add ipcluster engines; fix ssh process shutdown

IPython/zmq/parallel/ipclusterapp.py

@@ -1,505 +1,580 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
 
 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'
 
 
 _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 = 1
         self.default_config.Global.reset_config = False
         self.default_config.Global.clean_logs = True
-        self.default_config.Global.signal = 2
+        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):
+    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.
+
+        
         el_class = import_item(config.Global.engine_launcher)
         self.engine_launcher = el_class(
             work_dir=self.cluster_dir, config=config, logname=self.log.name
         )
-        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 signals
         signal.signal(signal.SIGINT, self.sigint_handler)
 
-        # 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)
-        # d1.addCallback(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.
-        
         # Start the controller and engines
         self._stopping = False  # Make sure stop_launchers is not called 2x.
-        d = self.start_controller()
-        dc = ioloop.DelayedCallback(self.start_engines, 1000*config.Global.delay, self.loop)
+        if controller:
+            self.start_controller()
+        dc = ioloop.DelayedCallback(self.start_engines, 1000*config.Global.delay*controller, self.loop)
         dc.start()
         self.startup_message()
-        # d.addCallback(self.start_engines)
-        # d.addCallback(self.startup_message)
-        # If the controller or engines fail to start, stop everything
-        # d.addErrback(self.stop_launchers)
-        return d
 
     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.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:
             self.log.info("stopping...")
         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:
+            self.log.fatal("stopping...")
+        # 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/launcher.py

@@ -1,835 +1,844 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 Facilities for launching IPython processes asynchronously.
 """
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2008-2009  The IPython Development Team
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 
 import os
 import re
 import sys
 import logging
 
 from signal import SIGINT, SIGTERM
 try:
     from signal import SIGKILL
 except ImportError:
     SIGKILL=SIGTERM
 
 from subprocess import Popen, PIPE, STDOUT
 try:
-    from subprocess import check_open
+    from subprocess import check_output
 except ImportError:
     # pre-2.7:
     from StringIO import StringIO
     
-    def check_open(*args, **kwargs):
+    def check_output(*args, **kwargs):
         sio = StringIO()
         kwargs.update(dict(stdout=PIPE, stderr=STDOUT))
         p = Popen(*args, **kwargs)
         out,err = p.communicate()
         return out
 
 from zmq.eventloop import ioloop
 
 from IPython.external import Itpl
 # from IPython.config.configurable import Configurable
 from IPython.utils.traitlets import Str, Int, List, Unicode, Instance
 from IPython.utils.path import get_ipython_module_path
 from IPython.utils.process import find_cmd, pycmd2argv, FindCmdError
 
 from factory import LoggingFactory
 
 # load winhpcjob from IPython.kernel
 try:
     from IPython.kernel.winhpcjob import (
         IPControllerTask, IPEngineTask,
         IPControllerJob, IPEngineSetJob
     )
 except ImportError:
     pass
 
 
 #-----------------------------------------------------------------------------
 # Paths to the kernel apps
 #-----------------------------------------------------------------------------
 
 
 ipcluster_cmd_argv = pycmd2argv(get_ipython_module_path(
     'IPython.zmq.parallel.ipclusterapp'
 ))
 
 ipengine_cmd_argv = pycmd2argv(get_ipython_module_path(
     'IPython.zmq.parallel.ipengineapp'
 ))
 
 ipcontroller_cmd_argv = pycmd2argv(get_ipython_module_path(
     'IPython.zmq.parallel.ipcontrollerapp'
 ))
 
 #-----------------------------------------------------------------------------
 # Base launchers and errors
 #-----------------------------------------------------------------------------
 
 
 class LauncherError(Exception):
     pass
 
 
 class ProcessStateError(LauncherError):
     pass
 
 
 class UnknownStatus(LauncherError):
     pass
 
 
 class BaseLauncher(LoggingFactory):
     """An asbtraction for starting, stopping and signaling a process."""
 
