add check_pid, and handle stale PID info in ipcluster.

    
ipcluster now actually tries to check if a program with a PID is running, better handling stale (or fake) PID files.
    
Checks are via `kernel32.OpenProcess` on Windows and `ps x` on everything else.  If the operation fails, then the process is assumed to exist, effectively falling back on prior behavior.

closes gh-428

IPython/parallel/apps/clusterdir.py

@@ -1,537 +1,566 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 The IPython cluster directory
 """
 
 #-----------------------------------------------------------------------------
 #  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 os
 import logging
 import re
 import shutil
 import sys
 
+from subprocess import Popen, PIPE
+
 from IPython.config.loader import PyFileConfigLoader
 from IPython.config.configurable import Configurable
 from IPython.core.application import Application, BaseAppConfigLoader
 from IPython.core.crashhandler import CrashHandler
 from IPython.core import release
 from IPython.utils.path import (
     get_ipython_package_dir,
     expand_path
 )
 from IPython.utils.traitlets import Unicode
 
 #-----------------------------------------------------------------------------
 # Module errors
 #-----------------------------------------------------------------------------
 
 class ClusterDirError(Exception):
     pass
 
 
 class PIDFileError(Exception):
     pass
 
 
 #-----------------------------------------------------------------------------
 # Class for managing cluster directories
 #-----------------------------------------------------------------------------
 
 class ClusterDir(Configurable):
     """An object to manage the cluster directory and its resources.
 
     The cluster directory is used by :command:`ipengine`, 
     :command:`ipcontroller` and :command:`ipclsuter` to manage the
     configuration, logging and security of these applications.
 
     This object knows how to find, create and manage these directories. This
     should be used by any code that want's to handle cluster directories.
     """
 
     security_dir_name = Unicode('security')
     log_dir_name = Unicode('log')
     pid_dir_name = Unicode('pid')
     security_dir = Unicode(u'')
     log_dir = Unicode(u'')
     pid_dir = Unicode(u'')
     location = Unicode(u'')
 
     def __init__(self, location=u''):
         super(ClusterDir, self).__init__(location=location)
 
     def _location_changed(self, name, old, new):
         if not os.path.isdir(new):
             os.makedirs(new)
         self.security_dir = os.path.join(new, self.security_dir_name)
         self.log_dir = os.path.join(new, self.log_dir_name)
         self.pid_dir = os.path.join(new, self.pid_dir_name)
         self.check_dirs()
 
     def _log_dir_changed(self, name, old, new):
         self.check_log_dir()
 
     def check_log_dir(self):
         if not os.path.isdir(self.log_dir):
             os.mkdir(self.log_dir)
 
     def _security_dir_changed(self, name, old, new):
         self.check_security_dir()
 
     def check_security_dir(self):
         if not os.path.isdir(self.security_dir):
             os.mkdir(self.security_dir, 0700)
         os.chmod(self.security_dir, 0700)
 
     def _pid_dir_changed(self, name, old, new):
         self.check_pid_dir()
 
     def check_pid_dir(self):
         if not os.path.isdir(self.pid_dir):
             os.mkdir(self.pid_dir, 0700)
         os.chmod(self.pid_dir, 0700)
 
     def check_dirs(self):
         self.check_security_dir()
         self.check_log_dir()
         self.check_pid_dir()
 
     def load_config_file(self, filename):
         """Load a config file from the top level of the cluster dir.
 
         Parameters
         ----------
         filename : unicode or str
             The filename only of the config file that must be located in
             the top-level of the cluster directory.
         """
         loader = PyFileConfigLoader(filename, self.location)
         return loader.load_config()
 
     def copy_config_file(self, config_file, path=None, overwrite=False):
         """Copy a default config file into the active cluster directory.
 
