Merging from upstream

IPython/kernel/multiengineclient.py

@@ -1,896 +1,951 b''
 # encoding: utf-8
 # -*- test-case-name: IPython.kernel.test.test_multiengineclient -*-
 
 """General Classes for IMultiEngine clients."""
 
 __docformat__ = "restructuredtext en"
 
 #-------------------------------------------------------------------------------
 #  Copyright (C) 2008  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 sys
 import cPickle as pickle
 from types import FunctionType
 import linecache
 
 from twisted.internet import reactor
 from twisted.python import components, log
 from twisted.python.failure import Failure
 from zope.interface import Interface, implements, Attribute
 
 from IPython.ColorANSI import TermColors
 
 from IPython.kernel.twistedutil import blockingCallFromThread
 from IPython.kernel import error
 from IPython.kernel.parallelfunction import ParallelFunction
 from IPython.kernel.mapper import (
     MultiEngineMapper, 
     IMultiEngineMapperFactory,
     IMapper
 )
 from IPython.kernel import map as Map
 from IPython.kernel import multiengine as me
 from IPython.kernel.multiengine import (IFullMultiEngine,
     IFullSynchronousMultiEngine)
 
 
 #-------------------------------------------------------------------------------
 # Pending Result things
 #-------------------------------------------------------------------------------
 
 class IPendingResult(Interface):
     """A representation of a result that is pending.
     
     This class is similar to Twisted's `Deferred` object, but is designed to be
     used in a synchronous context.
     """
     
     result_id=Attribute("ID of the deferred on the other side")
     client=Attribute("A client that I came from")
     r=Attribute("An attribute that is a property that calls and returns get_result")
     
     def get_result(default=None, block=True):
         """
         Get a result that is pending.
                 
         :Parameters:
             default
                 The value to return if the result is not ready.
             block : boolean
                 Should I block for the result.
                 
         :Returns: The actual result or the default value.
         """
         
     def add_callback(f, *args, **kwargs):
         """
         Add a callback that is called with the result.
         
         If the original result is foo, adding a callback will cause
         f(foo, *args, **kwargs) to be returned instead.  If multiple
         callbacks are registered, they are chained together: the result of
         one is passed to the next and so on.  
         
         Unlike Twisted's Deferred object, there is no errback chain.  Thus
         any exception raised will not be caught and handled.  User must 
         catch these by hand when calling `get_result`.
         """
 
 
 class PendingResult(object):
     """A representation of a result that is not yet ready.
     
     A user should not create a `PendingResult` instance by hand.
     
     Methods
     =======
     
     * `get_result`
     * `add_callback`
     
     Properties
     ==========
     * `r`
     """
     
     def __init__(self, client, result_id):
         """Create a PendingResult with a result_id and a client instance.
         
         The client should implement `_getPendingResult(result_id, block)`.
         """
         self.client = client
         self.result_id = result_id
         self.called = False
         self.raised = False
         self.callbacks = []
         
     def get_result(self, default=None, block=True):
         """Get a result that is pending.
                 
         This method will connect to an IMultiEngine adapted controller
         and see if the result is ready.  If the action triggers an exception
         raise it and record it.  This method records the result/exception once it is 
         retrieved.  Calling `get_result` again will get this cached result or will
         re-raise the exception.  The .r attribute is a property that calls
         `get_result` with block=True.
         
         :Parameters:
             default
                 The value to return if the result is not ready.
             block : boolean
                 Should I block for the result.
                 
         :Returns: The actual result or the default value.
         """
         
         if self.called:
             if self.raised:
                 raise self.result[0], self.result[1], self.result[2]
             else:
                 return self.result
         try:
             result = self.client.get_pending_deferred(self.result_id, block)
         except error.ResultNotCompleted:
             return default
         except:
             # Reraise other error, but first record them so they can be reraised
             # later if .r or get_result is called again.
             self.result = sys.exc_info()
             self.called = True
             self.raised = True
             raise
         else:
             for cb in self.callbacks:
                 result = cb[0](result, *cb[1], **cb[2])
             self.result = result
             self.called = True
             return result
         
     def add_callback(self, f, *args, **kwargs):
         """Add a callback that is called with the result.
         
         If the original result is result, adding a callback will cause
         f(result, *args, **kwargs) to be returned instead.  If multiple
         callbacks are registered, they are chained together: the result of
         one is passed to the next and so on.  
         
