add default ip<x>z_config files

IPython/config/default/ipclusterz_config.py

@@ -1,184 +1,184 b''
 import os
 
 c = get_config()
 
 #-----------------------------------------------------------------------------
 # Select which launchers to use
 #-----------------------------------------------------------------------------
 
 # This allows you to control what method is used to start the controller
 # and engines.  The following methods are currently supported:
 # - Start as a regular process on localhost.
 # - Start using mpiexec.
 # - Start using the Windows HPC Server 2008 scheduler
 # - Start using PBS
-# - Start using SSH (currently broken)
+# - Start using SSH
 
 
 # The selected launchers can be configured below.
 
 # Options are:
 # - LocalControllerLauncher
 # - MPIExecControllerLauncher
 # - PBSControllerLauncher
 # - WindowsHPCControllerLauncher
-# c.Global.controller_launcher = 'IPython.kernel.launcher.LocalControllerLauncher'
+# c.Global.controller_launcher = 'IPython.zmq.parallel.launcher.LocalControllerLauncher'
 
 # Options are:
 # - LocalEngineSetLauncher
 # - MPIExecEngineSetLauncher
 # - PBSEngineSetLauncher
 # - WindowsHPCEngineSetLauncher
-# c.Global.engine_launcher = 'IPython.kernel.launcher.LocalEngineSetLauncher'
+# c.Global.engine_launcher = 'IPython.zmq.parallel.launcher.LocalEngineSetLauncher'
 
 #-----------------------------------------------------------------------------
 # Global configuration
 #-----------------------------------------------------------------------------
 
 # The default number of engines that will be started. This is overridden by
 # the -n command line option: "ipcluster start -n 4"
 # c.Global.n = 2
 
 # Log to a file in cluster_dir/log, otherwise just log to sys.stdout.
 # c.Global.log_to_file = False
 
 # Remove old logs from cluster_dir/log before starting.
 # c.Global.clean_logs = True
 
 # The working directory for the process. The application will use os.chdir
 # to change to this directory before starting.
 # c.Global.work_dir = os.getcwd()
 
 
 #-----------------------------------------------------------------------------
 # Local process launchers
 #-----------------------------------------------------------------------------
 
 # The command line arguments to call the controller with.
 # c.LocalControllerLauncher.controller_args = \
 #    ['--log-to-file','--log-level', '40']
 
 # The working directory for the controller
 # c.LocalEngineSetLauncher.work_dir = u''
 
 # Command line argument passed to the engines.
 # c.LocalEngineSetLauncher.engine_args = ['--log-to-file','--log-level', '40']
 
 #-----------------------------------------------------------------------------
 # MPIExec launchers
 #-----------------------------------------------------------------------------
 
 # The mpiexec/mpirun command to use in started the controller.
 # c.MPIExecControllerLauncher.mpi_cmd = ['mpiexec']
 
 # Additional arguments to pass to the actual mpiexec command.
 # c.MPIExecControllerLauncher.mpi_args = []
 
 # The command line argument to call the controller with.
 # c.MPIExecControllerLauncher.controller_args = \
 #     ['--log-to-file','--log-level', '40']
 
 
 # The mpiexec/mpirun command to use in started the controller.
 # c.MPIExecEngineSetLauncher.mpi_cmd = ['mpiexec']
 
 # Additional arguments to pass to the actual mpiexec command.
 # c.MPIExecEngineSetLauncher.mpi_args = []
 
 # Command line argument passed to the engines.
 # c.MPIExecEngineSetLauncher.engine_args = ['--log-to-file','--log-level', '40']
 
 # The default number of engines to start if not given elsewhere.
 # c.MPIExecEngineSetLauncher.n = 1
 
 #-----------------------------------------------------------------------------
 # SSH launchers
 #-----------------------------------------------------------------------------
 
 # Todo
 
 
 #-----------------------------------------------------------------------------
 # Unix batch (PBS) schedulers launchers
 #-----------------------------------------------------------------------------
 
 # The command line program to use to submit a PBS job.
 # c.PBSControllerLauncher.submit_command = 'qsub'
 
 # The command line program to use to delete a PBS job.
 # c.PBSControllerLauncher.delete_command = 'qdel'
 
 # A regular expression that takes the output of qsub and find the job id.
 # c.PBSControllerLauncher.job_id_regexp = r'\d+'
 
 # The batch submission script used to start the controller. This is where
 # environment variables would be setup, etc. This string is interpolated using
 # the Itpl module in IPython.external. Basically, you can use ${n} for the
 # number of engine and ${cluster_dir} for the cluster_dir.
 # c.PBSControllerLauncher.batch_template = """"""
 
 # The name of the instantiated batch script that will actually be used to
 # submit the job. This will be written to the cluster directory.
 # c.PBSControllerLauncher.batch_file_name = u'pbs_batch_script_controller'
 
 
 # The command line program to use to submit a PBS job.
 # c.PBSEngineSetLauncher.submit_command = 'qsub'
 
 # The command line program to use to delete a PBS job.
 # c.PBSEngineSetLauncher.delete_command = 'qdel'
 
 # A regular expression that takes the output of qsub and find the job id.
 # c.PBSEngineSetLauncher.job_id_regexp = r'\d+'
 
 # The batch submission script used to start the engines. This is where
 # environment variables would be setup, etc. This string is interpolated using
 # the Itpl module in IPython.external. Basically, you can use ${n} for the
 # number of engine and ${cluster_dir} for the cluster_dir.
 # c.PBSEngineSetLauncher.batch_template = """"""
 
 # The name of the instantiated batch script that will actually be used to
 # submit the job. This will be written to the cluster directory.
 # c.PBSEngineSetLauncher.batch_file_name = u'pbs_batch_script_engines'
 
 #-----------------------------------------------------------------------------
 # Windows HPC Server 2008 launcher configuration
 #-----------------------------------------------------------------------------
 
 # c.IPControllerJob.job_name = 'IPController'
 # c.IPControllerJob.is_exclusive = False
 # c.IPControllerJob.username = r'USERDOMAIN\USERNAME'
 # c.IPControllerJob.priority = 'Highest'
 # c.IPControllerJob.requested_nodes = ''
 # c.IPControllerJob.project = 'MyProject'
 
 # c.IPControllerTask.task_name = 'IPController'
 # c.IPControllerTask.controller_cmd = [u'ipcontroller.exe']
 # c.IPControllerTask.controller_args = ['--log-to-file', '--log-level', '40']
 # c.IPControllerTask.environment_variables = {}
 
 # c.WindowsHPCControllerLauncher.scheduler = 'HEADNODE'
 # c.WindowsHPCControllerLauncher.job_file_name = u'ipcontroller_job.xml'
 
 
 # c.IPEngineSetJob.job_name = 'IPEngineSet'
 # c.IPEngineSetJob.is_exclusive = False
 # c.IPEngineSetJob.username = r'USERDOMAIN\USERNAME'
 # c.IPEngineSetJob.priority = 'Highest'
 # c.IPEngineSetJob.requested_nodes = ''
 # c.IPEngineSetJob.project = 'MyProject'
 
 # c.IPEngineTask.task_name = 'IPEngine'
 # c.IPEngineTask.engine_cmd = [u'ipengine.exe']
 # c.IPEngineTask.engine_args = ['--log-to-file', '--log-level', '40']
 # c.IPEngineTask.environment_variables = {}
 
 # c.WindowsHPCEngineSetLauncher.scheduler = 'HEADNODE'
 # c.WindowsHPCEngineSetLauncher.job_file_name = u'ipengineset_job.xml'
 
 
 
 
 
 
 

IPython/config/default/ipcontrollerz_config.py

@@ -1,136 +1,136 b''
 from IPython.config.loader import Config
 
 c = get_config()
 
 #-----------------------------------------------------------------------------
 # Global configuration
 #-----------------------------------------------------------------------------
 
 # Basic Global config attributes
 
 # Start up messages are logged to stdout using the logging module.
 # These all happen before the twisted reactor is started and are
 # useful for debugging purposes. Can be (10=DEBUG,20=INFO,30=WARN,40=CRITICAL) 
 # and smaller is more verbose.
 # c.Global.log_level = 20
 
 # Log to a file in cluster_dir/log, otherwise just log to sys.stdout.
 # c.Global.log_to_file = False
 
 # Remove old logs from cluster_dir/log before starting.
 # c.Global.clean_logs = True
 
 # A list of Python statements that will be run before starting the 
 # controller. This is provided because occasionally certain things need to 
 # be imported in the controller for pickling to work.
 # c.Global.import_statements = ['import math']
 
-# Reuse the controller's FURL files. If False, FURL files are regenerated
+# Reuse the controller's JSON files. If False, JSON files are regenerated
 # each time the controller is run. If True, they will be reused, *but*, you
 # also must set the network ports by hand. If set, this will override the
 # values set for the client and engine connections below.
-# c.Global.reuse_furls = True
+# c.Global.reuse_files = True
 
-# Enable SSL encryption on all connections to the controller. If set, this
-# will override the values set for the client and engine connections below.
+# Enable exec_key authentication on all messages. Default is True
 # c.Global.secure = True
 
 # The working directory for the process. The application will use os.chdir
 # to change to this directory before starting.
 # c.Global.work_dir = os.getcwd()
 
-#-----------------------------------------------------------------------------
-# Configure the client services
-#-----------------------------------------------------------------------------
+# The log url for logging to an `iploggerz` application.  This will override
+# log-to-file.
+# c.Global.log_url = 'tcp://127.0.0.1:20202'
 
-# Basic client service config attributes
+# The specific external IP that is used to disambiguate multi-interface URLs.
+# The default behavior is to guess from external IPs gleaned from `socket`.
+# c.Global.location = '192.168.1.123'
 
-# The network interface the controller will listen on for client connections.
-# This should be an IP address or hostname of the controller's host. The empty
-# string means listen on all interfaces.
-# c.FCClientServiceFactory.ip = ''
+# The ssh server remote clients should use to connect to this controller.
+# It must be a machine that can see the interface specified in client_ip.
+# The default for client_ip is localhost, in which case the sshserver must
+# be an external IP of the controller machine.
+# c.Global.sshserver = 'controller.example.com'
+
+# the url to use for registration.  If set, this overrides engine-ip,
+# engine-transport client-ip,client-transport, and regport.
+# c.RegistrationFactory.url = 'tcp://*:12345'
 
-# The TCP/IP port the controller will listen on for client connections. If 0
-# a random port will be used. If the controller's host has a firewall running
-# it must allow incoming traffic on this port.
-# c.FCClientServiceFactory.port = 0
+# the port to use for registration.  Clients and Engines both use this
+# port for registration.
+# c.RegistrationFactory.regport = 10101
 
-# The client learns how to connect to the controller by looking at the
-# location field embedded in the FURL. If this field is empty, all network
-# interfaces that the controller is listening on will be listed. To have the
-# client connect on a particular interface, list it here.
-# c.FCClientServiceFactory.location = ''
+#-----------------------------------------------------------------------------
+# Configure the Task Scheduler
+#-----------------------------------------------------------------------------
 