         Default configuration files are kept in :mod:`IPython.config.default`.
         This function moves these from that location to the working cluster
         directory.
         """
         if path is None:
             import IPython.config.default
             path = IPython.config.default.__file__.split(os.path.sep)[:-1]
             path = os.path.sep.join(path)
         src = os.path.join(path, config_file)
         dst = os.path.join(self.location, config_file)
         if not os.path.isfile(dst) or overwrite:
             shutil.copy(src, dst)
 
     def copy_all_config_files(self, path=None, overwrite=False):
         """Copy all config files into the active cluster directory."""
         for f in [u'ipcontroller_config.py', u'ipengine_config.py',
                   u'ipcluster_config.py']:
             self.copy_config_file(f, path=path, overwrite=overwrite)
 
     @classmethod
     def create_cluster_dir(csl, cluster_dir):
         """Create a new cluster directory given a full path.
 
         Parameters
         ----------
         cluster_dir : str
             The full path to the cluster directory.  If it does exist, it will
             be used.  If not, it will be created.
         """
         return ClusterDir(location=cluster_dir)
 
     @classmethod
     def create_cluster_dir_by_profile(cls, path, profile=u'default'):
         """Create a cluster dir by profile name and path.
 
         Parameters
         ----------
         path : str
             The path (directory) to put the cluster directory in.
         profile : str
             The name of the profile.  The name of the cluster directory will
             be "cluster_<profile>".
         """
         if not os.path.isdir(path):
             raise ClusterDirError('Directory not found: %s' % path)
         cluster_dir = os.path.join(path, u'cluster_' + profile)
         return ClusterDir(location=cluster_dir)
 
     @classmethod
     def find_cluster_dir_by_profile(cls, ipython_dir, profile=u'default'):
         """Find an existing cluster dir by profile name, return its ClusterDir.
 
         This searches through a sequence of paths for a cluster dir.  If it
         is not found, a :class:`ClusterDirError` exception will be raised.
 
         The search path algorithm is:
         1. ``os.getcwd()``
         2. ``ipython_dir``
         3. The directories found in the ":" separated 
            :env:`IPCLUSTER_DIR_PATH` environment variable.
 
         Parameters
         ----------
         ipython_dir : unicode or str
             The IPython directory to use.
         profile : unicode or str
             The name of the profile.  The name of the cluster directory
             will be "cluster_<profile>".
         """
         dirname = u'cluster_' + profile
         cluster_dir_paths = os.environ.get('IPCLUSTER_DIR_PATH','')
         if cluster_dir_paths:
             cluster_dir_paths = cluster_dir_paths.split(':')
         else:
             cluster_dir_paths = []
         paths = [os.getcwd(), ipython_dir] + cluster_dir_paths
         for p in paths:
             cluster_dir = os.path.join(p, dirname)
             if os.path.isdir(cluster_dir):
                 return ClusterDir(location=cluster_dir)
         else:
             raise ClusterDirError('Cluster directory not found in paths: %s' % dirname)
 
     @classmethod
     def find_cluster_dir(cls, cluster_dir):
         """Find/create a cluster dir and return its ClusterDir.
 
         This will create the cluster directory if it doesn't exist.
 
         Parameters
         ----------
         cluster_dir : unicode or str
             The path of the cluster directory.  This is expanded using
             :func:`IPython.utils.genutils.expand_path`.
         """
         cluster_dir = expand_path(cluster_dir)
         if not os.path.isdir(cluster_dir):
             raise ClusterDirError('Cluster directory not found: %s' % cluster_dir)
         return ClusterDir(location=cluster_dir)
 
