  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
 
 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, Int, Unicode, Dict, Set, Tuple, CStr
         )
 from IPython.utils.jsonutil import ISO8601, extract_dates
 
 from IPython.parallel import error, util
-from IPython.parallel.factory import RegistrationFactory, LoggingFactory
+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 = extract_dates(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=Int(0)
     queue=CStr()
     control=CStr()
     registration=CStr()
     heartbeat=CStr()
     pending=Set()
 
 class HubFactory(RegistrationFactory):
     """The Configurable for setting up a Hub."""
     
     # port-pairs for monitoredqueues:
     hb = Tuple(Int,Int,config=True,
         help="""XREQ/SUB Port pair for Engine heartbeats""")
     def _hb_default(self):
         return tuple(util.select_random_ports(2))
 
     mux = Tuple(Int,Int,config=True,
         help="""Engine/Client Port pair for MUX queue""")
     
     def _mux_default(self):
         return tuple(util.select_random_ports(2))
     
     task = Tuple(Int,Int,config=True,
         help="""Engine/Client Port pair for Task queue""")
     def _task_default(self):
         return tuple(util.select_random_ports(2))
 
     control = Tuple(Int,Int,config=True,
         help="""Engine/Client Port pair for Control queue""")
     
     def _control_default(self):
         return tuple(util.select_random_ports(2))
 
     iopub = Tuple(Int,Int,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 = Int(config=True,
         help="""Monitor (SUB) port for queue traffic""")
 
     def _mon_port_default(self):
         return util.select_random_ports(1)[0]
     
     notifier_port = Int(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 = Unicode('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.XREP), loop)
         q.bind(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))
         
         ### Engine connections ###
 
         # heartbeat
         hpub = ctx.socket(zmq.PUB)
         hpub.bind(engine_iface % self.hb[0])
         hrep = ctx.socket(zmq.XREP)
         hrep.bind(engine_iface % self.hb[1])
-        self.heartmonitor = HeartMonitor(loop=loop, pingstream=ZMQStream(hpub,loop), pongstream=ZMQStream(hrep,loop), 
-                                config=self.config)
+        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, "")
         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)
         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)
 
         # resubmit stream
         r = ZMQStream(ctx.socket(zmq.XREQ), loop)
         url = util.disambiguate_url(self.client_info['task'][-1])
         r.setsockopt(zmq.IDENTITY, self.session.session)
         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,
-                logname=self.log.name)
+                log=self.log)
     
 
-class Hub(LoggingFactory):
+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=Int()
     _idcounter=Int(0)
     
     # objects from constructor:
-    loop=Instance(ioloop.IOLoop)
     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 = { '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,
                                 '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])
         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)
             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)
         
     
     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)
             return
         client_id = idents[0]
         try:
             msg = self.session.unpack_message(msg, content=True)
         except Exception:
             content = error.wrap_exception()
             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( idents, msg)
         # print client_id, header, parent, content
         #switch on message type:
         msg_type = msg['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.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)
         if heart not in self.incoming_registrations:
             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)
         eid = self.hearts.get(heart, None)
         queue = self.engines[eid].queue
         if eid is None:
             self.log.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):
         if len(idents) < 2:
             self.log.error("invalid identity prefix: %r"%idents)
             return
         queue_id, client_id = idents[:2]
         try:
             msg = self.session.unpack_message(msg, content=False)
         except Exception:
             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()))
             return
             
         header = msg['header']
         msg_id = header['msg_id']
         record = init_record(msg)
         record['engine_uuid'] = queue_id
         record['client_uuid'] = client_id
         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))
                 elif evalue and not rvalue:
                     record[key] = evalue
             self.db.update_record(msg_id, record)
         except KeyError:
             self.db.add_record(msg_id, record)
         
         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)
             return
             
         client_id, queue_id = idents[:2]
         try:
             msg = self.session.unpack_message(msg, content=False)
         except Exception:
             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)
             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)
         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)
             return
         # update record anyway, because the unregistration could have been premature
         rheader = extract_dates(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)
         
             
     #--------------------- 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 Exception:
             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
         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))
                 elif evalue and not rvalue:
                     record[key] = evalue
             self.db.update_record(msg_id, record)
         except KeyError:
             self.db.add_record(msg_id, record)
         except Exception:
             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.unpack_message(msg, content=False)
         except Exception:
             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)
             return
         msg_id = parent['msg_id']
         if msg_id in self.unassigned:
             self.unassigned.remove(msg_id)
         
         header = extract_dates(msg['header'])
         engine_uuid = header.get('engine', None)
         eid = self.by_ident.get(engine_uuid, None)
         
         if msg_id in self.pending:
             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)
             
         else:
             self.log.debug("task::unknown task %r finished"%msg_id)
     
     def save_task_destination(self, idents, msg):
         try:
             msg = self.session.unpack_message(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[engine_uuid]
         
         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)
             
     
     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.unpack_message(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)
             return
         msg_id = parent['msg_id']
         msg_type = msg['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)
         
     
             
     #-------------------------------------------------------------------------
     # Registration requests
     #-------------------------------------------------------------------------
         
     def connection_request(self, client_id, msg):
         """Reply with connection addresses for clients."""
         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
         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 = content['queue']
         except KeyError:
             self.log.error("registration::queue not specified", exc_info=True)
             return
         heart = content.get('heartbeat', None)
         """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))
         
         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)
         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)
                 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)
                         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)
                         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']))
         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)
             return
         self.log.info("registration::unregister_engine(%r)"%eid)
         # print (eid)
         uuid = self.keytable[eid]
         content=dict(id=eid, queue=uuid)
         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)
                 