-# Use SSL encryption for the client connection.
-# c.FCClientServiceFactory.secure =  True
+# The routing scheme. 'pure' will use the pure-ZMQ scheduler. Any other 
+# value will use a Python scheduler with various routing schemes.
+# python schemes are: lru, weighted, random, twobin. Default is 'weighted'.
+# Note that the pure ZMQ scheduler does not support many features, such as
+# dying engines, dependencies, or engine-subset load-balancing.
+# c.ControllerFactory.scheme = 'pure'
 
-# Reuse the client FURL each time the controller is started. If set, you must
-# also pick a specific network port above (FCClientServiceFactory.port).
-# c.FCClientServiceFactory.reuse_furls = False
+# The pure ZMQ scheduler can limit the number of outstanding tasks per engine
+# by using the ZMQ HWM option.  This allows engines with long-running tasks
+# to not steal too many tasks from other engines. The default is 0, which
+# means agressively distribute messages, never waiting for them to finish.
+# c.ControllerFactory.hwm = 1
+
+# Whether to use Threads or Processes to start the Schedulers.  Threads will
+# use less resources, but potentially reduce throughput. Default is to 
+# use processes.  Note that the a Python scheduler will always be in a Process.
+# c.ControllerFactory.usethreads
 
 #-----------------------------------------------------------------------------
-# Configure the engine services
+# Configure the Hub
 #-----------------------------------------------------------------------------
 
-# Basic config attributes for the engine services.
+# Which class to use for the db backend.  Currently supported are DictDB (the
+# default), and MongoDB. Uncomment this line to enable MongoDB, which will
+# slow-down the Hub's responsiveness, but also reduce its memory footprint.
+# c.HubFactory.db_class = 'IPython.zmq.parallel.mongodb.MongoDB'
 
-# The network interface the controller will listen on for engine connections.
-# This should be an IP address or hostname of the controller's host. The empty
-# string means listen on all interfaces.
-# c.FCEngineServiceFactory.ip = ''
+# The heartbeat ping frequency.  This is the frequency (in ms) at which the
+# Hub pings engines for heartbeats.  This determines how quickly the Hub
+# will react to engines coming and going.  A lower number means faster response
+# time, but more network activity.  The default is 100ms
+# c.HubFactory.ping = 100
 
-# The TCP/IP port the controller will listen on for engine connections. If 0
-# a random port will be used. If the controller's host has a firewall running
-# it must allow incoming traffic on this port.
-# c.FCEngineServiceFactory.port = 0
+# HubFactory queue port pairs, to set by name: mux, iopub, control, task.  Set
+# each as a tuple of length 2 of ints.  The default is to find random
+# available ports
+# c.HubFactory.mux = (10102,10112)
 
-# The engine learns how to connect to the controller by looking at the
-# location field embedded in the FURL. If this field is empty, all network
-# interfaces that the controller is listening on will be listed. To have the
-# client connect on a particular interface, list it here.
-# c.FCEngineServiceFactory.location = ''
+#-----------------------------------------------------------------------------
+# Configure the client connections
+#-----------------------------------------------------------------------------
 
-# Use SSL encryption for the engine connection.
-# c.FCEngineServiceFactory.secure = True
+# Basic client connection config attributes
+
+# The network interface the controller will listen on for client connections.
+# This should be an IP address or interface on the controller. An asterisk
+# means listen on all interfaces. The transport can be any transport
+# supported by zeromq (tcp,epgm,pgm,ib,ipc):
+# c.HubFactory.client_ip = '*'
+# c.HubFactory.client_transport = 'tcp'
 
-# Reuse the client FURL each time the controller is started. If set, you must
-# also pick a specific network port above (FCClientServiceFactory.port).
-# c.FCEngineServiceFactory.reuse_furls = False
+# individual client ports to configure by name: query_port, notifier_port
+# c.HubFactory.query_port = 12345
 
 #-----------------------------------------------------------------------------
-# Developer level configuration attributes
+# Configure the engine connections
 #-----------------------------------------------------------------------------
 
-# You shouldn't have to modify anything in this section. These attributes
-# are more for developers who want to change the behavior of the controller
-# at a fundamental level.
-
-# c.FCClientServiceFactory.cert_file = u'ipcontroller-client.pem'
-
-# default_client_interfaces = Config()
-# default_client_interfaces.Task.interface_chain = [
-#     'IPython.kernel.task.ITaskController',
-#     'IPython.kernel.taskfc.IFCTaskController'
-# ]
-# 
-# default_client_interfaces.Task.furl_file = u'ipcontroller-tc.furl'
-# 
-# default_client_interfaces.MultiEngine.interface_chain = [
-#     'IPython.kernel.multiengine.IMultiEngine',
-#     'IPython.kernel.multienginefc.IFCSynchronousMultiEngine'
-# ]
-# 
-# default_client_interfaces.MultiEngine.furl_file = u'ipcontroller-mec.furl'
-# 
-# c.FCEngineServiceFactory.interfaces = default_client_interfaces
-
-# c.FCEngineServiceFactory.cert_file = u'ipcontroller-engine.pem'
-
-# default_engine_interfaces = Config()
-# default_engine_interfaces.Default.interface_chain = [
-#     'IPython.kernel.enginefc.IFCControllerBase'
-# ]
-# 
-# default_engine_interfaces.Default.furl_file = u'ipcontroller-engine.furl'
-# 
-# c.FCEngineServiceFactory.interfaces = default_engine_interfaces
+# Basic config attributes for the engine connections.
+
+# The network interface the controller will listen on for engine connections.
+# This should be an IP address or interface on the controller. An asterisk
+# means listen on all interfaces. The transport can be any transport
+# supported by zeromq (tcp,epgm,pgm,ib,ipc):
+# c.HubFactory.engine_ip = '*'
+# c.HubFactory.engine_transport = 'tcp'
+
+# set the engine heartbeat ports to use:
+# c.HubFactory.hb = (10303,10313)
+

IPython/config/default/ipenginez_config.py

@@ -1,90 +1,85 b''
 c = get_config()
 
 #-----------------------------------------------------------------------------
 # Global configuration
 #-----------------------------------------------------------------------------
 
 # Start up messages are logged to stdout using the logging module.
 # These all happen before the twisted reactor is started and are
 # useful for debugging purposes. Can be (10=DEBUG,20=INFO,30=WARN,40=CRITICAL) 
 # and smaller is more verbose.
 # c.Global.log_level = 20
 
 # Log to a file in cluster_dir/log, otherwise just log to sys.stdout.
 # c.Global.log_to_file = False
 
 # Remove old logs from cluster_dir/log before starting.
 # c.Global.clean_logs = True
 
 # A list of strings that will be executed in the users namespace on the engine
 # before it connects to the controller.
 # c.Global.exec_lines = ['import numpy']
 
 # The engine will try to connect to the controller multiple times, to allow
 # the controller time to startup and write its FURL file. These parameters 
 # control the number of retries (connect_max_tries) and the initial delay
 # (connect_delay) between attemps. The actual delay between attempts gets
 # longer each time by a factor of 1.5 (delay[i] = 1.5*delay[i-1])
 # those attemps.
 # c.Global.connect_delay = 0.1
 # c.Global.connect_max_tries = 15
 
-# By default, the engine will look for the controller's FURL file in its own
-# cluster directory. Sometimes, the FURL file will be elsewhere and this 
-# attribute can be set to the full path of the FURL file.
-# c.Global.furl_file = u''
+# By default, the engine will look for the controller's JSON file in its own
+# cluster directory. Sometimes, the JSON file will be elsewhere and this 
+# attribute can be set to the full path of the JSON file.
+# c.Global.url_file = u'/path/to/my/ipcontroller-engine.json'
 
 # The working directory for the process. The application will use os.chdir
 # to change to this directory before starting.
 # c.Global.work_dir = os.getcwd()
 
 #-----------------------------------------------------------------------------
 # MPI configuration
 #-----------------------------------------------------------------------------
 
 # Upon starting the engine can be configured to call MPI_Init. This section
 # configures that.
 
 # Select which MPI section to execute to setup MPI. The value of this 
 # attribute must match the name of another attribute in the MPI config 
 # section (mpi4py, pytrilinos, etc.). This can also be set by the --mpi
 # command line option.
 # c.MPI.use = ''
 
 # Initialize MPI using mpi4py. To use this, set c.MPI.use = 'mpi4py' to use
 # --mpi=mpi4py at the command line.
 # c.MPI.mpi4py = """from mpi4py import MPI as mpi
 # mpi.size = mpi.COMM_WORLD.Get_size()
 # mpi.rank = mpi.COMM_WORLD.Get_rank()
 # """
 
 # Initialize MPI using pytrilinos. To use this, set c.MPI.use = 'pytrilinos' 
 # to use --mpi=pytrilinos at the command line.
 # c.MPI.pytrilinos = """from PyTrilinos import Epetra
 # class SimpleStruct:
 # pass
 # mpi = SimpleStruct()
 # mpi.rank = 0
 # mpi.size = 0
 # """
 
 #-----------------------------------------------------------------------------
 # Developer level configuration attributes
 #-----------------------------------------------------------------------------
 
 # You shouldn't have to modify anything in this section. These attributes
 # are more for developers who want to change the behavior of the controller
 # at a fundamental level.
 
 # You should not have to change these attributes.
 
-# c.Global.shell_class = 'IPython.kernel.core.interpreter.Interpreter'
-
-# c.Global.furl_file_name = u'ipcontroller-engine.furl'
-
-
-
+# c.Global.url_file_name = u'ipcontroller-engine.furl'
 
 
 
 

IPython/zmq/parallel/controller.py

@@ -1,108 +1,117 b''
 #!/usr/bin/env python
 """The IPython Controller with 0MQ
 This is a collection of one Hub and several Schedulers.
 """
 #-----------------------------------------------------------------------------
 #  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 logging
 from multiprocessing import Process
 
 import zmq
-
+from zmq.devices import ProcessMonitoredQueue
 # internal:
 from IPython.utils.importstring import import_item
 from IPython.utils.traitlets import Int, Str, Instance, List, Bool
 
 from entry_point import signal_children
 
 
 from scheduler import launch_scheduler
 from hub import Hub, HubFactory
 
 #-----------------------------------------------------------------------------
 # Configurable
 #-----------------------------------------------------------------------------
 
 
 class ControllerFactory(HubFactory):
     """Configurable for setting up a Hub and Schedulers."""
     
     usethreads = Bool(False, config=True)
+    # pure-zmq downstream HWM
+    hwm = Int(0, config=True)
     
     # internal
     children = List()
     mq_class = Str('zmq.devices.ProcessMonitoredQueue')
     
     def _usethreads_changed(self, name, old, new):
         self.mq_class = 'zmq.devices.%sMonitoredQueue'%('Thread' if new else 'Process')
         
     def __init__(self, **kwargs):
         super(ControllerFactory, self).__init__(**kwargs)
         self.subconstructors.append(self.construct_schedulers)
     
     def start(self):
         super(ControllerFactory, self).start()
+        child_procs = []
         for child in self.children:
             child.start()
-        if not self.usethreads:
-            signal_children([ getattr(c, 'launcher', c) for c in self.children ])
+            if isinstance(child, ProcessMonitoredQueue):
+                child_procs.append(child.launcher)
+            elif isinstance(child, Process):
+                child_procs.append(child)
+        if child_procs:
+            signal_children(child_procs)
         
     
     def construct_schedulers(self):
         children = self.children
         mq = import_item(self.mq_class)
         