     # In all of the launchers, the work_dir is where child processes will be
     # run. This will usually be the cluster_dir, but may not be. any work_dir
     # passed into the __init__ method will override the config value.
     # This should not be used to set the work_dir for the actual engine
     # and controller. Instead, use their own config files or the
     # controller_args, engine_args attributes of the launchers to add
     # the --work-dir option.
     work_dir = Unicode(u'.')
     loop = Instance('zmq.eventloop.ioloop.IOLoop')
     def _loop_default(self):
         return ioloop.IOLoop.instance()
 
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(BaseLauncher, self).__init__(work_dir=work_dir, config=config, **kwargs)
         self.state = 'before' # can be before, running, after
         self.stop_callbacks = []
         self.start_data = None
         self.stop_data = None
 
     @property
     def args(self):
         """A list of cmd and args that will be used to start the process.
 
         This is what is passed to :func:`spawnProcess` and the first element
         will be the process name.
         """
         return self.find_args()
 
     def find_args(self):
         """The ``.args`` property calls this to find the args list.
 
         Subcommand should implement this to construct the cmd and args.
         """
         raise NotImplementedError('find_args must be implemented in a subclass')
 
     @property
     def arg_str(self):
         """The string form of the program arguments."""
         return ' '.join(self.args)
 
     @property
     def running(self):
         """Am I running."""
         if self.state == 'running':
             return True
         else:
             return False
 
     def start(self):
         """Start the process.
 
         This must return a deferred that fires with information about the
         process starting (like a pid, job id, etc.).
         """
         raise NotImplementedError('start must be implemented in a subclass')
 
     def stop(self):
         """Stop the process and notify observers of stopping.
 
         This must return a deferred that fires with information about the
         processing stopping, like errors that occur while the process is
         attempting to be shut down. This deferred won't fire when the process
         actually stops. To observe the actual process stopping, see
         :func:`observe_stop`.
         """
         raise NotImplementedError('stop must be implemented in a subclass')
 
     def on_stop(self, f):
         """Get a deferred that will fire when the process stops.
 
         The deferred will fire with data that contains information about
         the exit status of the process.
         """
         if self.state=='after':
             return f(self.stop_data)
         else:
             self.stop_callbacks.append(f)
 
     def notify_start(self, data):
         """Call this to trigger startup actions.
 
         This logs the process startup and sets the state to 'running'.  It is
         a pass-through so it can be used as a callback.
         """
 
         self.log.info('Process %r started: %r' % (self.args[0], data))
         self.start_data = data
         self.state = 'running'
         return data
 
     def notify_stop(self, data):
         """Call this to trigger process stop actions.
 
         This logs the process stopping and sets the state to 'after'. Call
         this to trigger all the deferreds from :func:`observe_stop`."""
 
         self.log.info('Process %r stopped: %r' % (self.args[0], data))
         self.stop_data = data
         self.state = 'after'
         for i in range(len(self.stop_callbacks)):
             d = self.stop_callbacks.pop()
             d(data)
         return data
 
     def signal(self, sig):
         """Signal the process.
 
         Return a semi-meaningless deferred after signaling the process.
 
         Parameters
         ----------
         sig : str or int
             'KILL', 'INT', etc., or any signal number
         """
         raise NotImplementedError('signal must be implemented in a subclass')
 
 
 #-----------------------------------------------------------------------------
 # Local process launchers
 #-----------------------------------------------------------------------------
 
 
 class LocalProcessLauncher(BaseLauncher):
     """Start and stop an external process in an asynchronous manner.
 
     This will launch the external process with a working directory of
     ``self.work_dir``.
     """
 
     # This is used to to construct self.args, which is passed to 
     # spawnProcess.
     cmd_and_args = List([])
     poll_frequency = Int(100) # in ms
 
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(LocalProcessLauncher, self).__init__(
             work_dir=work_dir, config=config, **kwargs
         )
         self.process = None
         self.start_deferred = None
         self.poller = None
 
     def find_args(self):
         return self.cmd_and_args
 
     def start(self):
         if self.state == 'before':
             self.process = Popen(self.args,
                 stdout=PIPE,stderr=PIPE,stdin=PIPE,
                 env=os.environ,
                 cwd=self.work_dir
             )
             