     
     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))
         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)
         if self.notifier:
             self.session.send(self.notifier, "registration_notification", content=content)
         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)
         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[bytes(t)] = {'queue': queue, 'completed': completed , 'tasks': tasks}
         content['unassigned'] = list(self.unassigned) if verbose else len(self.unassigned)
         
         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']
         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
                     msg_ids = self.completed.pop(eid)
                     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))
             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
             now_s = now.strftime(ISO8601)
             for rec in records:
                 header = rec['header']
                 # include resubmitted in header to prevent digest collision
                 header['resubmitted'] = now_s
                 msg = self.session.msg(header['msg_type'])
                 msg['content'] = rec['content']
                 msg['header'] = header
                 msg['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(str, 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)
         query = util.extract_dates(query)
         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.session.send(self.query, "db_reply", content=content, 
                                             parent=msg, ident=client_id,
                                             buffers=buffers)
 

IPython/parallel/controller/scheduler.py

@@ -1,687 +1,688 b''
 """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.
 """
 #-----------------------------------------------------------------------------
 #  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.loader import Config
 from IPython.utils.traitlets import Instance, Dict, List, Set, Int, Str, Enum
 
 from IPython.parallel import error
 from IPython.parallel.factory import SessionFactory
 from IPython.parallel.util import connect_logger, local_logger
 
 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 = Int(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')
     
     
     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...%r"%self)
     
     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.unpack_message(msg)
         except ValueError:
             self.log.warn("task::Unauthorized message from: %r"%idents)
             return
         
         msg_type = msg['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(str(msg['content']['queue']))
             except KeyError:
                 self.log.error("task::Invalid notification msg: %r"%msg)
     
     @logged
     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)
 
     
     @logged
     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
     #-----------------------------------------------------------------------
     @logged
     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.unpack_message(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(['intask']+raw_msg, copy=False)
         
         header = msg['header']
         msg_id = header['msg_id']
         self.all_ids.add(msg_id)
         
         # targets
         targets = set(header.get('targets', []))
         retries = header.get('retries', 0)
         self.retries[msg_id] = retries
         
         # time dependencies
         after = Dependency(header.get('after', []))
         if after.all:
             if after.success:
                 after.difference_update(self.all_completed)
             if after.failure:
                 after.difference_update(self.all_failed)
         if after.check(self.all_completed, self.all_failed):
             # recast as empty set, if `after` already met,
             # to prevent unnecessary set comparisons
             after = MET
         
         # location dependencies
         follow = Dependency(header.get('follow', []))
         
         # turn timeouts into datetime objects:
         timeout = header.get('timeout', None)
         if timeout:
             timeout = datetime.now() + timedelta(0,timeout,0)
         
         args = [raw_msg, targets, after, follow, timeout]
         
         # validate and reduce dependencies:
         for dep in after,follow:
             # 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)
     
     # @logged
     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)
                 
     @logged
     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=['outtask']+idents)
         
         self.update_graph(msg_id, success=False)
     
     @logged
     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
             
     @logged
     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)
     
     @logged
     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)
         self.session.send(self.mon_stream, 'task_destination', content=content, 
                         ident=['tracktask',self.session.session])
         
     
     #-----------------------------------------------------------------------
     # Result Handling
     #-----------------------------------------------------------------------
     @logged
     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.unpack_message(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(['outtask']+raw_msg, copy=False)
         else:
             self.handle_unmet_dependency(idents, parent)
         
     @logged
     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)
         
     @logged
     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)
         
         
     
     @logged
     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):
+            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='ZMQ', 
-                            log_url=None, loglevel=logging.DEBUG,
-                            identity=b'task'):
+def launch_scheduler(in_addr, out_addr, mon_addr, not_addr, config=None,
+                        logname='root', log_url=None, loglevel=logging.DEBUG,
+                        identity=b'task'):
     from zmq.eventloop import ioloop
     from zmq.eventloop.zmqstream import ZMQStream
     
     if config:
         # unwrap dict back into Config
         config = Config(config)
 
     ctx = zmq.Context()
     loop = ioloop.IOLoop()
     ins = ZMQStream(ctx.socket(zmq.XREP),loop)
     ins.setsockopt(zmq.IDENTITY, identity)
     ins.bind(in_addr)
     
     outs = ZMQStream(ctx.socket(zmq.XREP),loop)
     outs.setsockopt(zmq.IDENTITY, identity)
     outs.bind(out_addr)
     mons = ZMQStream(ctx.socket(zmq.PUB),loop)
     mons.connect(mon_addr)
     nots = ZMQStream(ctx.socket(zmq.SUB),loop)
     nots.setsockopt(zmq.SUBSCRIBE, '')
     nots.connect(not_addr)
     
     # setup logging. Note that these will not work in-process, because they clobber
     # existing loggers.
     if log_url:
-        connect_logger(logname, ctx, log_url, root="scheduler", loglevel=loglevel)
+        log = connect_logger(logname, ctx, log_url, root="scheduler", loglevel=loglevel)
     else:
-        local_logger(logname, loglevel)
+        log = local_logger(logname, loglevel)
     
     scheduler = TaskScheduler(client_stream=ins, engine_stream=outs,
                             mon_stream=mons, notifier_stream=nots,
-                            loop=loop, logname=logname,
+                            loop=loop, log=log,
                             config=config)
     scheduler.start()
     try:
         loop.start()
     except KeyboardInterrupt:
         print ("interrupted, exiting...", file=sys.__stderr__)
 

IPython/parallel/factory.py

@@ -1,99 +1,72 b''
 """Base config factories."""
 