+        maybe_inproc = 'inproc://monitor' if self.usethreads else self.monitor_url
         # IOPub relay (in a Process)
         q = mq(zmq.PUB, zmq.SUB, zmq.PUB, 'N/A','iopub')
         q.bind_in(self.client_info['iopub'])
         q.bind_out(self.engine_info['iopub'])
         q.setsockopt_out(zmq.SUBSCRIBE, '')
-        q.connect_mon(self.monitor_url)
+        q.connect_mon(maybe_inproc)
         q.daemon=True
         children.append(q)
 
         # Multiplexer Queue (in a Process)
         q = mq(zmq.XREP, zmq.XREP, zmq.PUB, 'in', 'out')
         q.bind_in(self.client_info['mux'])
         q.bind_out(self.engine_info['mux'])
-        q.connect_mon(self.monitor_url)
+        q.connect_mon(maybe_inproc)
         q.daemon=True
         children.append(q)
 
         # Control Queue (in a Process)
         q = mq(zmq.XREP, zmq.XREP, zmq.PUB, 'incontrol', 'outcontrol')
         q.bind_in(self.client_info['control'])
         q.bind_out(self.engine_info['control'])
-        q.connect_mon(self.monitor_url)
+        q.connect_mon(maybe_inproc)
         q.daemon=True
         children.append(q)
         # Task Queue (in a Process)
         if self.scheme == 'pure':
             self.log.warn("task::using pure XREQ Task scheduler")
             q = mq(zmq.XREP, zmq.XREQ, zmq.PUB, 'intask', 'outtask')
+            q.setsockopt_out(zmq.HWM, self.hwm)
             q.bind_in(self.client_info['task'][1])
             q.bind_out(self.engine_info['task'])
-            q.connect_mon(self.monitor_url)
+            q.connect_mon(maybe_inproc)
             q.daemon=True
             children.append(q)
         elif self.scheme == 'none':
             self.log.warn("task::using no Task scheduler")
             
         else:
             self.log.info("task::using Python %s Task scheduler"%self.scheme)
             sargs = (self.client_info['task'][1], self.engine_info['task'], self.monitor_url, self.client_info['notification'])
             kwargs = dict(scheme=self.scheme,logname=self.log.name, loglevel=self.log.level, config=self.config)
             q = Process(target=launch_scheduler, args=sargs, kwargs=kwargs)
             q.daemon=True
             children.append(q)
 

IPython/zmq/parallel/hub.py

@@ -1,1053 +1,1054 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
 from datetime import datetime
 import time
 import logging
 
 import zmq
 from zmq.eventloop import ioloop
 from zmq.eventloop.zmqstream import ZMQStream
 
 # internal:
 from IPython.config.configurable import Configurable
-from IPython.utils.traitlets import HasTraits, Instance, Int, Str, Dict, Set, List, Bool
+from IPython.utils.traitlets import HasTraits, Instance, Int, CStr, Str, Dict, Set, List, Bool
 from IPython.utils.importstring import import_item
 
 from entry_point import select_random_ports
 from factory import RegistrationFactory, LoggingFactory
 
 from streamsession import Message, wrap_exception, ISO8601
 from heartmonitor import HeartMonitor
 from util import validate_url_container
 
 try:
     from pymongo.binary import Binary
 except ImportError:
     MongoDB=None
 else:
     from mongodb import MongoDB
     
 #-----------------------------------------------------------------------------
 # Code
 #-----------------------------------------------------------------------------
 
 def _passer(*args, **kwargs):
     return
 
 def _printer(*args, **kwargs):
     print (args)
     print (kwargs)
 
 def init_record(msg):
     """Initialize a TaskRecord based on a request."""
     header = msg['header']
     return {
         'msg_id' : header['msg_id'],
         'header' : header,
         'content': msg['content'],
         'buffers': msg['buffers'],
         'submitted': datetime.strptime(header['date'], ISO8601),
         '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=Str()
     control=Str()
     registration=Str()
     heartbeat=Str()
     pending=Set()
 
 class HubFactory(RegistrationFactory):
     """The Configurable for setting up a Hub."""
     
     # name of a scheduler scheme
     scheme = Str('leastload', config=True)
     
     # port-pairs for monitoredqueues:
     hb = Instance(list, config=True)
     def _hb_default(self):
         return select_random_ports(2)
     
     mux = Instance(list, config=True)
     def _mux_default(self):
         return select_random_ports(2)
     
     task = Instance(list, config=True)
     def _task_default(self):
         return select_random_ports(2)
     
     control = Instance(list, config=True)
     def _control_default(self):
         return select_random_ports(2)
     
     iopub = Instance(list, config=True)
     def _iopub_default(self):
         return select_random_ports(2)
     
     # single ports:
     mon_port = Instance(int, config=True)
     def _mon_port_default(self):
         return select_random_ports(1)[0]
     
     query_port = Instance(int, config=True)
     def _query_port_default(self):
         return select_random_ports(1)[0]
     
     notifier_port = Instance(int, config=True)
     def _notifier_port_default(self):
         return select_random_ports(1)[0]
     
     ping = Int(1000, config=True) # ping frequency
     
-    engine_ip = Str('127.0.0.1', config=True)
-    engine_transport = Str('tcp', config=True)
+    engine_ip = CStr('127.0.0.1', config=True)
+    engine_transport = CStr('tcp', config=True)
     
-    client_ip = Str('127.0.0.1', config=True)
-    client_transport = Str('tcp', config=True)
+    client_ip = CStr('127.0.0.1', config=True)
+    client_transport = CStr('tcp', config=True)
     
-    monitor_ip = Str('127.0.0.1', config=True)
-    monitor_transport = Str('tcp', config=True)
+    monitor_ip = CStr('127.0.0.1', config=True)
+    monitor_transport = CStr('tcp', config=True)
     
-    monitor_url = Str('')
+    monitor_url = CStr('')
     
-    db_class = Str('IPython.zmq.parallel.dictdb.DictDB', config=True)
+    db_class = CStr('IPython.zmq.parallel.dictdb.DictDB', config=True)
     
     # not configurable
     db = Instance('IPython.zmq.parallel.dictdb.BaseDB')
     heartmonitor = Instance('IPython.zmq.parallel.heartmonitor.HeartMonitor')
     subconstructors = List()
     _constructed = Bool(False)
     
     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()
         # self.on_trait_change(self._sync_ips, 'ip')
         # self.on_trait_change(self._sync_transports, 'transport')
         self.subconstructors.append(self.construct_hub)
     
     
     def construct(self):
         assert not self._constructed, "already constructed!"
         
         for subc in self.subconstructors:
             subc()
         
         self._constructed = True
         
     
     def start(self):
         assert self._constructed, "must be constructed by self.construct() first!"
         self.heartmonitor.start()
         self.log.info("Heartmonitor started")
     
     def construct_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
         reg = ZMQStream(ctx.socket(zmq.XREP), loop)
         reg.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:
             reg.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), 
                                 period=self.ping, logname=self.log.name)
 
         ### Client connections ###
         # Clientele socket
         c = ZMQStream(ctx.socket(zmq.XREP), loop)
         c.bind(client_iface%self.query_port)
         # 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.db = import_item(self.db_class)(self.session.session)
         time.sleep(.25)
 
         # 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],
             'query': client_iface%self.query_port,
             'mux': client_iface%self.mux[0],
             'task' : (self.scheme, client_iface%self.task[0]),
             'iopub' : client_iface%self.iopub[0],
             'notification': client_iface%self.notifier_port
             }
         self.log.debug("hub::Hub engine addrs: %s"%self.engine_info)
         self.log.debug("hub::Hub client addrs: %s"%self.client_info)
         self.hub = Hub(loop=loop, session=self.session, monitor=sub, heartmonitor=self.heartmonitor,
                 registrar=reg, clientele=c, notifier=n, db=self.db,
                 engine_info=self.engine_info, client_info=self.client_info,
                 logname=self.log.name)
     
 
 class Hub(LoggingFactory):
     """The IPython Controller Hub with 0MQ connections
     
     Parameters
     ==========
     loop: zmq IOLoop instance
     session: StreamSession object
     <removed> context: zmq context for creating new connections (?)
     queue: ZMQStream for monitoring the command queue (SUB)
     registrar: ZMQStream for engine registration requests (XREP)
     heartbeat: HeartMonitor object checking the pulse of the engines
     clientele: ZMQStream for client connections (XREP)
                 not used for jobs, only query/control commands
     notifier: ZMQStream for broadcasting engine registration changes (PUB)
     db: connection to db for out of memory logging of commands
                 NotImplemented
     engine_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
     # mia=None
     incoming_registrations=Dict()
     registration_timeout=Int()
     _idcounter=Int(0)
     
     # objects from constructor:
     loop=Instance(ioloop.IOLoop)
     registrar=Instance(ZMQStream)
     clientele=Instance(ZMQStream)
     monitor=Instance(ZMQStream)
     heartmonitor=Instance(HeartMonitor)
     notifier=Instance(ZMQStream)
     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
         registrar: ZMQStream for engine registration
         heartbeat: HeartMonitor object for tracking engines
         # client:
         clientele: ZMQStream for client connections
         # extra:
         db: ZMQStream for db connection (NotImplemented)
         engine_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':
                 validate_url_container(v[1])
             else:
                 validate_url_container(v)
         # validate_url_container(self.client_info)
         validate_url_container(self.engine_info)
         
         # register our callbacks
         self.registrar.on_recv(self.dispatch_register_request)
         self.clientele.on_recv(self.dispatch_client_msg)
         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.client_handlers = {'queue_request': self.queue_status,
                                 'result_request': self.get_results,
                                 'purge_request': self.purge_results,
                                 'load_request': self.check_load,
                                 'resubmit_request': self.resubmit_task,
                                 'shutdown_request': self.shutdown_request,
                                 }
         
         self.registrar_handlers = {'registration_request' : self.register_engine,
                                 'unregistration_request' : self.unregister_engine,
                                 'connection_request': self.connection_request,
         }
         
         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
     
     def _validate_client_msg(self, msg):
         """validates and unpacks headers of a message. Returns False if invalid,
         (ident, header, parent, content)"""
         client_id = msg[0]
         try:
             msg = self.session.unpack_message(msg[1:], content=True)
         except:
             self.log.error("client::Invalid Message %s"%msg, exc_info=True)
             return False
         
         msg_type = msg.get('msg_type', None)
         if msg_type is None:
             return False
         header = msg.get('header')
         # session doesn't handle split content for now:
         return client_id, msg
         