 
 #-----------------------------------------------------------------------------
 # Command line options
 #-----------------------------------------------------------------------------
 
 class ClusterDirConfigLoader(BaseAppConfigLoader):
 
     def _add_cluster_profile(self, parser):
         paa = parser.add_argument
         paa('-p', '--profile',
             dest='Global.profile',type=unicode,
             help=
             """The string name of the profile to be used. This determines the name
             of the cluster dir as: cluster_<profile>. The default profile is named
             'default'.  The cluster directory is resolve this way if the
             --cluster-dir option is not used.""",
             metavar='Global.profile')
 
     def _add_cluster_dir(self, parser):
         paa = parser.add_argument
         paa('--cluster-dir',
             dest='Global.cluster_dir',type=unicode,
             help="""Set the cluster dir. This overrides the logic used by the
             --profile option.""",
             metavar='Global.cluster_dir')
 
     def _add_work_dir(self, parser):
         paa = parser.add_argument
         paa('--work-dir',
             dest='Global.work_dir',type=unicode,
             help='Set the working dir for the process.',
             metavar='Global.work_dir')
 
     def _add_clean_logs(self, parser):
         paa = parser.add_argument
         paa('--clean-logs',
             dest='Global.clean_logs', action='store_true',
             help='Delete old log flies before starting.')
 
     def _add_no_clean_logs(self, parser):
         paa = parser.add_argument
         paa('--no-clean-logs',
             dest='Global.clean_logs', action='store_false',
             help="Don't Delete old log flies before starting.")
 
     def _add_arguments(self):
         super(ClusterDirConfigLoader, self)._add_arguments()
         self._add_cluster_profile(self.parser)
         self._add_cluster_dir(self.parser)
         self._add_work_dir(self.parser)
         self._add_clean_logs(self.parser)
         self._add_no_clean_logs(self.parser)
 
 
 #-----------------------------------------------------------------------------
 # Crash handler for this application
 #-----------------------------------------------------------------------------
 
 
 _message_template = """\
 Oops, $self.app_name crashed. We do our best to make it stable, but...
 
 A crash report was automatically generated with the following information:
   - A verbatim copy of the crash traceback.
   - Data on your current $self.app_name configuration.
 
 It was left in the file named:
 \t'$self.crash_report_fname'
 If you can email this file to the developers, the information in it will help
 them in understanding and correcting the problem.
 
 You can mail it to: $self.contact_name at $self.contact_email
 with the subject '$self.app_name Crash Report'.
 
 If you want to do it now, the following command will work (under Unix):
 mail -s '$self.app_name Crash Report' $self.contact_email < $self.crash_report_fname
 
 To ensure accurate tracking of this issue, please file a report about it at:
 $self.bug_tracker
 """
 
 class ClusterDirCrashHandler(CrashHandler):
     """sys.excepthook for IPython itself, leaves a detailed report on disk."""
 
     message_template = _message_template
 
     def __init__(self, app):
         contact_name = release.authors['Brian'][0]
         contact_email = release.authors['Brian'][1]
         bug_tracker = 'http://github.com/ipython/ipython/issues'
         super(ClusterDirCrashHandler,self).__init__(
             app, contact_name, contact_email, bug_tracker
         )
 
 
 #-----------------------------------------------------------------------------
 # Main application
 #-----------------------------------------------------------------------------
 
 class ApplicationWithClusterDir(Application):
     """An application that puts everything into a cluster directory.
 
     Instead of looking for things in the ipython_dir, this type of application
     will use its own private directory called the "cluster directory"
     for things like config files, log files, etc.
 
     The cluster directory is resolved as follows:
 
     * If the ``--cluster-dir`` option is given, it is used.
     * If ``--cluster-dir`` is not given, the application directory is 
       resolve using the profile name as ``cluster_<profile>``. The search 
       path for this directory is then i) cwd if it is found there
       and ii) in ipython_dir otherwise.
 
     The config file for the application is to be put in the cluster
     dir and named the value of the ``config_file_name`` class attribute.
     """
 
     command_line_loader = ClusterDirConfigLoader
     crash_handler_class = ClusterDirCrashHandler
     auto_create_cluster_dir = True
     # temporarily override default_log_level to INFO
     default_log_level = logging.INFO
 
     def create_default_config(self):
         super(ApplicationWithClusterDir, self).create_default_config()
         self.default_config.Global.profile = u'default'
         self.default_config.Global.cluster_dir = u''
         self.default_config.Global.work_dir = os.getcwd()
         self.default_config.Global.log_to_file = False
         self.default_config.Global.log_url = None
         self.default_config.Global.clean_logs = False
 
     def find_resources(self):
         """This resolves the cluster directory.
 