 #-----------------------------------------------------------------------------
-#  Copyright (C) 2008-2009  The IPython Development Team
+#  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
 #-----------------------------------------------------------------------------
 
 
 import logging
 import os
 
 import zmq
 from zmq.eventloop.ioloop import IOLoop
 
 from IPython.config.configurable import Configurable
 from IPython.utils.traitlets import Int, Instance, Unicode
 
 from IPython.parallel.util import select_random_ports
-from IPython.zmq.session import Session
+from IPython.zmq.session import Session, SessionFactory
 
 #-----------------------------------------------------------------------------
 # Classes
 #-----------------------------------------------------------------------------
-class LoggingFactory(Configurable):
-    """A most basic class, that has a `log` (type:`Logger`) attribute, set via a `logname` Trait."""
-    log = Instance('logging.Logger', ('ZMQ', logging.WARN))
-    logname = Unicode('ZMQ')
-    def _logname_changed(self, name, old, new):
-        self.log = logging.getLogger(new)
-    
 
-class SessionFactory(LoggingFactory):
-    """The Base factory from which every factory in IPython.parallel inherits"""
-    
-    # not configurable:
-    context = Instance('zmq.Context')
-    def _context_default(self):
-        return zmq.Context.instance()
-    
-    session = Instance('IPython.zmq.session.Session')
-    loop = Instance('zmq.eventloop.ioloop.IOLoop', allow_none=False)
-    def _loop_default(self):
-        return IOLoop.instance()
-    
-    
-    def __init__(self, **kwargs):
-        super(SessionFactory, self).__init__(**kwargs)
-        
-        # construct the session
-        self.session = Session(**kwargs)
-    
 
 class RegistrationFactory(SessionFactory):
     """The Base Configurable for objects that involve registration."""
     
     url = Unicode('', config=True,
         help="""The 0MQ url used for registration. This sets transport, ip, and port
         in one variable. For example: url='tcp://127.0.0.1:12345' or
         url='epgm://*:90210'""") # url takes precedence over ip,regport,transport
     transport = Unicode('tcp', config=True,
         help="""The 0MQ transport for communications.  This will likely be
         the default of 'tcp', but other values include 'ipc', 'epgm', 'inproc'.""")
     ip = Unicode('127.0.0.1', config=True,
         help="""The IP address for registration.  This is generally either
         '127.0.0.1' for loopback only or '*' for all interfaces.
         [default: '127.0.0.1']""")
     regport = Int(config=True,
         help="""The port on which the Hub listens for registration.""")
     def _regport_default(self):
         return select_random_ports(1)[0]
     
     def __init__(self, **kwargs):
         super(RegistrationFactory, self).__init__(**kwargs)
         self._propagate_url()
         self._rebuild_url()
         self.on_trait_change(self._propagate_url, 'url')
         self.on_trait_change(self._rebuild_url, 'ip')
         self.on_trait_change(self._rebuild_url, 'transport')
         self.on_trait_change(self._rebuild_url, 'regport')
     
     def _rebuild_url(self):
         self.url = "%s://%s:%i"%(self.transport, self.ip, self.regport)
         
     def _propagate_url(self):
         """Ensure self.url contains full transport://interface:port"""
         if self.url:
             iface = self.url.split('://',1)
             if len(iface) == 2:
                 self.transport,iface = iface
             iface = iface.split(':')
             self.ip = iface[0]
             if iface[1]:
                 self.regport = int(iface[1])

IPython/parallel/util.py

@@ -1,466 +1,468 b''
 """some generic utilities for dealing with classes, urls, and serialization"""
 #-----------------------------------------------------------------------------
 #  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
 #-----------------------------------------------------------------------------
 
 # Standard library imports.
 import logging
 import os
 import re
 import stat
 import socket
 import sys
 from signal import signal, SIGINT, SIGABRT, SIGTERM
 try:
     from signal import SIGKILL
 except ImportError:
     SIGKILL=None
 
 try:
     import cPickle
     pickle = cPickle
 except:
     cPickle = None
     import pickle
 
 # System library imports
 import zmq
 from zmq.log import handlers
 
 # IPython imports
 from IPython.utils.pickleutil import can, uncan, canSequence, uncanSequence
 from IPython.utils.newserialized import serialize, unserialize
 from IPython.zmq.log import EnginePUBHandler
 
 #-----------------------------------------------------------------------------
 # Classes
 #-----------------------------------------------------------------------------
 
 class Namespace(dict):
     """Subclass of dict for attribute access to keys."""
     
     def __getattr__(self, key):
         """getattr aliased to getitem"""
         if key in self.iterkeys():
             return self[key]
         else:
             raise NameError(key)
 
     def __setattr__(self, key, value):
         """setattr aliased to setitem, with strict"""
         if hasattr(dict, key):
             raise KeyError("Cannot override dict keys %r"%key)
         self[key] = value
     
 
 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._reverse.pop(value)
         return value
     
     def get(self, key, default=None):
         try:
             return self[key]
         except KeyError:
             return default
 
 #-----------------------------------------------------------------------------
 # Functions
 #-----------------------------------------------------------------------------
 
 def validate_url(url):
     """validate a url for zeromq"""
     if not isinstance(url, basestring):
         raise TypeError("url must be a string, not %r"%type(url))
     url = url.lower()
     
     proto_addr = url.split('://')
     assert len(proto_addr) == 2, 'Invalid url: %r'%url
     proto, addr = proto_addr
     assert proto in ['tcp','pgm','epgm','ipc','inproc'], "Invalid protocol: %r"%proto
     