     
     #-----------------------------------------------------------------------------
     # dispatch methods (1 per stream)
     #-----------------------------------------------------------------------------
     
     def dispatch_register_request(self, msg):
         """"""
         self.log.debug("registration::dispatch_register_request(%s)"%msg)
         idents,msg = self.session.feed_identities(msg)
         if not idents:
             self.log.error("Bad Queue Message: %s"%msg, exc_info=True)
             return
         try:
             msg = self.session.unpack_message(msg,content=True)
         except:
             self.log.error("registration::got bad registration message: %s"%msg, exc_info=True)
             return
         
         msg_type = msg['msg_type']
         content = msg['content']
         
         handler = self.registrar_handlers.get(msg_type, None)
         if handler is None:
             self.log.error("registration::got bad registration message: %s"%msg)
         else:
             handler(idents, msg)
     
     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: %s"%msg[:2])
         switch = msg[0]
         idents, msg = self.session.feed_identities(msg[1:])
         if not idents:
             self.log.error("Bad Monitor Message: %s"%msg)
             return
         handler = self.monitor_handlers.get(switch, None)
         if handler is not None:
             handler(idents, msg)
         else:
             self.log.error("Invalid monitor topic: %s"%switch)
         
     
     def dispatch_client_msg(self, msg):
         """Route messages from clients"""
         idents, msg = self.session.feed_identities(msg)
         if not idents:
             self.log.error("Bad Client Message: %s"%msg)
             return
         client_id = idents[0]
         try:
             msg = self.session.unpack_message(msg, content=True)
         except:
             content = wrap_exception()
             self.log.error("Bad Client Message: %s"%msg, exc_info=True)
             self.session.send(self.clientele, "hub_error", ident=client_id, 
                     content=content)
             return
         
         # print client_id, header, parent, content
         #switch on message type:
         msg_type = msg['msg_type']
         self.log.info("client:: client %s requested %s"%(client_id, msg_type))
         handler = self.client_handlers.get(msg_type, None)
         try:
             assert handler is not None, "Bad Message Type: %s"%msg_type
         except:
             content = wrap_exception()
             self.log.error("Bad Message Type: %s"%msg_type, exc_info=True)
             self.session.send(self.clientele, "hub_error", ident=client_id, 
                     content=content)
             return
         else:
             handler(client_id, msg)
             
     def dispatch_db(self, msg):
         """"""
         raise NotImplementedError
     
     #---------------------------------------------------------------------------
     # handler methods (1 per event)
     #---------------------------------------------------------------------------
     
     #----------------------- Heartbeat --------------------------------------
     
     def handle_new_heart(self, heart):
         """handler to attach to heartbeater.
         Called when a new heart starts to beat.
         Triggers completion of registration."""
         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: %s"%idents)
             return
         queue_id, client_id = idents[:2]
         try:
             msg = self.session.unpack_message(msg, content=False)
         except:
             self.log.error("queue::client %r sent invalid message to %r: %s"%(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: %s"%(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'
         if MongoDB is not None and isinstance(self.db, MongoDB):
             record['buffers'] = map(Binary, record['buffers'])
         self.pending.add(msg_id)
         self.queues[eid].append(msg_id)
         self.db.add_record(msg_id, record)
     
     def save_queue_result(self, idents, msg):
         if len(idents) < 2:
             self.log.error("invalid identity prefix: %s"%idents)
             return
             
         client_id, queue_id = idents[:2]
         try:
             msg = self.session.unpack_message(msg, content=False)
         except:
             self.log.error("queue::engine %r sent invalid message to %r: %s"%(
                     queue_id,client_id, msg), exc_info=True)
             return
         
         eid = self.by_ident.get(queue_id, None)
         if eid is None:
             self.log.error("queue::unknown engine %r is sending a reply: "%queue_id)
             self.log.debug("queue::       %s"%msg[2:])
             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)
             rheader = msg['header']
             completed = datetime.strptime(rheader['date'], ISO8601)
             started = rheader.get('started', None)
             if started is not None:
                 started = datetime.strptime(started, ISO8601)
             result = {
                 'result_header' : rheader,
                 'result_content': msg['content'],
                 'started' : started,
                 'completed' : completed
             }
             if MongoDB is not None and isinstance(self.db, MongoDB):
                 result['result_buffers'] = map(Binary, msg['buffers'])
             else:
                 result['result_buffers'] = msg['buffers']
             self.db.update_record(msg_id, result)
         else:
             self.log.debug("queue:: unknown msg finished %s"%msg_id)
             
     #--------------------- Task Queue Traffic ------------------------------
     
     def save_task_request(self, idents, msg):
         """Save the submission of a task."""
         client_id = idents[0]
         
         try:
             msg = self.session.unpack_message(msg, content=False)
         except:
             self.log.error("task::client %r sent invalid task message: %s"%(
                     client_id, msg), exc_info=True)
             return
         record = init_record(msg)
         if MongoDB is not None and isinstance(self.db, MongoDB):
             record['buffers'] = map(Binary, record['buffers'])
         record['client_uuid'] = client_id
         record['queue'] = 'task'
         header = msg['header']
         msg_id = header['msg_id']
         self.pending.add(msg_id)
         self.db.add_record(msg_id, record)
     
     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:
             self.log.error("task::invalid task result message send to %r: %s"%(
                     client_id, msg), exc_info=True)
             raise
             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']
         
         header = msg['header']
         engine_uuid = header.get('engine', None)
         eid = self.by_ident.get(engine_uuid, None)
         
         if msg_id in self.pending:
             self.pending.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 = datetime.strptime(header['date'], ISO8601)
             started = header.get('started', None)
             if started is not None:
                 started = datetime.strptime(started, ISO8601)
             result = {
                 'result_header' : header,
                 'result_content': msg['content'],
                 'started' : started,
                 'completed' : completed,
                 'engine_uuid': engine_uuid
             }
             if MongoDB is not None and isinstance(self.db, MongoDB):
                 result['result_buffers'] = map(Binary, msg['buffers'])
             else:
                 result['result_buffers'] = msg['buffers']
             self.db.update_record(msg_id, result)
             
         else:
             self.log.debug("task::unknown task %s finished"%msg_id)
     
     def save_task_destination(self, idents, msg):
         try:
             msg = self.session.unpack_message(msg, content=True)
         except:
             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 %s arrived on %s"%(msg_id, eid))
         # if msg_id in self.mia:
         #     self.mia.remove(msg_id)
         # else:
         #     self.log.debug("task::task %s not listed as MIA?!"%(msg_id))
         
         self.tasks[eid].append(msg_id)
         # self.pending[msg_id][1].update(received=datetime.now(),engine=(eid,engine_uuid))
         self.db.update_record(msg_id, dict(engine_uuid=engine_uuid))
     
     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:
             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: %s"%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:
             self.log.error("iopub::IOPub message has invalid parent", exc_info=True)
             return
         # 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
         else:
             d[msg_type] = content['data']
         
         self.db.update_record(msg_id, d)
         
     
             
     #-------------------------------------------------------------------------
     # Registration requests
     #-------------------------------------------------------------------------
         
     def connection_request(self, client_id, msg):
         """Reply with connection addresses for clients."""
         self.log.info("client::client %s connected"%client_id)
         content = dict(status='ok')
         content.update(self.client_info)
         jsonable = {}
         for k,v in self.keytable.iteritems():
             jsonable[str(k)] = v
         content['engines'] = jsonable
         self.session.send(self.registrar, 'connection_reply', content, parent=msg, ident=client_id)
     
     def register_engine(self, reg, msg):
         """Register a new engine."""
         content = msg['content']
         try:
             queue = content['queue']
         except KeyError:
             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 = 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 = 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 = 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 = wrap_exception()
                     break
         
         msg = self.session.send(self.registrar, "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: %s"%(eid, content['evalue']))
         return eid
     
     def unregister_engine(self, ident, msg):
         """Unregister an engine that explicitly requested to leave."""
         try:
             eid = msg['content']['id']
         except:
             self.log.error("registration::bad engine id for unregistration: %s"%ident, exc_info=True)
             return
         self.log.info("registration::unregister_engine(%s)"%eid)
         content=dict(id=eid, queue=self.engines[eid].queue)
         self.ids.remove(eid)
         self.keytable.pop(eid)
         ec = self.engines.pop(eid)
         self.hearts.pop(ec.heartbeat)
         self.by_ident.pop(ec.queue)
         self.completed.pop(eid)
         for msg_id in self.queues.pop(eid):
             msg = self.pending.remove(msg_id)
             ############## TODO: HANDLE IT ################
         
         if self.notifier:
             self.session.send(self.notifier, "unregistration_notification", content=content)
     
     def finish_registration(self, heart):
         """Second half of engine registration, called after our HeartMonitor
         has received a beat from the Engine's Heart."""
         try: 
             (eid,queue,reg,purge) = self.incoming_registrations.pop(heart)
         except KeyError:
             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."""
         # s = self.context.socket(zmq.XREQ)
         # s.connect(self.client_connections['mux'])
         # time.sleep(0.1)
         # for eid,ec in self.engines.iteritems():
         #     self.session.send(s, 'shutdown_request', content=dict(restart=False), ident=ec.queue)
         # time.sleep(1)
         self.session.send(self.clientele, 'shutdown_reply', content={'status': 'ok'}, ident=client_id)
         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 = wrap_exception()
             self.session.send(self.clientele, "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.clientele, "load_reply", content=content, ident=client_id)
             
     
     def queue_status(self, client_id, msg):
         """Return the Queue status of one or more targets.
         if verbose: return the msg_ids
         else: return len of each type.
         keys: queue (pending MUX jobs)
             tasks (pending Task jobs)
             completed (finished jobs from both queues)"""
         content = msg['content']
         targets = content['targets']
         try:
             targets = self._validate_targets(targets)
         except:
             content = wrap_exception()
             self.session.send(self.clientele, "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}
             # pending
         self.session.send(self.clientele, "queue_reply", content=content, ident=client_id)
     
     def purge_results(self, client_id, msg):
         """Purge results from memory. This method is more valuable before we move
         to a DB based message storage mechanism."""
         content = msg['content']
         msg_ids = content.get('msg_ids', [])
         reply = dict(status='ok')
         if msg_ids == 'all':
             self.db.drop_matching_records(dict(completed={'$ne':None}))
         else:
             for msg_id in msg_ids:
                 if msg_id in self.all_completed:
                     self.db.drop_record(msg_id)
                 else:
                     if msg_id in self.pending:
                         try:
                             raise IndexError("msg pending: %r"%msg_id)
                         except:
                             reply = wrap_exception()
                     else:
                         try:
                             raise IndexError("No such msg: %r"%msg_id)
                         except:
                             reply = wrap_exception()
                     break
             eids = content.get('engine_ids', [])
             for eid in eids:
                 if eid not in self.engines:
                     try:
                         raise IndexError("No such engine: %i"%eid)
                     except:
                         reply = wrap_exception()
                     break
                 msg_ids = self.completed.pop(eid)
                 uid = self.engines[eid].queue
                 self.db.drop_matching_records(dict(engine_uuid=uid, completed={'$ne':None}))
         
         self.session.send(self.clientele, 'purge_reply', content=reply, ident=client_id)
     
     def resubmit_task(self, client_id, msg, buffers):
         """Resubmit a task."""
         raise NotImplementedError
     
     def get_results(self, client_id, msg):
         """Get the result of 1 or more messages."""
         content = msg['content']
         msg_ids = 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:
             content['results'] = {}
             records = self.db.find_records(dict(msg_id={'$in':msg_ids}))
         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:
                     rec = records[msg_id]
                     io_dict = {}
                     for key in 'pyin pyout pyerr stdout stderr'.split():
                             io_dict[key] = rec[key]
                     content[msg_id] = { 'result_content': rec['result_content'],
                                         'header': rec['header'],
                                         'result_header' : rec['result_header'],
                                         'io' : io_dict,
                                       }
                     buffers.extend(map(str, rec['result_buffers']))
             else:
                 try:
                     raise KeyError('No such message: '+msg_id)
                 except:
                     content = wrap_exception()
                 break
         self.session.send(self.clientele, "result_reply", content=content, 
                                             parent=msg, ident=client_id,
                                             buffers=buffers)
 