         This tries to find the cluster directory and if successful, it will
         have done:
         * Sets ``self.cluster_dir_obj`` to the :class:`ClusterDir` object for 
           the application.
         * Sets ``self.cluster_dir`` attribute of the application and config
           objects.
 
         The algorithm used for this is as follows:
         1. Try ``Global.cluster_dir``.
         2. Try using ``Global.profile``.
         3. If both of these fail and ``self.auto_create_cluster_dir`` is
            ``True``, then create the new cluster dir in the IPython directory.
         4. If all fails, then raise :class:`ClusterDirError`.
         """
 
         try:
             cluster_dir = self.command_line_config.Global.cluster_dir
         except AttributeError:
             cluster_dir = self.default_config.Global.cluster_dir
         cluster_dir = expand_path(cluster_dir)
         try:
             self.cluster_dir_obj = ClusterDir.find_cluster_dir(cluster_dir)
         except ClusterDirError:
             pass
         else:
             self.log.info('Using existing cluster dir: %s' % \
                 self.cluster_dir_obj.location
             )
             self.finish_cluster_dir()
             return
 
         try:
             self.profile = self.command_line_config.Global.profile
         except AttributeError:
             self.profile = self.default_config.Global.profile
         try:
             self.cluster_dir_obj = ClusterDir.find_cluster_dir_by_profile(
                 self.ipython_dir, self.profile)
         except ClusterDirError:
             pass
         else:
             self.log.info('Using existing cluster dir: %s' % \
                 self.cluster_dir_obj.location
             )
             self.finish_cluster_dir()
             return
 
         if self.auto_create_cluster_dir:
             self.cluster_dir_obj = ClusterDir.create_cluster_dir_by_profile(
                 self.ipython_dir, self.profile
             )
             self.log.info('Creating new cluster dir: %s' % \
                 self.cluster_dir_obj.location
             )
             self.finish_cluster_dir()
         else:
             raise ClusterDirError('Could not find a valid cluster directory.')
 
     def finish_cluster_dir(self):
         # Set the cluster directory
         self.cluster_dir = self.cluster_dir_obj.location
         
         # These have to be set because they could be different from the one
         # that we just computed.  Because command line has the highest
         # priority, this will always end up in the master_config.
         self.default_config.Global.cluster_dir = self.cluster_dir
         self.command_line_config.Global.cluster_dir = self.cluster_dir
 
     def find_config_file_name(self):
         """Find the config file name for this application."""
         # For this type of Application it should be set as a class attribute.
         if not hasattr(self, 'default_config_file_name'):
             self.log.critical("No config filename found")
         else:
             self.config_file_name = self.default_config_file_name
 
     def find_config_file_paths(self):
         # Set the search path to to the cluster directory. We should NOT
         # include IPython.config.default here as the default config files
         # are ALWAYS automatically moved to the cluster directory.
         conf_dir = os.path.join(get_ipython_package_dir(), 'config', 'default')
         self.config_file_paths = (self.cluster_dir,)
 
     def pre_construct(self):
         # The log and security dirs were set earlier, but here we put them
         # into the config and log them.
         config = self.master_config
         sdir = self.cluster_dir_obj.security_dir
         self.security_dir = config.Global.security_dir = sdir
         ldir = self.cluster_dir_obj.log_dir
         self.log_dir = config.Global.log_dir = ldir
         pdir = self.cluster_dir_obj.pid_dir
         self.pid_dir = config.Global.pid_dir = pdir
         self.log.info("Cluster directory set to: %s" % self.cluster_dir)
         config.Global.work_dir = unicode(expand_path(config.Global.work_dir))
         # Change to the working directory. We do this just before construct
         # is called so all the components there have the right working dir.
         self.to_work_dir()
 