     # domain pattern adapted from http://www.regexlib.com/REDetails.aspx?regexp_id=391
     # author: Remi Sabourin
     pat = re.compile(r'^([\w\d]([\w\d\-]{0,61}[\w\d])?\.)*[\w\d]([\w\d\-]{0,61}[\w\d])?$')
     
     if proto == 'tcp':
         lis = addr.split(':')
         assert len(lis) == 2, 'Invalid url: %r'%url
         addr,s_port = lis
         try:
             port = int(s_port)
         except ValueError:
             raise AssertionError("Invalid port %r in url: %r"%(port, url))
         
         assert addr == '*' or pat.match(addr) is not None, 'Invalid url: %r'%url
         
     else:
         # only validate tcp urls currently
         pass
     
     return True
 
 
 def validate_url_container(container):
     """validate a potentially nested collection of urls."""
     if isinstance(container, basestring):
         url = container
         return validate_url(url)
     elif isinstance(container, dict):
         container = container.itervalues()
     
     for element in container:
         validate_url_container(element)
 
 
 def split_url(url):
     """split a zmq url (tcp://ip:port) into ('tcp','ip','port')."""
     proto_addr = url.split('://')
     assert len(proto_addr) == 2, 'Invalid url: %r'%url
     proto, addr = proto_addr
     lis = addr.split(':')
     assert len(lis) == 2, 'Invalid url: %r'%url
     addr,s_port = lis
     return proto,addr,s_port
     
 def disambiguate_ip_address(ip, location=None):
     """turn multi-ip interfaces '0.0.0.0' and '*' into connectable
     ones, based on the location (default interpretation of location is localhost)."""
     if ip in ('0.0.0.0', '*'):
         external_ips = socket.gethostbyname_ex(socket.gethostname())[2]
         if location is None or location in external_ips:
             ip='127.0.0.1'
         elif location:
             return location
     return ip
 
 def disambiguate_url(url, location=None):
     """turn multi-ip interfaces '0.0.0.0' and '*' into connectable
     ones, based on the location (default interpretation is localhost).
     
     This is for zeromq urls, such as tcp://*:10101."""
     try:
         proto,ip,port = split_url(url)
     except AssertionError:
         # probably not tcp url; could be ipc, etc.
         return url
     
     ip = disambiguate_ip_address(ip,location)
     
     return "%s://%s:%s"%(proto,ip,port)
 
 
 def rekey(dikt):
     """Rekey a dict that has been forced to use str keys where there should be
     ints by json.  This belongs in the jsonutil added by fperez."""
     for k in dikt.iterkeys():
         if isinstance(k, str):
             ik=fk=None
             try:
                 ik = int(k)
             except ValueError:
                 try:
                     fk = float(k)
                 except ValueError:
                     continue
             if ik is not None:
                 nk = ik
             else:
                 nk = fk
             if nk in dikt:
                 raise KeyError("already have key %r"%nk)
             dikt[nk] = dikt.pop(k)
     return dikt
 
 def serialize_object(obj, threshold=64e-6):
     """Serialize an object into a list of sendable buffers.
     
     Parameters
     ----------
     
     obj : object
         The object to be serialized
     threshold : float
         The threshold for not double-pickling the content.
         
     
     Returns
     -------
     ('pmd', [bufs]) :
         where pmd is the pickled metadata wrapper,
         bufs is a list of data buffers
     """
     databuffers = []
     if isinstance(obj, (list, tuple)):
         clist = canSequence(obj)
         slist = map(serialize, clist)
         for s in slist:
             if s.typeDescriptor in ('buffer', 'ndarray') or s.getDataSize() > threshold:
                 databuffers.append(s.getData())
                 s.data = None
         return pickle.dumps(slist,-1), databuffers
     elif isinstance(obj, dict):
         sobj = {}
         for k in sorted(obj.iterkeys()):
             s = serialize(can(obj[k]))
             if s.typeDescriptor in ('buffer', 'ndarray') or s.getDataSize() > threshold:
                 databuffers.append(s.getData())
                 s.data = None
             sobj[k] = s
         return pickle.dumps(sobj,-1),databuffers
     else:
         s = serialize(can(obj))
         if s.typeDescriptor in ('buffer', 'ndarray') or s.getDataSize() > threshold:
             databuffers.append(s.getData())
             s.data = None
         return pickle.dumps(s,-1),databuffers
             
         
 def unserialize_object(bufs):
     """reconstruct an object serialized by serialize_object from data buffers."""
     bufs = list(bufs)
     sobj = pickle.loads(bufs.pop(0))
     if isinstance(sobj, (list, tuple)):
         for s in sobj:
             if s.data is None:
                 s.data = bufs.pop(0)
         return uncanSequence(map(unserialize, sobj)), bufs
     elif isinstance(sobj, dict):
         newobj = {}
         for k in sorted(sobj.iterkeys()):
             s = sobj[k]
             if s.data is None:
                 s.data = bufs.pop(0)
             newobj[k] = uncan(unserialize(s))
         return newobj, bufs
     else:
         if sobj.data is None:
             sobj.data = bufs.pop(0)
         return uncan(unserialize(sobj)), bufs
 