             self.loop.add_handler(self.process.stdout.fileno(), self.handle_stdout, self.loop.READ)
             self.loop.add_handler(self.process.stderr.fileno(), self.handle_stderr, self.loop.READ)
             self.poller = ioloop.PeriodicCallback(self.poll, self.poll_frequency, self.loop)
             self.poller.start()
             self.notify_start(self.process.pid)
         else:
             s = 'The process was already started and has state: %r' % self.state
             raise ProcessStateError(s)
 
     def stop(self):
         return self.interrupt_then_kill()
 
     def signal(self, sig):
         if self.state == 'running':
             self.process.send_signal(sig)
 
     def interrupt_then_kill(self, delay=2.0):
         """Send INT, wait a delay and then send KILL."""
         self.signal(SIGINT)
         self.killer  = ioloop.DelayedCallback(lambda : self.signal(SIGKILL), delay*1000, self.loop)
         self.killer.start()
 
     # callbacks, etc:
     
     def handle_stdout(self, fd, events):
         line = self.process.stdout.readline()
         # a stopped process will be readable but return empty strings
         if line:
             self.log.info(line[:-1])
         else:
             self.poll()
     
     def handle_stderr(self, fd, events):
         line = self.process.stderr.readline()
         # a stopped process will be readable but return empty strings
         if line:
             self.log.error(line[:-1])
         else:
             self.poll()
     
     def poll(self):
         status = self.process.poll()
         if status is not None:
             self.poller.stop()
             self.loop.remove_handler(self.process.stdout.fileno())
             self.loop.remove_handler(self.process.stderr.fileno())
             self.notify_stop(dict(exit_code=status, pid=self.process.pid))
         return status
 
 class LocalControllerLauncher(LocalProcessLauncher):
     """Launch a controller as a regular external process."""
 
     controller_cmd = List(ipcontroller_cmd_argv, config=True)
     # Command line arguments to ipcontroller.
     controller_args = List(['--log-to-file','--log-level', str(logging.INFO)], config=True)
 
     def find_args(self):
         return self.controller_cmd + self.controller_args
 
     def start(self, cluster_dir):
         """Start the controller by cluster_dir."""
         self.controller_args.extend(['--cluster-dir', cluster_dir])
         self.cluster_dir = unicode(cluster_dir)
         self.log.info("Starting LocalControllerLauncher: %r" % self.args)
         return super(LocalControllerLauncher, self).start()
 
 
 class LocalEngineLauncher(LocalProcessLauncher):
     """Launch a single engine as a regular externall process."""
 
     engine_cmd = List(ipengine_cmd_argv, config=True)
     # Command line arguments for ipengine.
     engine_args = List(
         ['--log-to-file','--log-level', str(logging.INFO)], config=True
     )
 
     def find_args(self):
         return self.engine_cmd + self.engine_args
 
     def start(self, cluster_dir):
         """Start the engine by cluster_dir."""
         self.engine_args.extend(['--cluster-dir', cluster_dir])
         self.cluster_dir = unicode(cluster_dir)
         return super(LocalEngineLauncher, self).start()
 
 
 class LocalEngineSetLauncher(BaseLauncher):
     """Launch a set of engines as regular external processes."""
 
     # Command line arguments for ipengine.
     engine_args = List(
         ['--log-to-file','--log-level', str(logging.INFO)], config=True
     )
     # launcher class
     launcher_class = LocalEngineLauncher
     
     def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(LocalEngineSetLauncher, self).__init__(
             work_dir=work_dir, config=config, **kwargs
         )
         self.launchers = {}
         self.stop_data = {}
 
     def start(self, n, cluster_dir):
         """Start n engines by profile or cluster_dir."""
         self.cluster_dir = unicode(cluster_dir)
         dlist = []
         for i in range(n):
             el = self.launcher_class(work_dir=self.work_dir, config=self.config, logname=self.log.name)
             # Copy the engine args over to each engine launcher.
             import copy
             el.engine_args = copy.deepcopy(self.engine_args)
             el.on_stop(self._notice_engine_stopped)
             d = el.start(cluster_dir)
             if i==0:
                 self.log.info("Starting LocalEngineSetLauncher: %r" % el.args)
             self.launchers[i] = el
             dlist.append(d)
         self.notify_start(dlist)
         # The consumeErrors here could be dangerous
         # dfinal = gatherBoth(dlist, consumeErrors=True)
         # dfinal.addCallback(self.notify_start)
         return dlist
 