IPython/zmq/parallel/ipclusterapp.py

@@ -1,593 +1,593 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 re
 import logging
 import os
 import signal
 import logging
 import errno
 
 import zmq
 from zmq.eventloop import ioloop
 
 from IPython.external.argparse import ArgumentParser, SUPPRESS
 from IPython.utils.importstring import import_item
 from IPython.zmq.parallel.clusterdir import (
     ApplicationWithClusterDir, ClusterDirConfigLoader,
     ClusterDirError, PIDFileError
 )
 
 
 #-----------------------------------------------------------------------------
 # Module level variables
 #-----------------------------------------------------------------------------
 
 
-default_config_file_name = u'ipcluster_config.py'
+default_config_file_name = u'ipclusterz_config.py'
 
 
 _description = """\
 Start an IPython cluster for parallel computing.\n\n
 
 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 'ipclusterz start -n 4'. For more complex usage 
 you will typically do 'ipclusterz create -p mycluster', then edit
 configuration files, followed by 'ipclusterz start -p 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
 
 
 #-----------------------------------------------------------------------------
 # Command line options
 #-----------------------------------------------------------------------------
 
 
 class IPClusterAppConfigLoader(ClusterDirConfigLoader):
 
     def _add_arguments(self):
         # Don't call ClusterDirConfigLoader._add_arguments as we don't want
         # its defaults on self.parser. Instead, we will put those on
         # default options on our subparsers.
         
         # This has all the common options that all subcommands use
         parent_parser1 = ArgumentParser(
             add_help=False,
             argument_default=SUPPRESS
         )
         self._add_ipython_dir(parent_parser1)
         self._add_log_level(parent_parser1)
 
         # This has all the common options that other subcommands use
         parent_parser2 = ArgumentParser(
             add_help=False,
             argument_default=SUPPRESS
         )
         self._add_cluster_profile(parent_parser2)
         self._add_cluster_dir(parent_parser2)
         self._add_work_dir(parent_parser2)
         paa = parent_parser2.add_argument
         paa('--log-to-file',
             action='store_true', dest='Global.log_to_file', 
             help='Log to a file in the log directory (default is stdout)')
 
         # Create the object used to create the subparsers.
         subparsers = self.parser.add_subparsers(
             dest='Global.subcommand',
             title='ipcluster subcommands',
             description=
             """ipcluster has a variety of subcommands. The general way of 
             running ipcluster is 'ipclusterz <cmd> [options]'. To get help
             on a particular subcommand do 'ipclusterz <cmd> -h'."""
             # help="For more help, type 'ipclusterz <cmd> -h'",
         )
 
         # The "list" subcommand parser
         parser_list = subparsers.add_parser(
             'list',
             parents=[parent_parser1],
             argument_default=SUPPRESS,
             help="List all clusters in cwd and ipython_dir.",
             description=
             """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>'."""
         )
 
         # The "create" subcommand parser
         parser_create = subparsers.add_parser(
             'create',
             parents=[parent_parser1, parent_parser2],
             argument_default=SUPPRESS,
             help="Create a new cluster directory.",
             description=
             """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, 
             'ipclusterz create -p mycluster', which will put the directory
             in '<ipython_dir>/cluster_mycluster'. 
             """
         )
         paa = parser_create.add_argument
         paa('--reset-config',
             dest='Global.reset_config', action='store_true',
             help=
             """Recopy the default config files to the cluster directory.
             You will loose any modifications you have made to these files.""")
 
         # The "start" subcommand parser
         parser_start = subparsers.add_parser(
             'start',
             parents=[parent_parser1, parent_parser2],
             argument_default=SUPPRESS,
             help="Start a cluster.",
             description=
             """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, 'ipclusterz start -n 4 -p <profile>`,
             otherwise use the '--cluster-dir' option.
             """
         )
         
         paa = parser_start.add_argument
         paa('-n', '--number',
             type=int, dest='Global.n',
             help='The number of engines to start.',
             metavar='Global.n')
         paa('--clean-logs',
             dest='Global.clean_logs', action='store_true',
             help='Delete old log flies before starting.')
         paa('--no-clean-logs',
             dest='Global.clean_logs', action='store_false',
             help="Don't delete old log flies before starting.")
         paa('--daemon',
             dest='Global.daemonize', action='store_true',
             help='Daemonize the ipcluster program. This implies --log-to-file')
         paa('--no-daemon',
             dest='Global.daemonize', action='store_false',
             help="Dont't daemonize the ipcluster program.")
         paa('--delay',
             type=float, dest='Global.delay',
             help="Specify the delay (in seconds) between starting the controller and starting the engine(s).")
 
         # The "stop" subcommand parser
         parser_stop = subparsers.add_parser(
             'stop',
             parents=[parent_parser1, parent_parser2],
             argument_default=SUPPRESS,
             help="Stop a running cluster.",
             description=
             """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, 'ipclusterz stop -p <profile>`, otherwise
             use the '--cluster-dir' option.
             """
         )
         paa = parser_stop.add_argument
         paa('--signal',
             dest='Global.signal', type=int,
             help="The signal number to use in stopping the cluster (default=2).",
             metavar="Global.signal")
         
         # the "engines" subcommand parser
         parser_engines = subparsers.add_parser(
             'engines',
             parents=[parent_parser1, parent_parser2],
             argument_default=SUPPRESS,
             help="Attach some engines to an existing controller or cluster.",
             description=
             """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, 'ipclusterz engines -n 4 -p <profile>`,
             otherwise use the '--cluster-dir' option.
             """
         )
         paa = parser_engines.add_argument
         paa('-n', '--number',
             type=int, dest='Global.n',
             help='The number of engines to start.',
             metavar='Global.n')
         paa('--daemon',
             dest='Global.daemonize', action='store_true',
             help='Daemonize the ipcluster program. This implies --log-to-file')
         paa('--no-daemon',
             dest='Global.daemonize', action='store_false',
             help="Dont't daemonize the ipcluster program.")
 
 #-----------------------------------------------------------------------------
 # Main application
 #-----------------------------------------------------------------------------
 
 
 class IPClusterApp(ApplicationWithClusterDir):
 
     name = u'ipclusterz'
     description = _description
     usage = None
     command_line_loader = IPClusterAppConfigLoader
     default_config_file_name = default_config_file_name
     default_log_level = logging.INFO
     auto_create_cluster_dir = False
 
     def create_default_config(self):
         super(IPClusterApp, self).create_default_config()
         self.default_config.Global.controller_launcher = \
             'IPython.zmq.parallel.launcher.LocalControllerLauncher'
         self.default_config.Global.engine_launcher = \
             'IPython.zmq.parallel.launcher.LocalEngineSetLauncher'
         self.default_config.Global.n = 2
         self.default_config.Global.delay = 2
         self.default_config.Global.reset_config = False
         self.default_config.Global.clean_logs = True
         self.default_config.Global.signal = signal.SIGINT
         self.default_config.Global.daemonize = False
 
     def find_resources(self):
         subcommand = self.command_line_config.Global.subcommand
         if subcommand=='list':
             self.list_cluster_dirs()
             # Exit immediately because there is nothing left to do.
             self.exit()
         elif subcommand=='create':
             self.auto_create_cluster_dir = True
             super(IPClusterApp, self).find_resources()
         elif subcommand=='start' or subcommand=='stop':
             self.auto_create_cluster_dir = True
             try:
                 super(IPClusterApp, self).find_resources()
             except ClusterDirError:
                 raise ClusterDirError(
                     "Could not find a cluster directory. A cluster dir must "
                     "be created before running 'ipclusterz start'.  Do "
                     "'ipclusterz create -h' or 'ipclusterz list -h' for more "
                     "information about creating and listing cluster dirs."
                 )
         elif subcommand=='engines':
             self.auto_create_cluster_dir = False
             try:
                 super(IPClusterApp, self).find_resources()
             except ClusterDirError:
                 raise ClusterDirError(
                     "Could not find a cluster directory. A cluster dir must "
                     "be created before running 'ipclusterz start'.  Do "
                     "'ipclusterz create -h' or 'ipclusterz list -h' for more "
                     "information about creating and listing cluster dirs."
                 )
 
     def list_cluster_dirs(self):
         # Find the search paths
         cluster_dir_paths = os.environ.get('IPCLUSTER_DIR_PATH','')
         if cluster_dir_paths:
             cluster_dir_paths = cluster_dir_paths.split(':')
         else:
             cluster_dir_paths = []
         try:
             ipython_dir = self.command_line_config.Global.ipython_dir
         except AttributeError:
             ipython_dir = self.default_config.Global.ipython_dir
         paths = [os.getcwd(), ipython_dir] + \
             cluster_dir_paths
         paths = list(set(paths))
 
         self.log.info('Searching for cluster dirs 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('cluster_'):
                     profile = full_path.split('_')[-1]
                     start_cmd = 'ipclusterz start -p %s -n 4' % profile
                     print start_cmd + " ==> " + full_path
 
     def pre_construct(self):
         # IPClusterApp.pre_construct() is where we cd to the working directory.
         super(IPClusterApp, self).pre_construct()
         config = self.master_config
         try:
             daemon = config.Global.daemonize
             if daemon:
                 config.Global.log_to_file = True
         except AttributeError:
             pass
 
     def construct(self):
         config = self.master_config
         subcmd = config.Global.subcommand
         reset = config.Global.reset_config
         if subcmd == 'list':
             return
         if subcmd == 'create':
             self.log.info('Copying default config files to cluster directory '
             '[overwrite=%r]' % (reset,))
             self.cluster_dir_obj.copy_all_config_files(overwrite=reset)
         if subcmd =='start':
             self.cluster_dir_obj.copy_all_config_files(overwrite=False)
             self.start_logging()
             self.loop = ioloop.IOLoop.instance()
             # reactor.callWhenRunning(self.start_launchers)
             dc = ioloop.DelayedCallback(self.start_launchers, 0, self.loop)
             dc.start()
         if subcmd == 'engines':
             self.start_logging()
             self.loop = ioloop.IOLoop.instance()
             # reactor.callWhenRunning(self.start_launchers)
             engine_only = lambda : self.start_launchers(controller=False)
             dc = ioloop.DelayedCallback(engine_only, 0, self.loop)
             dc.start()
 
     def start_launchers(self, controller=True):
         config = self.master_config
 
         # Create the launchers. In both bases, we set the work_dir of
         # the launcher to the cluster_dir. This is where the launcher's
         # subprocesses will be launched. It is not where the controller
         # and engine will be launched.
         if controller:
             cl_class = import_item(config.Global.controller_launcher)
             self.controller_launcher = cl_class(
                 work_dir=self.cluster_dir, config=config,
                 logname=self.log.name
             )
             # Setup the observing of stopping. If the controller dies, shut
             # everything down as that will be completely fatal for the engines.
             self.controller_launcher.on_stop(self.stop_launchers)
             # But, we don't monitor the stopping of engines. An engine dying
             # is just fine and in principle a user could start a new engine.
             # Also, if we did monitor engine stopping, it is difficult to
             # know what to do when only some engines die. Currently, the
             # observing of engine stopping is inconsistent. Some launchers
             # might trigger on a single engine stopping, other wait until
             # all stop.  TODO: think more about how to handle this.
         else:
             self.controller_launcher = None
         
         el_class = import_item(config.Global.engine_launcher)
         self.engine_launcher = el_class(
             work_dir=self.cluster_dir, config=config, logname=self.log.name
         )
 