 def pack_apply_message(f, args, kwargs, threshold=64e-6):
     """pack up a function, args, and kwargs to be sent over the wire
     as a series of buffers. Any object whose data is larger than `threshold`
     will not have their data copied (currently only numpy arrays support zero-copy)"""
     msg = [pickle.dumps(can(f),-1)]
     databuffers = [] # for large objects
     sargs, bufs = serialize_object(args,threshold)
     msg.append(sargs)
     databuffers.extend(bufs)
     skwargs, bufs = serialize_object(kwargs,threshold)
     msg.append(skwargs)
     databuffers.extend(bufs)
     msg.extend(databuffers)
     return msg
 
 def unpack_apply_message(bufs, g=None, copy=True):
     """unpack f,args,kwargs from buffers packed by pack_apply_message()
     Returns: original f,args,kwargs"""
     bufs = list(bufs) # allow us to pop
     assert len(bufs) >= 3, "not enough buffers!"
     if not copy:
         for i in range(3):
             bufs[i] = bufs[i].bytes
     cf = pickle.loads(bufs.pop(0))
     sargs = list(pickle.loads(bufs.pop(0)))
     skwargs = dict(pickle.loads(bufs.pop(0)))
     # print sargs, skwargs
     f = uncan(cf, g)
     for sa in sargs:
         if sa.data is None:
             m = bufs.pop(0)
             if sa.getTypeDescriptor() in ('buffer', 'ndarray'):
                 # always use a buffer, until memoryviews get sorted out
                 sa.data = buffer(m)
                 # disable memoryview support
                 # if copy:
                 #     sa.data = buffer(m)
                 # else:
                 #     sa.data = m.buffer
             else:
                 if copy:
                     sa.data = m
                 else:
                     sa.data = m.bytes
     
     args = uncanSequence(map(unserialize, sargs), g)
     kwargs = {}
     for k in sorted(skwargs.iterkeys()):
         sa = skwargs[k]
         if sa.data is None:
             m = bufs.pop(0)
             if sa.getTypeDescriptor() in ('buffer', 'ndarray'):
                 # always use a buffer, until memoryviews get sorted out
                 sa.data = buffer(m)
                 # disable memoryview support
                 # if copy:
                 #     sa.data = buffer(m)
                 # else:
                 #     sa.data = m.buffer
             else:
                 if copy:
                     sa.data = m
                 else:
                     sa.data = m.bytes
 
         kwargs[k] = uncan(unserialize(sa), g)
     
     return f,args,kwargs
 
 #--------------------------------------------------------------------------
 # helpers for implementing old MEC API via view.apply
 #--------------------------------------------------------------------------
 
 def interactive(f):
     """decorator for making functions appear as interactively defined.
     This results in the function being linked to the user_ns as globals()
     instead of the module globals().
     """
     f.__module__ = '__main__'
     return f
 
 @interactive
 def _push(ns):
     """helper method for implementing `client.push` via `client.apply`"""
     globals().update(ns)
 
 @interactive
 def _pull(keys):
     """helper method for implementing `client.pull` via `client.apply`"""
     user_ns = globals()
     if isinstance(keys, (list,tuple, set)):
         for key in keys:
             if not user_ns.has_key(key):
                 raise NameError("name '%s' is not defined"%key)
         return map(user_ns.get, keys)
     else:
         if not user_ns.has_key(keys):
             raise NameError("name '%s' is not defined"%keys)
         return user_ns.get(keys)
 
 @interactive
 def _execute(code):
     """helper method for implementing `client.execute` via `client.apply`"""
     exec code in globals()
 
 #--------------------------------------------------------------------------
 # extra process management utilities
 #--------------------------------------------------------------------------
 
 _random_ports = set()
 
 def select_random_ports(n):
     """Selects and return n random ports that are available."""
     ports = []
     for i in xrange(n):
         sock = socket.socket()
         sock.bind(('', 0))
         while sock.getsockname()[1] in _random_ports:
             sock.close()
             sock = socket.socket()
             sock.bind(('', 0))
         ports.append(sock)
     for i, sock in enumerate(ports):
         port = sock.getsockname()[1]
         sock.close()
         ports[i] = port
         _random_ports.add(port)
     return ports
 
 def signal_children(children):
     """Relay interupt/term signals to children, for more solid process cleanup."""
     def terminate_children(sig, frame):
         logging.critical("Got signal %i, terminating children..."%sig)
         for child in children:
             child.terminate()
         
         sys.exit(sig != SIGINT)
         # sys.exit(sig)
     for sig in (SIGINT, SIGABRT, SIGTERM):
         signal(sig, terminate_children)
 
 def generate_exec_key(keyfile):
     import uuid
     newkey = str(uuid.uuid4())
     with open(keyfile, 'w') as f:
         # f.write('ipython-key ')
         f.write(newkey+'\n')
     # set user-only RW permissions (0600)
     # this will have no effect on Windows
     os.chmod(keyfile, stat.S_IRUSR|stat.S_IWUSR)
 
 
 def integer_loglevel(loglevel):
     try:
         loglevel = int(loglevel)
     except ValueError:
         if isinstance(loglevel, str):
             loglevel = getattr(logging, loglevel)
     return loglevel
 
 def connect_logger(logname, context, iface, root="ip", loglevel=logging.DEBUG):
     logger = logging.getLogger(logname)
     if any([isinstance(h, handlers.PUBHandler) for h in logger.handlers]):
         # don't add a second PUBHandler
         return
     loglevel = integer_loglevel(loglevel)
     lsock = context.socket(zmq.PUB)
     lsock.connect(iface)
     handler = handlers.PUBHandler(lsock)
     handler.setLevel(loglevel)
     handler.root_topic = root
     logger.addHandler(handler)
     logger.setLevel(loglevel)
 