     def find_args(self):
         return ['engine set']
 
     def signal(self, sig):
         dlist = []
         for el in self.launchers.itervalues():
             d = el.signal(sig)
             dlist.append(d)
         # dfinal = gatherBoth(dlist, consumeErrors=True)
         return dlist
 
     def interrupt_then_kill(self, delay=1.0):
         dlist = []
         for el in self.launchers.itervalues():
             d = el.interrupt_then_kill(delay)
             dlist.append(d)
         # dfinal = gatherBoth(dlist, consumeErrors=True)
         return dlist
 
     def stop(self):
         return self.interrupt_then_kill()
     
     def _notice_engine_stopped(self, data):
         print "notice", data
         pid = data['pid']
         for idx,el in self.launchers.iteritems():
             if el.process.pid == pid:
                 break
         self.launchers.pop(idx)
         self.stop_data[idx] = data
         if not self.launchers:
             self.notify_stop(self.stop_data)
 
 
 #-----------------------------------------------------------------------------
 # MPIExec launchers
 #-----------------------------------------------------------------------------
 
 
 class MPIExecLauncher(LocalProcessLauncher):
     """Launch an external process using mpiexec."""
 
     # The mpiexec command to use in starting the process.
     mpi_cmd = List(['mpiexec'], config=True)
     # The command line arguments to pass to mpiexec.
     mpi_args = List([], config=True)
     # The program to start using mpiexec.
     program = List(['date'], config=True)
     # The command line argument to the program.
     program_args = List([], config=True)
     # The number of instances of the program to start.
     n = Int(1, config=True)
 
     def find_args(self):
         """Build self.args using all the fields."""
         return self.mpi_cmd + ['-n', self.n] + self.mpi_args + \
                self.program + self.program_args
 
     def start(self, n):
         """Start n instances of the program using mpiexec."""
         self.n = n
         return super(MPIExecLauncher, self).start()
 
 
 class MPIExecControllerLauncher(MPIExecLauncher):
     """Launch a controller using mpiexec."""
 
     controller_cmd = List(ipcontroller_cmd_argv, config=True)
     # Command line arguments to ipcontroller.
     controller_args = List(['--log-to-file','--log-level', str(logging.INFO)], config=True)
     n = Int(1, config=False)
 
     def start(self, cluster_dir):
         """Start the controller by cluster_dir."""
         self.controller_args.extend(['--cluster-dir', cluster_dir])
         self.cluster_dir = unicode(cluster_dir)
         self.log.info("Starting MPIExecControllerLauncher: %r" % self.args)
         return super(MPIExecControllerLauncher, self).start(1)
 
     def find_args(self):
         return self.mpi_cmd + ['-n', self.n] + self.mpi_args + \
                self.controller_cmd + self.controller_args
 
 
 class MPIExecEngineSetLauncher(MPIExecLauncher):
 
     engine_cmd = List(ipengine_cmd_argv, config=True)
     # Command line arguments for ipengine.
     engine_args = List(
         ['--log-to-file','--log-level', str(logging.INFO)], config=True
     )
     n = Int(1, config=True)
 
     def start(self, n, cluster_dir):
         """Start n engines by profile or cluster_dir."""
         self.engine_args.extend(['--cluster-dir', cluster_dir])
         self.cluster_dir = unicode(cluster_dir)
         self.n = n
         self.log.info('Starting MPIExecEngineSetLauncher: %r' % self.args)
         return super(MPIExecEngineSetLauncher, self).start(n)
 
     def find_args(self):
         return self.mpi_cmd + ['-n', self.n] + self.mpi_args + \
                self.engine_cmd + self.engine_args
 
 
 #-----------------------------------------------------------------------------
 # SSH launchers
 #-----------------------------------------------------------------------------
 
 # TODO: Get SSH Launcher working again.
 
 class SSHLauncher(LocalProcessLauncher):
     """A minimal launcher for ssh.
 