         # Setup signals
         signal.signal(signal.SIGINT, self.sigint_handler)
 
         # Start the controller and engines
         self._stopping = False  # Make sure stop_launchers is not called 2x.
         if controller:
             self.start_controller()
         dc = ioloop.DelayedCallback(self.start_engines, 1000*config.Global.delay*controller, self.loop)
         dc.start()
         self.startup_message()
 
     def startup_message(self, r=None):
         self.log.info("IPython cluster: started")
         return r
         
     def start_controller(self, r=None):
         # self.log.info("In start_controller")
         config = self.master_config
         d = self.controller_launcher.start(
             cluster_dir=config.Global.cluster_dir
         )
         return d
             
     def start_engines(self, r=None):
         # self.log.info("In start_engines")
         config = self.master_config
         
         d = self.engine_launcher.start(
             config.Global.n,
             cluster_dir=config.Global.cluster_dir
         )
         return d
 
     def stop_controller(self, r=None):
         # self.log.info("In stop_controller")
         if self.controller_launcher and self.controller_launcher.running:
             return self.controller_launcher.stop()
 
     def stop_engines(self, r=None):
         # self.log.info("In stop_engines")
         if self.engine_launcher.running:
             d = self.engine_launcher.stop()
             # d.addErrback(self.log_err)
             return d
         else:
             return None
 
     def log_err(self, f):
         self.log.error(f.getTraceback())
         return None
         
     def stop_launchers(self, r=None):
         if not self._stopping:
             self._stopping = True
             # if isinstance(r, failure.Failure):
             #     self.log.error('Unexpected error in ipcluster:')
             #     self.log.info(r.getTraceback())
             self.log.error("IPython cluster: stopping")
             # These return deferreds. We are not doing anything with them
             # but we are holding refs to them as a reminder that they 
             # do return deferreds.
             d1 = self.stop_engines()
             d2 = self.stop_controller()
             # Wait a few seconds to let things shut down.
             dc = ioloop.DelayedCallback(self.loop.stop, 4000, self.loop)
             dc.start()
             # reactor.callLater(4.0, reactor.stop)
 
     def sigint_handler(self, signum, frame):
         self.stop_launchers()
         
     def start_logging(self):
         # Remove old log files of the controller and engine
         if self.master_config.Global.clean_logs:
             log_dir = self.master_config.Global.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(IPClusterApp, self).start_logging()
 
     def start_app(self):
         """Start the application, depending on what subcommand is used."""
         subcmd = self.master_config.Global.subcommand
         if subcmd=='create' or subcmd=='list':
             return
         elif subcmd=='start':
             self.start_app_start()
         elif subcmd=='stop':
             self.start_app_stop()
         elif subcmd=='engines':
             self.start_app_engines()
 
     def start_app_start(self):
         """Start the app for the start subcommand."""
         config = self.master_config
         # First see if the cluster is already running
         try:
             pid = self.get_pid_from_file()
         except PIDFileError:
             pass
         else:
             self.log.critical(
                 'Cluster is already running with [pid=%s]. '
                 'use "ipclusterz 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)
 
         # Now log and daemonize
         self.log.info(
             'Starting ipclusterz with [daemon=%r]' % config.Global.daemonize
         )
         # TODO: Get daemonize working on Windows or as a Windows Server.
         if config.Global.daemonize:
             if os.name=='posix':
                 from twisted.scripts._twistd_unix import daemonize
                 daemonize()
 
         # 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
         self.remove_pid_file()
 
     def start_app_engines(self):
         """Start the app for the start subcommand."""
         config = self.master_config
         # First see if the cluster is already running
         
         # Now log and daemonize
         self.log.info(
             'Starting engines with [daemon=%r]' % config.Global.daemonize
         )
         # TODO: Get daemonize working on Windows or as a Windows Server.
         if config.Global.daemonize:
             if os.name=='posix':
                 from twisted.scripts._twistd_unix import daemonize
                 daemonize()
 
         # 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
         # self.remove_pid_file()
     
     def start_app_stop(self):
         """Start the app for the stop subcommand."""
         config = self.master_config
         try:
             pid = self.get_pid_from_file()
         except PIDFileError:
             self.log.critical(
                 'Problem reading 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.exit(ALREADY_STOPPED)
         else:
             if os.name=='posix':
                 sig = config.Global.signal
                 self.log.info(
                     "Stopping cluster [pid=%r] with [signal=%r]" % (pid, sig)
                 )
                 os.kill(pid, sig)
             elif os.name=='nt':
                 # As of right now, we don't support daemonize on Windows, so
                 # stop will not do anything. Minimally, it should clean up the
                 # old .pid files.
                 self.remove_pid_file()
 
 
 def launch_new_instance():
     """Create and run the IPython cluster."""
     app = IPClusterApp()
     app.start()
 
 
 if __name__ == '__main__':
     launch_new_instance()
 

IPython/zmq/parallel/ipcontrollerapp.py

@@ -1,390 +1,428 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 The IPython controller 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
 #-----------------------------------------------------------------------------
 
 from __future__ import with_statement
 
 import copy
 import sys
 import os
 import logging
 import stat
 import socket
 
 import uuid
 
 import zmq
 from zmq.log.handlers import PUBHandler
 from zmq.utils import jsonapi as json
 
 from IPython.config.loader import Config
 from IPython.zmq.parallel import factory
 from IPython.zmq.parallel.controller import ControllerFactory
 from IPython.zmq.parallel.clusterdir import (
     ApplicationWithClusterDir,
     ClusterDirConfigLoader
 )
 # from IPython.kernel.fcutil import FCServiceFactory, FURLError
 from IPython.utils.traitlets import Instance, Unicode
 
 from util import disambiguate_ip_address, split_url
 
 
 #-----------------------------------------------------------------------------
 # Module level variables
 #-----------------------------------------------------------------------------
 
 
 #: The default config file name for this application
-default_config_file_name = u'ipcontroller_config.py'
+default_config_file_name = u'ipcontrollerz_config.py'
 
 
 _description = """Start the IPython controller for parallel computing.
 
 The IPython controller provides a gateway between the IPython engines and
 clients. The controller needs to be started before the engines and 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
+your ipython directory and named as "cluster_<profile>". See the --profile
 and --cluster-dir options for details.
 """
 
 #-----------------------------------------------------------------------------
 # Default interfaces
 #-----------------------------------------------------------------------------
 
 # The default client interfaces for FCClientServiceFactory.interfaces
 default_client_interfaces = Config()
 default_client_interfaces.Default.url_file = 'ipcontroller-client.url'
 
 # Make this a dict we can pass to Config.__init__ for the default
 default_client_interfaces = dict(copy.deepcopy(default_client_interfaces.items()))
 
 
 
 # The default engine interfaces for FCEngineServiceFactory.interfaces
 default_engine_interfaces = Config()
 default_engine_interfaces.Default.url_file = u'ipcontroller-engine.url'
 
 # Make this a dict we can pass to Config.__init__ for the default
 default_engine_interfaces = dict(copy.deepcopy(default_engine_interfaces.items()))
 
 
 #-----------------------------------------------------------------------------
 # Service factories
 #-----------------------------------------------------------------------------
 
 # 
 # class FCClientServiceFactory(FCServiceFactory):
 #     """A Foolscap implementation of the client services."""
 # 
 #     cert_file = Unicode(u'ipcontroller-client.pem', config=True)
 #     interfaces = Instance(klass=Config, kw=default_client_interfaces,
 #                           allow_none=False, config=True)
 # 
 # 
 # class FCEngineServiceFactory(FCServiceFactory):
 #     """A Foolscap implementation of the engine services."""
 # 
 #     cert_file = Unicode(u'ipcontroller-engine.pem', config=True)
 #     interfaces = Instance(klass=dict, kw=default_engine_interfaces,
 #                           allow_none=False, config=True)
 # 
 
 #-----------------------------------------------------------------------------
 # Command line options
 #-----------------------------------------------------------------------------
 
 
 class IPControllerAppConfigLoader(ClusterDirConfigLoader):
 
     def _add_arguments(self):
         super(IPControllerAppConfigLoader, self)._add_arguments()
         paa = self.parser.add_argument
         
         ## Hub Config:
         paa('--mongodb', 
             dest='HubFactory.db_class', action='store_const',
             const='IPython.zmq.parallel.mongodb.MongoDB', 
             help='Use MongoDB task storage [default: in-memory]')
         paa('--hb',
             type=int, dest='HubFactory.hb', nargs=2,
             help='The (2) ports the Hub\'s Heartmonitor will use for the heartbeat '
             'connections [default: random]',
             metavar='Hub.hb_ports')
         paa('--ping',
             type=int, dest='HubFactory.ping',
             help='The frequency at which the Hub pings the engines for heartbeats '
             ' (in ms) [default: 100]',
             metavar='Hub.ping')
         
         # Client config
         paa('--client-ip',
             type=str, dest='HubFactory.client_ip', 
             help='The IP address or hostname the Hub will listen on for '
             'client connections. Both engine-ip and client-ip can be set simultaneously '
             'via --ip [default: loopback]',
             metavar='Hub.client_ip')
         paa('--client-transport',
             type=str, dest='HubFactory.client_transport', 
             help='The ZeroMQ transport the Hub will use for '
             'client connections. Both engine-transport and client-transport can be set simultaneously '
             'via --transport [default: tcp]',
             metavar='Hub.client_transport')
         paa('--query',
             type=int, dest='HubFactory.query_port', 
             help='The port on which the Hub XREP socket will listen for result queries from clients [default: random]',
             metavar='Hub.query_port')
         paa('--notifier',
             type=int, dest='HubFactory.notifier_port', 
             help='The port on which the Hub PUB socket will listen for notification connections [default: random]',
             metavar='Hub.notifier_port')
         
         # Engine config
         paa('--engine-ip',
             type=str, dest='HubFactory.engine_ip', 
             help='The IP address or hostname the Hub will listen on for '
             'engine connections. This applies to the Hub and its schedulers'
             'engine-ip and client-ip can be set simultaneously '
             'via --ip [default: loopback]',
             metavar='Hub.engine_ip')
         paa('--engine-transport',
             type=str, dest='HubFactory.engine_transport', 
             help='The ZeroMQ transport the Hub will use for '
             'client connections. Both engine-transport and client-transport can be set simultaneously '
             'via --transport [default: tcp]',
             metavar='Hub.engine_transport')
         