         Unlike Twisted's Deferred object, there is no errback chain.  Thus
         any exception raised will not be caught and handled.  User must 
         catch these by hand when calling `get_result`.
         """
         assert callable(f)
         self.callbacks.append((f, args, kwargs))
         
     def __cmp__(self, other):
         if self.result_id < other.result_id:
             return -1
         else:
             return 1
             
     def _get_r(self):
         return self.get_result(block=True)
     
     r = property(_get_r)
     """This property is a shortcut to a `get_result(block=True)`."""
 
 
 #-------------------------------------------------------------------------------
 # Pretty printing wrappers for certain lists
 #-------------------------------------------------------------------------------    
     
 class ResultList(list):
     """A subclass of list that pretty prints the output of `execute`/`get_result`."""
     
     def __repr__(self):
         output = []
         # These colored prompts were not working on Windows
         if sys.platform == 'win32':
             blue = normal = red = green = ''
         else:
             blue = TermColors.Blue
             normal = TermColors.Normal
             red = TermColors.Red
             green = TermColors.Green
         output.append("<Results List>\n")
         for cmd in self:
             if isinstance(cmd, Failure):
                 output.append(cmd)
             else:
                 target = cmd.get('id',None)
                 cmd_num = cmd.get('number',None)
                 cmd_stdin = cmd.get('input',{}).get('translated','No Input')
                 cmd_stdout = cmd.get('stdout', None)
                 cmd_stderr = cmd.get('stderr', None)
                 output.append("%s[%i]%s In [%i]:%s %s\n" % \
                     (green, target,
                     blue, cmd_num, normal, cmd_stdin))
                 if cmd_stdout:
                     output.append("%s[%i]%s Out[%i]:%s %s\n" % \
                         (green, target,
                         red, cmd_num, normal, cmd_stdout))
                 if cmd_stderr:
                     output.append("%s[%i]%s Err[%i]:\n%s %s" % \
                         (green, target,
                         red, cmd_num, normal, cmd_stderr))
         return ''.join(output)
 
 
 def wrapResultList(result):
     """A function that wraps the output of `execute`/`get_result` -> `ResultList`."""
     if len(result) == 0:
         result = [result]
     return ResultList(result)
 
 
 class QueueStatusList(list):
     """A subclass of list that pretty prints the output of `queue_status`."""
     
     def __repr__(self):
         output = []
         output.append("<Queue Status List>\n")
         for e in self:
             output.append("Engine: %s\n" % repr(e[0]))
             output.append("    Pending: %s\n" % repr(e[1]['pending']))
             for q in e[1]['queue']:
                 output.append("    Command: %s\n" % repr(q))
         return ''.join(output)
 
 
 #-------------------------------------------------------------------------------
 # InteractiveMultiEngineClient
 #-------------------------------------------------------------------------------    
 
 class InteractiveMultiEngineClient(object):
     """A mixin class that add a few methods to a multiengine client.
     
     The methods in this mixin class are designed for interactive usage.
     """
                                         
     def activate(self):
         """Make this `MultiEngineClient` active for parallel magic commands.
         
         IPython has a magic command syntax to work with `MultiEngineClient` objects.
         In a given IPython session there is a single active one.  While
         there can be many `MultiEngineClient` created and used by the user, 
         there is only one active one.  The active `MultiEngineClient` is used whenever 
         the magic commands %px and %autopx are used.
         
         The activate() method is called on a given `MultiEngineClient` to make it 
         active.  Once this has been done, the magic commands can be used.
         """
         
         try:
             __IPYTHON__.activeController = self
         except NameError:
             print "The IPython Controller magics only work within IPython."
                     
     def __setitem__(self, key, value):
         """Add a dictionary interface for pushing/pulling.
         
         This functions as a shorthand for `push`.
         
         :Parameters:
             key : str 
                 What to call the remote object.
             value : object
                 The local Python object to push.
         """
         targets, block = self._findTargetsAndBlock()
         return self.push({key:value}, targets=targets, block=block)
     
     def __getitem__(self, key):
         """Add a dictionary interface for pushing/pulling.
         
         This functions as a shorthand to `pull`.
         