     To be useful this will probably have to be extended to use the ``sshx``
     idea for environment variables.  There could be other things this needs
     as well.
     """
 
     ssh_cmd = List(['ssh'], config=True)
-    ssh_args = List([], config=True)
+    ssh_args = List(['-tt'], config=True)
     program = List(['date'], config=True)
     program_args = List([], config=True)
     hostname = Str('', config=True)
     user = Str(os.environ.get('USER','username'), config=True)
     location = Str('')
 
     def _hostname_changed(self, name, old, new):
         self.location = '%s@%s' % (self.user, new)
 
     def _user_changed(self, name, old, new):
         self.location = '%s@%s' % (new, self.hostname)
 
     def find_args(self):
         return self.ssh_cmd + self.ssh_args + [self.location] + \
                self.program + self.program_args
 
     def start(self, cluster_dir, hostname=None, user=None):
+        print self.config
         if hostname is not None:
             self.hostname = hostname
         if user is not None:
             self.user = user
+        print (self.location, hostname, user)
         return super(SSHLauncher, self).start()
+    
+    def signal(self, sig):
+        if self.state == 'running':
+            # send escaped ssh connection-closer
+            self.process.stdin.write('~.')
+            self.process.stdin.flush()
+        
 
 
 class SSHControllerLauncher(SSHLauncher):
 
     program = List(ipcontroller_cmd_argv, config=True)
     # Command line arguments to ipcontroller.
     program_args = List(['--log-to-file','--log-level', str(logging.INFO)], config=True)
 
 
 class SSHEngineLauncher(SSHLauncher):
     program = List(ipengine_cmd_argv, config=True)
     # Command line arguments for ipengine.
     program_args = List(
         ['--log-to-file','--log-level', str(logging.INFO)], config=True
     )
     
 class SSHEngineSetLauncher(LocalEngineSetLauncher):
     launcher_class = SSHEngineLauncher
 
 
 #-----------------------------------------------------------------------------
 # Windows HPC Server 2008 scheduler launchers
 #-----------------------------------------------------------------------------
 
 
 # This is only used on Windows.
 def find_job_cmd():
     if os.name=='nt':
         try:
             return find_cmd('job')
         except FindCmdError:
             return 'job'
     else:
         return 'job'
 
 
 class WindowsHPCLauncher(BaseLauncher):
 
     # A regular expression used to get the job id from the output of the 
     # submit_command.
     job_id_regexp = Str(r'\d+', config=True)
     # The filename of the instantiated job script.
     job_file_name = Unicode(u'ipython_job.xml', config=True)
     # The full path to the instantiated job script. This gets made dynamically
     # by combining the work_dir with the job_file_name.
     job_file = Unicode(u'')
     # The hostname of the scheduler to submit the job to
     scheduler = Str('', config=True)
     job_cmd = Str(find_job_cmd(), config=True)
 
-    def __init__(self, work_dir=u'.', config=None):
+    def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(WindowsHPCLauncher, self).__init__(
-            work_dir=work_dir, config=config
+            work_dir=work_dir, config=config, **kwargs
         )
 
     @property
     def job_file(self):
         return os.path.join(self.work_dir, self.job_file_name)
 
     def write_job_file(self, n):
         raise NotImplementedError("Implement write_job_file in a subclass.")
 
     def find_args(self):
         return ['job.exe']
         
     def parse_job_id(self, output):
         """Take the output of the submit command and return the job id."""
         m = re.search(self.job_id_regexp, output)
         if m is not None:
             job_id = m.group()
         else:
             raise LauncherError("Job id couldn't be determined: %s" % output)
         self.job_id = job_id
         self.log.info('Job started with job id: %r' % job_id)
         return job_id
 
     def start(self, n):
         """Start n copies of the process using the Win HPC job scheduler."""
         self.write_job_file(n)
         args = [
             'submit',
             '/jobfile:%s' % self.job_file,
             '/scheduler:%s' % self.scheduler
         ]
         self.log.info("Starting Win HPC Job: %s" % (self.job_cmd + ' ' + ' '.join(args),))
         # Twisted will raise DeprecationWarnings if we try to pass unicode to this
         output = check_output([self.job_cmd]+args,
             env=os.environ,
             cwd=self.work_dir,
             stderr=STDOUT
         )
         job_id = self.parse_job_id(output)
         # self.notify_start(job_id)
         return job_id
 