 def connect_engine_logger(context, iface, engine, loglevel=logging.DEBUG):
     logger = logging.getLogger()
     if any([isinstance(h, handlers.PUBHandler) for h in logger.handlers]):
         # don't add a second PUBHandler
         return
     loglevel = integer_loglevel(loglevel)
     lsock = context.socket(zmq.PUB)
     lsock.connect(iface)
     handler = EnginePUBHandler(engine, lsock)
     handler.setLevel(loglevel)
     logger.addHandler(handler)
     logger.setLevel(loglevel)
+    return logger
 
 def local_logger(logname, loglevel=logging.DEBUG):
     loglevel = integer_loglevel(loglevel)
     logger = logging.getLogger(logname)
     if any([isinstance(h, logging.StreamHandler) for h in logger.handlers]):
         # don't add a second StreamHandler
         return
     handler = logging.StreamHandler()
     handler.setLevel(loglevel)
     logger.addHandler(handler)
     logger.setLevel(loglevel)
+    return logger
 

IPython/zmq/session.py

@@ -1,479 +1,511 b''
 #!/usr/bin/env python
 """edited session.py to work with streams, and move msg_type to the header
 """
 #-----------------------------------------------------------------------------
 #  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
 #-----------------------------------------------------------------------------
 
 import hmac
+import logging
 import os
 import pprint
 import uuid
 from datetime import datetime
 
 try:
     import cPickle
     pickle = cPickle
 except:
     cPickle = None
     import pickle
 
 import zmq
 from zmq.utils import jsonapi
+from zmq.eventloop.ioloop import IOLoop
 from zmq.eventloop.zmqstream import ZMQStream
 
 from IPython.config.configurable import Configurable
 from IPython.utils.importstring import import_item
 from IPython.utils.jsonutil import date_default
 from IPython.utils.traitlets import CStr, Unicode, Bool, Any, Instance, Set
 
 #-----------------------------------------------------------------------------
 # utility functions
 #-----------------------------------------------------------------------------
 
 def squash_unicode(obj):
     """coerce unicode back to bytestrings."""
     if isinstance(obj,dict):
         for key in obj.keys():
             obj[key] = squash_unicode(obj[key])
             if isinstance(key, unicode):
                 obj[squash_unicode(key)] = obj.pop(key)
     elif isinstance(obj, list):
         for i,v in enumerate(obj):
             obj[i] = squash_unicode(v)
     elif isinstance(obj, unicode):
         obj = obj.encode('utf8')
     return obj
 
 #-----------------------------------------------------------------------------
 # globals and defaults
 #-----------------------------------------------------------------------------
 
 _default_key = 'on_unknown' if jsonapi.jsonmod.__name__ == 'jsonlib' else 'default'
 json_packer = lambda obj: jsonapi.dumps(obj, **{_default_key:date_default})
 json_unpacker = lambda s: squash_unicode(jsonapi.loads(s))
 
 pickle_packer = lambda o: pickle.dumps(o,-1)
 pickle_unpacker = pickle.loads
 
 default_packer = json_packer
 default_unpacker = json_unpacker
 
 
 DELIM="<IDS|MSG>"
 
 #-----------------------------------------------------------------------------
 # Classes
 #-----------------------------------------------------------------------------
 
+class SessionFactory(Configurable):
+    """The Base class for configurables that have a Session, Context, logger,
+    and IOLoop.
+    """
+    
+    log = Instance('logging.Logger', ('', logging.WARN))
+    
+    logname = Unicode('')
+    def _logname_changed(self, name, old, new):
+        self.log = logging.getLogger(new)
+    
+    # not configurable:
+    context = Instance('zmq.Context')
+    def _context_default(self):
+        return zmq.Context.instance()
+    
+    session = Instance('IPython.zmq.session.Session')
+    
+    loop = Instance('zmq.eventloop.ioloop.IOLoop', allow_none=False)
+    def _loop_default(self):
+        return IOLoop.instance()
+    
+    def __init__(self, **kwargs):
+        super(SessionFactory, self).__init__(**kwargs)
+        
+        if self.session is None:
+            # construct the session
+            self.session = Session(**kwargs)
+    
+
 class Message(object):
     """A simple message object that maps dict keys to attributes.
 
     A Message can be created from a dict and a dict from a Message instance
     simply by calling dict(msg_obj)."""
     
     def __init__(self, msg_dict):
         dct = self.__dict__
         for k, v in dict(msg_dict).iteritems():
             if isinstance(v, dict):
                 v = Message(v)
             dct[k] = v
 
     # Having this iterator lets dict(msg_obj) work out of the box.
     def __iter__(self):
         return iter(self.__dict__.iteritems())
     
     def __repr__(self):
         return repr(self.__dict__)
 
     def __str__(self):
         return pprint.pformat(self.__dict__)
 
     def __contains__(self, k):
         return k in self.__dict__
 
     def __getitem__(self, k):
         return self.__dict__[k]
 
 
 def msg_header(msg_id, msg_type, username, session):
     date=datetime.now()
     return locals()
 
 def extract_header(msg_or_header):
     """Given a message or header, return the header."""
     if not msg_or_header:
         return {}
     try:
         # See if msg_or_header is the entire message.
         h = msg_or_header['header']
     except KeyError:
         try:
             # See if msg_or_header is just the header
             h = msg_or_header['msg_id']
         except KeyError:
             raise
         else:
             h = msg_or_header
     if not isinstance(h, dict):
         h = dict(h)
     return h
 
 class Session(Configurable):
     """tweaked version of IPython.zmq.session.Session, for development in Parallel"""
     debug=Bool(False, config=True, help="""Debug output in the Session""")
     packer = Unicode('json',config=True,
             help="""The name of the packer for serializing messages.
             Should be one of 'json', 'pickle', or an import name
             for a custom serializer.""")
     def _packer_changed(self, name, old, new):
         if new.lower() == 'json':
             self.pack = json_packer
             self.unpack = json_unpacker
         elif new.lower() == 'pickle':
             self.pack = pickle_packer
             self.unpack = pickle_unpacker
         else:
             self.pack = import_item(new)
 