         # Scheduler config
         paa('--mux',
             type=int, dest='ControllerFactory.mux', nargs=2,
             help='The (2) ports the MUX scheduler will listen on for client,engine '
             'connections, respectively [default: random]',
             metavar='Scheduler.mux_ports')
         paa('--task',
             type=int, dest='ControllerFactory.task', nargs=2,
             help='The (2) ports the Task scheduler will listen on for client,engine '
             'connections, respectively [default: random]',
             metavar='Scheduler.task_ports')
         paa('--control',
             type=int, dest='ControllerFactory.control', nargs=2,
             help='The (2) ports the Control scheduler will listen on for client,engine '
             'connections, respectively [default: random]',
             metavar='Scheduler.control_ports')
         paa('--iopub',
             type=int, dest='ControllerFactory.iopub', nargs=2,
             help='The (2) ports the IOPub scheduler will listen on for client,engine '
             'connections, respectively [default: random]',
             metavar='Scheduler.iopub_ports')
+            
         paa('--scheme',
             type=str, dest='HubFactory.scheme',
             choices = ['pure', 'lru', 'plainrandom', 'weighted', 'twobin','leastload'],
             help='select the task scheduler scheme  [default: Python LRU]',
             metavar='Scheduler.scheme')
         paa('--usethreads',
             dest='ControllerFactory.usethreads', action="store_true",
             help='Use threads instead of processes for the schedulers',
             )
+        paa('--hwm',
+            dest='ControllerFactory.hwm', type=int,
+            help='specify the High Water Mark (HWM) for the downstream '
+            'socket in the pure ZMQ scheduler. This is the maximum number '
+            'of allowed outstanding tasks on each engine.',
+            )
         
         ## Global config
         paa('--log-to-file',
             action='store_true', dest='Global.log_to_file',
             help='Log to a file in the log directory (default is stdout)')
         paa('--log-url',
             type=str, dest='Global.log_url',
             help='Broadcast logs to an iploggerz process [default: disabled]')
-        paa('-r','--reuse-key', 
-            action='store_true', dest='Global.reuse_key',
-            help='Try to reuse existing execution keys.')
+        paa('-r','--reuse-files', 
+            action='store_true', dest='Global.reuse_files',
+            help='Try to reuse existing json connection files.')
         paa('--no-secure',
             action='store_false', dest='Global.secure',
             help='Turn off execution keys (default).')
         paa('--secure',
             action='store_true', dest='Global.secure',
             help='Turn on execution keys.')
         paa('--execkey',
             type=str, dest='Global.exec_key',
             help='path to a file containing an execution key.',
             metavar='keyfile')
         paa('--ssh',
             type=str, dest='Global.sshserver',
             help='ssh url for clients to use when connecting to the Controller '
             'processes. It should be of the form: [user@]server[:port]. The '
             'Controller\'s listening addresses must be accessible from the ssh server',
             metavar='Global.sshserver')
         paa('--location',
             type=str, dest='Global.location',
             help="The external IP or domain name of this machine, used for disambiguating "
             "engine and client connections.",
             metavar='Global.location')
         factory.add_session_arguments(self.parser)
         factory.add_registration_arguments(self.parser)
 
 
 #-----------------------------------------------------------------------------
 # The main application
 #-----------------------------------------------------------------------------
 
 
 class IPControllerApp(ApplicationWithClusterDir):
 
     name = u'ipcontrollerz'
     description = _description
     command_line_loader = IPControllerAppConfigLoader
     default_config_file_name = default_config_file_name
     auto_create_cluster_dir = True
     
 
     def create_default_config(self):
         super(IPControllerApp, self).create_default_config()
         # Don't set defaults for Global.secure or Global.reuse_furls
         # as those are set in a component.
         self.default_config.Global.import_statements = []
         self.default_config.Global.clean_logs = True
         self.default_config.Global.secure = True
-        self.default_config.Global.reuse_key = False
+        self.default_config.Global.reuse_files = False
         self.default_config.Global.exec_key = "exec_key.key"
         self.default_config.Global.sshserver = None
         self.default_config.Global.location = None
 
     def pre_construct(self):
         super(IPControllerApp, self).pre_construct()
         c = self.master_config
         # The defaults for these are set in FCClientServiceFactory and
         # FCEngineServiceFactory, so we only set them here if the global
         # options have be set to override the class level defaults.
         
         # if hasattr(c.Global, 'reuse_furls'):
         #     c.FCClientServiceFactory.reuse_furls = c.Global.reuse_furls
         #     c.FCEngineServiceFactory.reuse_furls = c.Global.reuse_furls
         #     del c.Global.reuse_furls
         # if hasattr(c.Global, 'secure'):
         #     c.FCClientServiceFactory.secure = c.Global.secure
         #     c.FCEngineServiceFactory.secure = c.Global.secure
         #     del c.Global.secure
     
     def save_connection_dict(self, fname, cdict):
         """save a connection dict to json file."""
         c = self.master_config
         url = cdict['url']
         location = cdict['location']
         if not location:
             try:
                 proto,ip,port = split_url(url)
             except AssertionError:
                 pass
             else:
                 location = socket.gethostbyname_ex(socket.gethostname())[2][-1]
             cdict['location'] = location
         fname = os.path.join(c.Global.security_dir, fname)
         with open(fname, 'w') as f:
             f.write(json.dumps(cdict, indent=2))
         os.chmod(fname, stat.S_IRUSR|stat.S_IWUSR)
+    
+    def load_config_from_json(self):
+        """load config from existing json connector files."""
+        c = self.master_config
+        # load from engine config
+        with open(os.path.join(c.Global.security_dir, 'ipcontroller-engine.json')) as f:
+            cfg = json.loads(f.read())
+        key = c.SessionFactory.exec_key = cfg['exec_key']
+        xport,addr = cfg['url'].split('://')
+        c.HubFactory.engine_transport = xport
+        ip,ports = addr.split(':')
+        c.HubFactory.engine_ip = ip
+        c.HubFactory.regport = int(ports)
+        c.Global.location = cfg['location']
         
+        # load client config
+        with open(os.path.join(c.Global.security_dir, 'ipcontroller-client.json')) as f:
+            cfg = json.loads(f.read())
+        assert key == cfg['exec_key'], "exec_key mismatch between engine and client keys"
+        xport,addr = cfg['url'].split('://')
+        c.HubFactory.client_transport = xport
+        ip,ports = addr.split(':')
+        c.HubFactory.client_ip = ip
+        c.Global.sshserver = cfg['ssh']
+        assert int(ports) == c.HubFactory.regport, "regport mismatch"
+    
     def construct(self):
         # This is the working dir by now.
         sys.path.insert(0, '')
         c = self.master_config
         
         self.import_statements()
-
-        if c.Global.secure:
+        reusing = c.Global.reuse_files
+        if reusing:
+            try:
+                self.load_config_from_json()
+            except (AssertionError,IOError):
+                reusing=False
+        # check again, because reusing may have failed:
+        if reusing:
+            pass
+        elif c.Global.secure:
             keyfile = os.path.join(c.Global.security_dir, c.Global.exec_key)
-            if not c.Global.reuse_key or not os.path.exists(keyfile):
-                key = str(uuid.uuid4())
-                with open(keyfile, 'w') as f:
-                    f.write(key)
-                os.chmod(keyfile, stat.S_IRUSR|stat.S_IWUSR)
-            else:
-                with open(keyfile) as f:
-                    key = f.read().strip()
+            key = str(uuid.uuid4())
+            with open(keyfile, 'w') as f:
+                f.write(key)
+            os.chmod(keyfile, stat.S_IRUSR|stat.S_IWUSR)
             c.SessionFactory.exec_key = key
         else:
             c.SessionFactory.exec_key = ''
             key = None
         
         try:
             self.factory = ControllerFactory(config=c, logname=self.log.name)
             self.start_logging()
             self.factory.construct()
         except:
             self.log.error("Couldn't construct the Controller", exc_info=True)
             self.exit(1)
         
-        f = self.factory
-        cdict = {'exec_key' : key,
-                'ssh' : c.Global.sshserver,
-                'url' : "%s://%s:%s"%(f.client_transport, f.client_ip, f.regport),
-                'location' : c.Global.location
-                }
-        self.save_connection_dict('ipcontroller-client.json', cdict)
-        edict = cdict
-        edict['url']="%s://%s:%s"%((f.client_transport, f.client_ip, f.regport))
-        self.save_connection_dict('ipcontroller-engine.json', edict)
+        if not reusing:
+            # save to new json config files
+            f = self.factory
+            cdict = {'exec_key' : key,
+                    'ssh' : c.Global.sshserver,
+                    'url' : "%s://%s:%s"%(f.client_transport, f.client_ip, f.regport),
+                    'location' : c.Global.location
+                    }
+            self.save_connection_dict('ipcontroller-client.json', cdict)
+            edict = cdict
+            edict['url']="%s://%s:%s"%((f.client_transport, f.client_ip, f.regport))
+            self.save_connection_dict('ipcontroller-engine.json', edict)
         
     
     def save_urls(self):
         """save the registration urls to files."""
         c = self.master_config
         
         sec_dir = c.Global.security_dir
         cf = self.factory
         
         with open(os.path.join(sec_dir, 'ipcontroller-engine.url'), 'w') as f:
             f.write("%s://%s:%s"%(cf.engine_transport, cf.engine_ip, cf.regport))
         
         with open(os.path.join(sec_dir, 'ipcontroller-client.url'), 'w') as f:
             f.write("%s://%s:%s"%(cf.client_transport, cf.client_ip, cf.regport))
         
     
     def import_statements(self):
         statements = self.master_config.Global.import_statements
         for s in statements:
             try:
                 self.log.msg("Executing statement: '%s'" % s)
                 exec s in globals(), locals()
             except:
                 self.log.msg("Error running statement: %s" % s)
 
     def start_logging(self):
         super(IPControllerApp, self).start_logging()
         if self.master_config.Global.log_url:
             context = self.factory.context
             lsock = context.socket(zmq.PUB)
             lsock.connect(self.master_config.Global.log_url)
             handler = PUBHandler(lsock)
             handler.root_topic = 'controller'
             handler.setLevel(self.log_level)
             self.log.addHandler(handler)
     # 
     def start_app(self):
         # Start the subprocesses:
         self.factory.start()
         self.write_pid_file(overwrite=True)
         try:
             self.factory.loop.start()
         except KeyboardInterrupt:
             self.log.critical("Interrupted, Exiting...\n")
 
 
 def launch_new_instance():
     """Create and run the IPython controller"""
     app = IPControllerApp()
     app.start()
 
 
 if __name__ == '__main__':
     launch_new_instance()