     def to_work_dir(self):
         wd = self.master_config.Global.work_dir
         if unicode(wd) != unicode(os.getcwd()):
             os.chdir(wd)
             self.log.info("Changing to working dir: %s" % wd)
 
     def start_logging(self):
         # Remove old log files
         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'%s-\d+\.(log|err|out)'%self.name,f):
                 # if f.startswith(self.name + u'-') and f.endswith('.log'):
                     os.remove(os.path.join(log_dir, f))
         # Start logging to the new log file
         if self.master_config.Global.log_to_file:
             log_filename = self.name + u'-' + str(os.getpid()) + u'.log'
             logfile = os.path.join(self.log_dir, log_filename)
             open_log_file = open(logfile, 'w')
         elif self.master_config.Global.log_url:
             open_log_file = None
         else:
             open_log_file = sys.stdout
         if open_log_file is not None:
             self.log.removeHandler(self._log_handler)
             self._log_handler = logging.StreamHandler(open_log_file)
             self._log_formatter = logging.Formatter("[%(name)s] %(message)s")
             self._log_handler.setFormatter(self._log_formatter)
             self.log.addHandler(self._log_handler)
         # log.startLogging(open_log_file)
 
     def write_pid_file(self, overwrite=False):
         """Create a .pid file in the pid_dir with my pid.
 
         This must be called after pre_construct, which sets `self.pid_dir`.
         This raises :exc:`PIDFileError` if the pid file exists already.
         """
         pid_file = os.path.join(self.pid_dir, self.name + u'.pid')
         if os.path.isfile(pid_file):
             pid = self.get_pid_from_file()
             if not overwrite:
                 raise PIDFileError(
                     'The pid file [%s] already exists. \nThis could mean that this '
                     'server is already running with [pid=%s].' % (pid_file, pid)
                 )
         with open(pid_file, 'w') as f:
             self.log.info("Creating pid file: %s" % pid_file)
             f.write(repr(os.getpid())+'\n')
 
     def remove_pid_file(self):
         """Remove the pid file.
 
         This should be called at shutdown by registering a callback with
         :func:`reactor.addSystemEventTrigger`. This needs to return
         ``None``.
         """
         pid_file = os.path.join(self.pid_dir, self.name + u'.pid')
         if os.path.isfile(pid_file):
             try:
                 self.log.info("Removing pid file: %s" % pid_file)
                 os.remove(pid_file)
             except:
                 self.log.warn("Error removing the pid file: %s" % pid_file)
 
     def get_pid_from_file(self):
         """Get the pid from the pid file.
 
         If the  pid file doesn't exist a :exc:`PIDFileError` is raised.
         """
         pid_file = os.path.join(self.pid_dir, self.name + u'.pid')
         if os.path.isfile(pid_file):
             with open(pid_file, 'r') as f:
                 pid = int(f.read().strip())
                 return pid
         else:
             raise PIDFileError('pid file not found: %s' % pid_file)
-
+    
+    def check_pid(self, pid):
+        if os.name == 'nt':
+            try:
+                import ctypes
+                # returns 0 if no such process (of ours) exists
+                # positive int otherwise
+                p = ctypes.windll.kernel32.OpenProcess(1,0,pid)
+            except Exception:
+                self.log.warn(
+                    "Could not determine whether pid %i is running via `OpenProcess`. "
+                    " Making the likely assumption that it is."%pid
+                )
+                return True
+            return bool(p)
+        else:
+            try:
+                p = Popen(['ps','x'], stdout=PIPE, stderr=PIPE)
+                output,_ = p.communicate()
+            except OSError:
+                self.log.warn(
+                    "Could not determine whether pid %i is running via `ps x`. "
+                    " Making the likely assumption that it is."%pid
+                )
+                return True
+            pids = map(int, re.findall(r'^\W*\d+', output, re.MULTILINE))
+            return pid in pids
+    
  No newline at end of file