-    unpacker = Unicode('json',config=True,
+    unpacker = Unicode('json', config=True,
         help="""The name of the unpacker for unserializing messages.
         Only used with custom functions for `packer`.""")
     def _unpacker_changed(self, name, old, new):
         if new.lower() == 'json':
             self.pack = json_packer
             self.unpack = json_unpacker
         elif new.lower() == 'pickle':
             self.pack = pickle_packer
             self.unpack = pickle_unpacker
         else:
             self.unpack = import_item(new)
         
-    session = CStr('',config=True,
+    session = CStr('', config=True,
         help="""The UUID identifying this session.""")
     def _session_default(self):
         return bytes(uuid.uuid4())
     username = Unicode(os.environ.get('USER','username'), config=True,
         help="""Username for the Session. Default is your system username.""")
     
     # message signature related traits:
     key = CStr('', config=True,
         help="""execution key, for extra authentication.""")
     def _key_changed(self, name, old, new):
         if new:
             self.auth = hmac.HMAC(new)
         else:
             self.auth = None
     auth = Instance(hmac.HMAC)
     counters = Instance('collections.defaultdict', (int,))
     digest_history = Set()
     
     keyfile = Unicode('', config=True,
         help="""path to file containing execution key.""")
     def _keyfile_changed(self, name, old, new):
         with open(new, 'rb') as f:
             self.key = f.read().strip()
 
     pack = Any(default_packer) # the actual packer function
     def _pack_changed(self, name, old, new):
         if not callable(new):
             raise TypeError("packer must be callable, not %s"%type(new))
         
     unpack = Any(default_unpacker) # the actual packer function
     def _unpack_changed(self, name, old, new):
         if not callable(new):
             raise TypeError("packer must be callable, not %s"%type(new))
 
     def __init__(self, **kwargs):
         super(Session, self).__init__(**kwargs)
         self.none = self.pack({})
 
     @property
     def msg_id(self):
         """always return new uuid"""
         return str(uuid.uuid4())
 
     def msg_header(self, msg_type):
         return msg_header(self.msg_id, msg_type, self.username, self.session)
 
     def msg(self, msg_type, content=None, parent=None, subheader=None):
         msg = {}
         msg['header'] = self.msg_header(msg_type)
         msg['msg_id'] = msg['header']['msg_id']
         msg['parent_header'] = {} if parent is None else extract_header(parent)
         msg['msg_type'] = msg_type
         msg['content'] = {} if content is None else content
         sub = {} if subheader is None else subheader
         msg['header'].update(sub)
         return msg
 
     def check_key(self, msg_or_header):
         """Check that a message's header has the right key"""
         if not self.key:
             return True
         header = extract_header(msg_or_header)
         return header.get('key', '') == self.key
     
     def sign(self, msg):
         """Sign a message with HMAC digest. If no auth, return b''."""
         if self.auth is None:
             return b''
         h = self.auth.copy()
         for m in msg:
             h.update(m)
         return h.hexdigest()
 
     def serialize(self, msg, ident=None):
         content = msg.get('content', {})
         if content is None:
             content = self.none
         elif isinstance(content, dict):
             content = self.pack(content)
         elif isinstance(content, bytes):
             # content is already packed, as in a relayed message
             pass
         elif isinstance(content, unicode):
             # should be bytes, but JSON often spits out unicode
             content = content.encode('utf8')
         else:
             raise TypeError("Content incorrect type: %s"%type(content))
         
         real_message = [self.pack(msg['header']), 
                         self.pack(msg['parent_header']), 
                         content
         ]
         
         to_send = []
 
         if isinstance(ident, list):
             # accept list of idents
             to_send.extend(ident)
         elif ident is not None:
             to_send.append(ident)
         to_send.append(DELIM)
         
         signature = self.sign(real_message)
         to_send.append(signature)
         
         to_send.extend(real_message)
 
         return to_send
         
     def send(self, stream, msg_or_type, content=None, parent=None, ident=None, 
                                     buffers=None, subheader=None, track=False):
         """Build and send a message via stream or socket.
         
         Parameters
         ----------
         
         stream : zmq.Socket or ZMQStream
             the socket-like object used to send the data
         msg_or_type : str or Message/dict
             Normally, msg_or_type will be a msg_type unless a message is being sent more
             than once.
         
         content : dict or None
             the content of the message (ignored if msg_or_type is a message)
         parent : Message or dict or None
             the parent or parent header describing the parent of this message
         ident : bytes or list of bytes
             the zmq.IDENTITY routing path
         subheader : dict or None
             extra header keys for this message's header
         buffers : list or None
             the already-serialized buffers to be appended to the message
         track : bool
             whether to track.  Only for use with Sockets, 
             because ZMQStream objects cannot track messages.
         