     def stop(self):
         args = [
             'cancel',
             self.job_id,
             '/scheduler:%s' % self.scheduler
         ]
         self.log.info("Stopping Win HPC Job: %s" % (self.job_cmd + ' ' + ' '.join(args),))
         try:
             output = check_output([self.job_cmd]+args,
                 env=os.environ,
                 cwd=self.work_dir,
                 stderr=STDOUT
             )
         except:
             output = 'The job already appears to be stoppped: %r' % self.job_id
         self.notify_stop(output)  # Pass the output of the kill cmd
         return output
 
 
 class WindowsHPCControllerLauncher(WindowsHPCLauncher):
 
     job_file_name = Unicode(u'ipcontroller_job.xml', config=True)
     extra_args = List([], config=False)
 
     def write_job_file(self, n):
         job = IPControllerJob(config=self.config)
 
         t = IPControllerTask(config=self.config)
         # The tasks work directory is *not* the actual work directory of 
         # the controller. It is used as the base path for the stdout/stderr
         # files that the scheduler redirects to.
         t.work_directory = self.cluster_dir
         # Add the --cluster-dir and from self.start().
         t.controller_args.extend(self.extra_args)
         job.add_task(t)
 
         self.log.info("Writing job description file: %s" % self.job_file)
         job.write(self.job_file)
 
     @property
     def job_file(self):
         return os.path.join(self.cluster_dir, self.job_file_name)
 
     def start(self, cluster_dir):
         """Start the controller by cluster_dir."""
         self.extra_args = ['--cluster-dir', cluster_dir]
         self.cluster_dir = unicode(cluster_dir)
         return super(WindowsHPCControllerLauncher, self).start(1)
 
 
 class WindowsHPCEngineSetLauncher(WindowsHPCLauncher):
 
     job_file_name = Unicode(u'ipengineset_job.xml', config=True)
     extra_args = List([], config=False)
 
     def write_job_file(self, n):
         job = IPEngineSetJob(config=self.config)
 
         for i in range(n):
             t = IPEngineTask(config=self.config)
             # The tasks work directory is *not* the actual work directory of 
             # the engine. It is used as the base path for the stdout/stderr
             # files that the scheduler redirects to.
             t.work_directory = self.cluster_dir
             # Add the --cluster-dir and from self.start().
             t.engine_args.extend(self.extra_args)
             job.add_task(t)
 
         self.log.info("Writing job description file: %s" % self.job_file)
         job.write(self.job_file)
 
     @property
     def job_file(self):
         return os.path.join(self.cluster_dir, self.job_file_name)
 
     def start(self, n, cluster_dir):
         """Start the controller by cluster_dir."""
         self.extra_args = ['--cluster-dir', cluster_dir]
         self.cluster_dir = unicode(cluster_dir)
         return super(WindowsHPCEngineSetLauncher, self).start(n)
 
 
 #-----------------------------------------------------------------------------
 # Batch (PBS) system launchers
 #-----------------------------------------------------------------------------
 
 # TODO: Get PBS launcher working again.
 
 class BatchSystemLauncher(BaseLauncher):
     """Launch an external process using a batch system.
 
     This class is designed to work with UNIX batch systems like PBS, LSF,
     GridEngine, etc.  The overall model is that there are different commands
     like qsub, qdel, etc. that handle the starting and stopping of the process.
 
     This class also has the notion of a batch script. The ``batch_template``
     attribute can be set to a string that is a template for the batch script.
     This template is instantiated using Itpl. Thus the template can use
     ${n} fot the number of instances. Subclasses can add additional variables
     to the template dict.
     """
 
     # Subclasses must fill these in.  See PBSEngineSet
     # The name of the command line program used to submit jobs.
     submit_command = Str('', config=True)
     # The name of the command line program used to delete jobs.
     delete_command = Str('', config=True)
     # A regular expression used to get the job id from the output of the 
     # submit_command.
     job_id_regexp = Str('', config=True)
     # The string that is the batch script template itself.
     batch_template = Str('', config=True)
     # The filename of the instantiated batch script.
     batch_file_name = Unicode(u'batch_script', config=True)
     # The full path to the instantiated batch script.
     batch_file = Unicode(u'')
 