IPython/parallel/apps/ipclusterapp.py

@@ -1,593 +1,617 b''
 #!/usr/bin/env python
 # encoding: utf-8
 """
 The ipcluster application.
 """
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2008-2009  The IPython Development Team
 #
 #  Distributed under the terms of the BSD License.  The full license is in
 #  the file COPYING, distributed as part of this software.
 #-----------------------------------------------------------------------------
 
 #-----------------------------------------------------------------------------
 # Imports
 #-----------------------------------------------------------------------------
 
 import errno
 import logging
 import os
 import re
 import signal
 
+from subprocess import check_call, CalledProcessError, PIPE
 import zmq
 from zmq.eventloop import ioloop
 
 from IPython.external.argparse import ArgumentParser, SUPPRESS
 from IPython.utils.importstring import import_item
 
 from IPython.parallel.apps.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 'ipcluster start -n 4'. For more complex usage 
 you will typically do 'ipcluster create -p mycluster', then edit
 configuration files, followed by 'ipcluster 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 'ipcluster <cmd> [options]'. To get help
             on a particular subcommand do 'ipcluster <cmd> -h'."""
             # help="For more help, type 'ipcluster <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, 
             'ipcluster 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, 'ipcluster 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, 'ipcluster 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, 'ipcluster 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'ipcluster'
     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.parallel.apps.launcher.LocalControllerLauncher'
         self.default_config.Global.engine_launcher = \
             'IPython.parallel.apps.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 'ipcluster start'.  Do "
                     "'ipcluster create -h' or 'ipcluster 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 'ipcluster start'.  Do "
                     "'ipcluster create -h' or 'ipcluster 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 = 'ipcluster 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 "ipcluster stop" to stop the cluster.' % pid
-            )
-            # Here I exit with a unusual exit status that other processes
-            # can watch for to learn how I existed.
-            self.exit(ALREADY_STARTED)
+            if self.check_pid(pid):
+                self.log.critical(
+                    'Cluster is already running with [pid=%s]. '
+                    'use "ipcluster stop" to stop the cluster.' % pid
+                )
+                # Here I exit with a unusual exit status that other processes
+                # can watch for to learn how I existed.
+                self.exit(ALREADY_STARTED)
+            else:
+                self.remove_pid_file()
+                
 
         # Now log and daemonize
         self.log.info(
             'Starting ipcluster with [daemon=%r]' % 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()
+        finally:
+            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.'
+                'Could not read pid file, cluster is probably not running.'
             )
             # Here I exit with a unusual exit status that other processes
             # can watch for to learn how I existed.
+            self.remove_pid_file()
             self.exit(ALREADY_STOPPED)
-        else:
-            if os.name=='posix':
-                sig = config.Global.signal
-                self.log.info(
-                    "Stopping cluster [pid=%r] with [signal=%r]" % (pid, sig)
-                )
+        
+        if not self.check_pid(pid):
+            self.log.critical(
+                'Cluster [pid=%r] is not running.' % pid
+            )
+            self.remove_pid_file()
+            # Here I exit with a unusual exit status that other processes
+            # can watch for to learn how I existed.
+            self.exit(ALREADY_STOPPED)
+            
+        elif os.name=='posix':
+            sig = config.Global.signal
+            self.log.info(
+                "Stopping cluster [pid=%r] with [signal=%r]" % (pid, sig)
+            )
+            try:
                 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.
+            except OSError:
+                self.log.error("Stopping cluster failed, assuming already dead.",
+                    exc_info=True)
                 self.remove_pid_file()
+        elif os.name=='nt':
+            try:
+                # kill the whole tree
+                p = check_call(['taskkill', '-pid', str(pid), '-t', '-f'], stdout=PIPE,stderr=PIPE)
+            except (CalledProcessError, OSError):
+                self.log.error("Stopping cluster failed, assuming already dead.",
+                    exc_info=True)
+            self.remove_pid_file()
 
 
 def launch_new_instance():
     """Create and run the IPython cluster."""
     app = IPClusterApp()
     app.start()
 
 
 if __name__ == '__main__':
     launch_new_instance()