IPython/zmq/parallel/ipengineapp.py

@@ -1,294 +1,294 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 The IPython engine 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 os
 import sys
 import json
 
 import zmq
 from zmq.eventloop import ioloop
 
 from IPython.zmq.parallel.clusterdir import (
     ApplicationWithClusterDir,
     ClusterDirConfigLoader
 )
 from IPython.zmq.log import EnginePUBHandler
 
 from IPython.zmq.parallel import factory
 from IPython.zmq.parallel.engine import EngineFactory
 from IPython.zmq.parallel.streamkernel import Kernel
 from IPython.utils.importstring import import_item
 
 from util import disambiguate_url
 
 #-----------------------------------------------------------------------------
 # Module level variables
 #-----------------------------------------------------------------------------
 
 #: The default config file name for this application
-default_config_file_name = u'ipengine_config.py'
+default_config_file_name = u'ipenginez_config.py'
 
 
 mpi4py_init = """from mpi4py import MPI as mpi
 mpi.size = mpi.COMM_WORLD.Get_size()
 mpi.rank = mpi.COMM_WORLD.Get_rank()
 """
 
 
 pytrilinos_init = """from PyTrilinos import Epetra
 class SimpleStruct:
 pass
 mpi = SimpleStruct()
 mpi.rank = 0
 mpi.size = 0
 """
 
 
 _description = """Start an IPython engine for parallel computing.\n\n
 
 IPython engines run in parallel and perform computations on behalf of a client
 and controller. A controller needs to be started before the engines. The
 engine 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>".
+usually located in your ipython directory and named as "cluster_<profile>".
 See the --profile and --cluster-dir options for details.
 """
 
 #-----------------------------------------------------------------------------
 # Command line options
 #-----------------------------------------------------------------------------
 
 
 class IPEngineAppConfigLoader(ClusterDirConfigLoader):
 
     def _add_arguments(self):
         super(IPEngineAppConfigLoader, self)._add_arguments()
         paa = self.parser.add_argument
         # Controller config
-        paa('--file',
+        paa('--file', '-f',
             type=unicode, dest='Global.url_file',
             help='The full location of the file containing the connection information fo '
             'controller. If this is not given, the file must be in the '
             'security directory of the cluster directory.  This location is '
             'resolved using the --profile and --app-dir options.',
             metavar='Global.url_file')
         # MPI
         paa('--mpi',
             type=str, dest='MPI.use',
             help='How to enable MPI (mpi4py, pytrilinos, or empty string to disable).',
             metavar='MPI.use')
         # Global config
         paa('--log-to-file',
             action='store_true', dest='Global.log_to_file',
             help='Log to a file in the log directory (default is stdout)')
         paa('--log-url',
             dest='Global.log_url',
             help="url of ZMQ logger, as started with iploggerz")
         # paa('--execkey',
         #     type=str, dest='Global.exec_key',
         #     help='path to a file containing an execution key.',
         #     metavar='keyfile')
         # paa('--no-secure',
         #     action='store_false', dest='Global.secure',
         #     help='Turn off execution keys.')
         # paa('--secure',
         #     action='store_true', dest='Global.secure',
         #     help='Turn on execution keys (default).')
         # init command
         paa('-c',
             type=str, dest='Global.extra_exec_lines',
             help='specify a command to be run at startup')
         
         factory.add_session_arguments(self.parser)
         factory.add_registration_arguments(self.parser)
 
 
 #-----------------------------------------------------------------------------
 # Main application
 #-----------------------------------------------------------------------------
 
 
 class IPEngineApp(ApplicationWithClusterDir):
 
     name = u'ipenginez'
     description = _description
     command_line_loader = IPEngineAppConfigLoader
     default_config_file_name = default_config_file_name
     auto_create_cluster_dir = True
 
     def create_default_config(self):
         super(IPEngineApp, self).create_default_config()
 
         # The engine should not clean logs as we don't want to remove the
         # active log files of other running engines.
         self.default_config.Global.clean_logs = False
         self.default_config.Global.secure = True
 
         # Global config attributes
         self.default_config.Global.exec_lines = []
         self.default_config.Global.extra_exec_lines = ''
 
         # Configuration related to the controller
         # This must match the filename (path not included) that the controller
         # used for the FURL file.
         self.default_config.Global.url_file = u''
         self.default_config.Global.url_file_name = u'ipcontroller-engine.json'
         # If given, this is the actual location of the controller's FURL file.
         # If not, this is computed using the profile, app_dir and furl_file_name
         # self.default_config.Global.key_file_name = u'exec_key.key'
         # self.default_config.Global.key_file = u''
 
         # MPI related config attributes
         self.default_config.MPI.use = ''
         self.default_config.MPI.mpi4py = mpi4py_init
         self.default_config.MPI.pytrilinos = pytrilinos_init
 
     def post_load_command_line_config(self):
         pass
 
     def pre_construct(self):
         super(IPEngineApp, self).pre_construct()
         # self.find_cont_url_file()
         self.find_url_file()
         if self.master_config.Global.extra_exec_lines:
             self.master_config.Global.exec_lines.append(self.master_config.Global.extra_exec_lines)
 
     # def find_key_file(self):
     #     """Set the key file.
     # 
     #     Here we don't try to actually see if it exists for is valid as that
     #     is hadled by the connection logic.
     #     """
     #     config = self.master_config
     #     # Find the actual controller key file
     #     if not config.Global.key_file:
     #         try_this = os.path.join(
     #             config.Global.cluster_dir, 
     #             config.Global.security_dir,
     #             config.Global.key_file_name
     #         )
     #         config.Global.key_file = try_this
         
     def find_url_file(self):
         """Set the key file.
 
         Here we don't try to actually see if it exists for is valid as that
         is hadled by the connection logic.
         """
         config = self.master_config
         # Find the actual controller key file
         if not config.Global.url_file:
             try_this = os.path.join(
                 config.Global.cluster_dir, 
                 config.Global.security_dir,
                 config.Global.url_file_name
             )
             config.Global.url_file = try_this
         
     def construct(self):
         # This is the working dir by now.
         sys.path.insert(0, '')
         config = self.master_config
         # if os.path.exists(config.Global.key_file) and config.Global.secure:
         #     config.SessionFactory.exec_key = config.Global.key_file
         if os.path.exists(config.Global.url_file):
             with open(config.Global.url_file) as f:
                 d = json.loads(f.read())
                 for k,v in d.iteritems():
                     if isinstance(v, unicode):
                         d[k] = v.encode()
             if d['exec_key']:
                 config.SessionFactory.exec_key = d['exec_key']
             d['url'] = disambiguate_url(d['url'], d['location'])
             config.RegistrationFactory.url=d['url']
             config.EngineFactory.location = d['location']
         
         
         
         config.Kernel.exec_lines = config.Global.exec_lines
 
         self.start_mpi()
 
         # Create the underlying shell class and EngineService
         # shell_class = import_item(self.master_config.Global.shell_class)
         try:
             self.engine = EngineFactory(config=config, logname=self.log.name)
         except:
             self.log.error("Couldn't start the Engine", exc_info=True)
             self.exit(1)
         
         self.start_logging()
 
         # Create the service hierarchy
         # self.main_service = service.MultiService()
         # self.engine_service.setServiceParent(self.main_service)
         # self.tub_service = Tub()
         # self.tub_service.setServiceParent(self.main_service)
         # # This needs to be called before the connection is initiated
         # self.main_service.startService()
 
         # This initiates the connection to the controller and calls
         # register_engine to tell the controller we are ready to do work
         # self.engine_connector = EngineConnector(self.tub_service)
 
         # self.log.info("Using furl file: %s" % self.master_config.Global.furl_file)
 
         # reactor.callWhenRunning(self.call_connect)
 
 
     def start_logging(self):
         super(IPEngineApp, self).start_logging()
         if self.master_config.Global.log_url:
             context = self.engine.context
             lsock = context.socket(zmq.PUB)
             lsock.connect(self.master_config.Global.log_url)
             handler = EnginePUBHandler(self.engine, lsock)
             handler.setLevel(self.log_level)
             self.log.addHandler(handler)
     
     def start_mpi(self):
         global mpi
         mpikey = self.master_config.MPI.use
         mpi_import_statement = self.master_config.MPI.get(mpikey, None)
         if mpi_import_statement is not None:
             try:
                 self.log.info("Initializing MPI:")
                 self.log.info(mpi_import_statement)
                 exec mpi_import_statement in globals()
             except:
                 mpi = None
         else:
             mpi = None
 
 
     def start_app(self):
         self.engine.start()
         try:
             self.engine.loop.start()
         except KeyboardInterrupt:
             self.log.critical("Engine Interrupted, shutting down...\n")
 
 
 def launch_new_instance():
     """Create and run the IPython controller"""
     app = IPEngineApp()
     app.start()
 
 
 if __name__ == '__main__':
     launch_new_instance()
 

IPython/zmq/parallel/iploggerapp.py

@@ -1,132 +1,132 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.zmq.parallel.clusterdir import (
     ApplicationWithClusterDir,
     ClusterDirConfigLoader
 )
 from IPython.zmq.parallel.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.\n\n
 
 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>".
+usually located in your ipython directory and named as "cluster_<profile>".
 See the --profile and --cluster-dir options for details.
 """
 
 #-----------------------------------------------------------------------------
 # Command line options
 #-----------------------------------------------------------------------------
 
 
 class IPLoggerAppConfigLoader(ClusterDirConfigLoader):
 
     def _add_arguments(self):
         super(IPLoggerAppConfigLoader, self)._add_arguments()
         paa = self.parser.add_argument
         # Controller config
         paa('--url',
             type=str, dest='LogWatcher.url',
             help='The url the LogWatcher will listen on',
             )
         # MPI
         paa('--topics',
             type=str, dest='LogWatcher.topics', nargs='+',
             help='What topics to subscribe to',
             metavar='topics')
         # Global config
         paa('--log-to-file',
             action='store_true', dest='Global.log_to_file',
             help='Log to a file in the log directory (default is stdout)')
         
 
 #-----------------------------------------------------------------------------
 # Main application
 #-----------------------------------------------------------------------------
 
 
 class IPLoggerApp(ApplicationWithClusterDir):
 
     name = u'iploggerz'
     description = _description
     command_line_loader = IPLoggerAppConfigLoader
     default_config_file_name = default_config_file_name
     auto_create_cluster_dir = True
 
     def create_default_config(self):
         super(IPLoggerApp, self).create_default_config()
 
         # The engine should not clean logs as we don't want to remove the
         # active log files of other running engines.
         self.default_config.Global.clean_logs = False
 
         # If given, this is the actual location of the logger's URL file.
         # If not, this is computed using the profile, app_dir and furl_file_name
         self.default_config.Global.url_file_name = u'iplogger.url'
         self.default_config.Global.url_file = u''
 
     def post_load_command_line_config(self):
         pass
 
     def pre_construct(self):
         super(IPLoggerApp, self).pre_construct()
 
     def construct(self):
         # This is the working dir by now.
         sys.path.insert(0, '')
 
         self.start_logging()
 
         try:
             self.watcher = LogWatcher(config=self.master_config, logname=self.log.name)
         except:
             self.log.error("Couldn't start the LogWatcher", exc_info=True)
             self.exit(1)
         
 
     def start_app(self):
         try:
             self.watcher.start()
             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()
     app.start()
 
 
 if __name__ == '__main__':
     launch_new_instance()