-    def __init__(self, work_dir=u'.', config=None):
+    def __init__(self, work_dir=u'.', config=None, **kwargs):
         super(BatchSystemLauncher, self).__init__(
-            work_dir=work_dir, config=config
+            work_dir=work_dir, config=config, **kwargs
         )
         self.batch_file = os.path.join(self.work_dir, self.batch_file_name)
         self.context = {}
 
     def parse_job_id(self, output):
         """Take the output of the submit command and return the job id."""
         m = re.match(self.job_id_regexp, output)
         if m is not None:
             job_id = m.group()
         else:
             raise LauncherError("Job id couldn't be determined: %s" % output)
         self.job_id = job_id
         self.log.info('Job started with job id: %r' % job_id)
         return job_id
 
     def write_batch_script(self, n):
         """Instantiate and write the batch script to the work_dir."""
         self.context['n'] = n
         script_as_string = Itpl.itplns(self.batch_template, self.context)
         self.log.info('Writing instantiated batch script: %s' % self.batch_file)
         f = open(self.batch_file, 'w')
         f.write(script_as_string)
         f.close()
 
     def start(self, n):
         """Start n copies of the process using a batch system."""
         self.write_batch_script(n)
         output = check_output([self.submit_command, self.batch_file], env=os.environ, stdout=STDOUT)
         job_id = self.parse_job_id(output)
         # self.notify_start(job_id)
         return job_id
 
     def stop(self):
-        output = Popen([self.delete_command, self.job_id], env=os.environ, stderr=STDOUT)
+        output = check_output([self.delete_command, self.job_id], env=os.environ, stderr=STDOUT)
         self.notify_stop(output)  # Pass the output of the kill cmd
         return output
 
 
 class PBSLauncher(BatchSystemLauncher):
     """A BatchSystemLauncher subclass for PBS."""
 
     submit_command = Str('qsub', config=True)
     delete_command = Str('qdel', config=True)
     job_id_regexp = Str(r'\d+', config=True)
     batch_template = Str('', config=True)
     batch_file_name = Unicode(u'pbs_batch_script', config=True)
     batch_file = Unicode(u'')
 
 
 class PBSControllerLauncher(PBSLauncher):
     """Launch a controller using PBS."""
 
     batch_file_name = Unicode(u'pbs_batch_script_controller', config=True)
 
     def start(self, cluster_dir):
         """Start the controller by profile or cluster_dir."""
         # Here we save profile and cluster_dir in the context so they
         # can be used in the batch script template as ${profile} and
         # ${cluster_dir}
         self.context['cluster_dir'] = cluster_dir
         self.cluster_dir = unicode(cluster_dir)
         self.log.info("Starting PBSControllerLauncher: %r" % self.args)
         return super(PBSControllerLauncher, self).start(1)
 
 
 class PBSEngineSetLauncher(PBSLauncher):
 
     batch_file_name = Unicode(u'pbs_batch_script_engines', config=True)
 
     def start(self, n, cluster_dir):
         """Start n engines by profile or cluster_dir."""
         self.program_args.extend(['--cluster-dir', cluster_dir])
         self.cluster_dir = unicode(cluster_dir)
         self.log.info('Starting PBSEngineSetLauncher: %r' % self.args)
         return super(PBSEngineSetLauncher, self).start(n)
 
 
 #-----------------------------------------------------------------------------
 # A launcher for ipcluster itself!
 #-----------------------------------------------------------------------------
 
 
 class IPClusterLauncher(LocalProcessLauncher):
     """Launch the ipcluster program in an external process."""
 
     ipcluster_cmd = List(ipcluster_cmd_argv, config=True)
     # Command line arguments to pass to ipcluster.
     ipcluster_args = List(
         ['--clean-logs', '--log-to-file', '--log-level', str(logging.INFO)], config=True)
     ipcluster_subcommand = Str('start')
     ipcluster_n = Int(2)
 
     def find_args(self):
         return self.ipcluster_cmd + [self.ipcluster_subcommand] + \
             ['-n', repr(self.ipcluster_n)] + self.ipcluster_args
 
     def start(self):
         self.log.info("Starting ipcluster: %r" % self.args)
         return super(IPClusterLauncher, self).start()