         Returns
         -------
         msg : message dict
             the constructed message
         (msg,tracker) : (message dict, MessageTracker)
             if track=True, then a 2-tuple will be returned, 
             the first element being the constructed
             message, and the second being the MessageTracker
             
         """
 
         if not isinstance(stream, (zmq.Socket, ZMQStream)):
             raise TypeError("stream must be Socket or ZMQStream, not %r"%type(stream))
         elif track and isinstance(stream, ZMQStream):
             raise TypeError("ZMQStream cannot track messages")
         
         if isinstance(msg_or_type, (Message, dict)):
             # we got a Message, not a msg_type
             # don't build a new Message
             msg = msg_or_type
         else:
             msg = self.msg(msg_or_type, content, parent, subheader)
         
         buffers = [] if buffers is None else buffers
         to_send = self.serialize(msg, ident)
         flag = 0
         if buffers:
             flag = zmq.SNDMORE
             _track = False
         else:
             _track=track
         if track:
             tracker = stream.send_multipart(to_send, flag, copy=False, track=_track)
         else:
             tracker = stream.send_multipart(to_send, flag, copy=False)
         for b in buffers[:-1]:
             stream.send(b, flag, copy=False)
         if buffers:
             if track:
                 tracker = stream.send(buffers[-1], copy=False, track=track)
             else:
                 tracker = stream.send(buffers[-1], copy=False)
                 
         # omsg = Message(msg)
         if self.debug:
             pprint.pprint(msg)
             pprint.pprint(to_send)
             pprint.pprint(buffers)
         
         msg['tracker'] = tracker
         
         return msg
     
     def send_raw(self, stream, msg, flags=0, copy=True, ident=None):
         """Send a raw message via ident path.
         
         Parameters
         ----------
         msg : list of sendable buffers"""
         to_send = []
         if isinstance(ident, bytes):
             ident = [ident]
         if ident is not None:
             to_send.extend(ident)
             
         to_send.append(DELIM)
         to_send.append(self.sign(msg))
         to_send.extend(msg)
         stream.send_multipart(msg, flags, copy=copy)
     
     def recv(self, socket, mode=zmq.NOBLOCK, content=True, copy=True):
         """receives and unpacks a message
         returns [idents], msg"""
         if isinstance(socket, ZMQStream):
             socket = socket.socket
         try:
             msg = socket.recv_multipart(mode)
         except zmq.ZMQError as e:
             if e.errno == zmq.EAGAIN:
                 # We can convert EAGAIN to None as we know in this case
                 # recv_multipart won't return None.
                 return None,None
             else:
                 raise
         # return an actual Message object
         # determine the number of idents by trying to unpack them.
         # this is terrible:
         idents, msg = self.feed_identities(msg, copy)
         try:
             return idents, self.unpack_message(msg, content=content, copy=copy)
         except Exception as e:
             print (idents, msg)
             # TODO: handle it
             raise e
     
     def feed_identities(self, msg, copy=True):
         """feed until DELIM is reached, then return the prefix as idents and remainder as
         msg. This is easily broken by setting an IDENT to DELIM, but that would be silly.
         
         Parameters
         ----------
         msg : a list of Message or bytes objects
             the message to be split
         copy : bool
             flag determining whether the arguments are bytes or Messages
         
         Returns
         -------
         (idents,msg) : two lists
             idents will always be a list of bytes - the indentity prefix
             msg will be a list of bytes or Messages, unchanged from input
             msg should be unpackable via self.unpack_message at this point.
         """
         if copy:
             idx = msg.index(DELIM)
             return msg[:idx], msg[idx+1:]
         else:
             failed = True
             for idx,m in enumerate(msg):
                 if m.bytes == DELIM:
                     failed = False
                     break
             if failed:
                 raise ValueError("DELIM not in msg")
             idents, msg = msg[:idx], msg[idx+1:]
             return [m.bytes for m in idents], msg
     
     def unpack_message(self, msg, content=True, copy=True):
         """Return a message object from the format
         sent by self.send.
         
         Parameters:
         -----------
         
         content : bool (True)
             whether to unpack the content dict (True), 
             or leave it serialized (False)
         
         copy : bool (True)
             whether to return the bytes (True), 
             or the non-copying Message object in each place (False)
         
         """
         minlen = 4
         message = {}
         if not copy:
             for i in range(minlen):
                 msg[i] = msg[i].bytes
         if self.auth is not None:
             signature = msg[0]
             if signature in self.digest_history:
                 raise ValueError("Duplicate Signature: %r"%signature)
             self.digest_history.add(signature)
             check = self.sign(msg[1:4])
             if not signature == check:
                 raise ValueError("Invalid Signature: %r"%signature)
         if not len(msg) >= minlen:
             raise TypeError("malformed message, must have at least %i elements"%minlen)
         message['header'] = self.unpack(msg[1])
         message['msg_type'] = message['header']['msg_type']
         message['parent_header'] = self.unpack(msg[2])
         if content:
             message['content'] = self.unpack(msg[3])
         else:
             message['content'] = msg[3]
         
         message['buffers'] = msg[4:]
         return message
 
 def test_msg2obj():
     am = dict(x=1)
     ao = Message(am)
     assert ao.x == am['x']
 
     am['y'] = dict(z=1)
     ao = Message(am)
     assert ao.y.z == am['y']['z']
     
     k1, k2 = 'y', 'z'
     assert ao[k1][k2] == am[k1][k2]
     
     am2 = dict(ao)
     assert am['x'] == am2['x']
     assert am['y']['z'] == am2['y']['z']