         :Parameters:
          - `key`: A string representing the key.
         """
         if isinstance(key, str):
             targets, block = self._findTargetsAndBlock()
             return self.pull(key, targets=targets, block=block)
         else:
             raise TypeError("__getitem__ only takes strs")
             
     def __len__(self):
         """Return the number of available engines."""
         return len(self.get_ids())
     
     #---------------------------------------------------------------------------
     # Make this a context manager for with
     #---------------------------------------------------------------------------
     
     def findsource_file(self,f):
         linecache.checkcache()
         s = findsource(f.f_code)
         lnum = f.f_lineno
         wsource = s[0][f.f_lineno:]
         return strip_whitespace(wsource)
 
     def findsource_ipython(self,f):
         from IPython import ipapi
         self.ip = ipapi.get()
         wsource = [l+'\n' for l in
                    self.ip.IP.input_hist_raw[-1].splitlines()[1:]] 
         return strip_whitespace(wsource)
         
     def __enter__(self):
         f = sys._getframe(1)
         local_ns = f.f_locals
         global_ns = f.f_globals
         if f.f_code.co_filename == '<ipython console>':
             s = self.findsource_ipython(f)
         else:
             s = self.findsource_file(f)
 
         self._with_context_result = self.execute(s)
 
     def __exit__ (self, etype, value, tb):
         if issubclass(etype,error.StopLocalExecution):
             return True
 
 
 def remote():
     m = 'Special exception to stop local execution of parallel code.'
     raise error.StopLocalExecution(m)
 
 def strip_whitespace(source):
     # Expand tabs to avoid any confusion.
     wsource = [l.expandtabs(4) for l in source]
     # Detect the indentation level
     done = False
     for line in wsource:
         if line.isspace():
             continue
         for col,char in enumerate(line):
             if char != ' ':
                 done = True
                 break
         if done:
             break
     # Now we know how much leading space there is in the code.  Next, we
     # extract up to the first line that has less indentation.
     # WARNINGS: we skip comments that may be misindented, but we do NOT yet
     # detect triple quoted strings that may have flush left text.
     for lno,line in enumerate(wsource):
         lead = line[:col]
         if lead.isspace():
             continue
         else:
             if not lead.lstrip().startswith('#'):
                 break
     # The real 'with' source is up to lno
     src_lines = [l[col:] for l in wsource[:lno+1]]
 
     # Finally, check that the source's first non-comment line begins with the
     # special call 'remote()'
     for nline,line in enumerate(src_lines):
         if line.isspace() or line.startswith('#'):
             continue
         if 'remote()' in line:
             break
         else:
             raise ValueError('remote() call missing at the start of code')
     src = ''.join(src_lines[nline+1:])
     #print 'SRC:\n<<<<<<<>>>>>>>\n%s<<<<<>>>>>>' % src  # dbg
     return src
 
 
 #-------------------------------------------------------------------------------
 # The top-level MultiEngine client adaptor
 #-------------------------------------------------------------------------------
 
 
 class IFullBlockingMultiEngineClient(Interface):
     pass
 
 
 class FullBlockingMultiEngineClient(InteractiveMultiEngineClient):
     """
     A blocking client to the `IMultiEngine` controller interface.
     
     This class allows users to use a set of engines for a parallel
     computation through the `IMultiEngine` interface.  In this interface,
     each engine has a specific id (an int) that is used to refer to the
     engine, run code on it, etc.
     """
     
     implements(
         IFullBlockingMultiEngineClient,
         IMultiEngineMapperFactory,
         IMapper
     )
     
     def __init__(self, smultiengine):
         self.smultiengine = smultiengine
         self.block = True
         self.targets = 'all'
     
     def _findBlock(self, block=None):
         if block is None:
             return self.block
         else:
             if block in (True, False):
                 return block
             else:
                 raise ValueError("block must be True or False")
     
     def _findTargets(self, targets=None):
         if targets is None:
             return self.targets
         else:
             if not isinstance(targets, (str,list,tuple,int)):
                 raise ValueError("targets must be a str, list, tuple or int")
             return targets
     
     def _findTargetsAndBlock(self, targets=None, block=None):
         return self._findTargets(targets), self._findBlock(block) 
     
     def _blockFromThread(self, function, *args, **kwargs):
         block = kwargs.get('block', None)
         if block is None:
             raise error.MissingBlockArgument("'block' keyword argument is missing")
         result = blockingCallFromThread(function, *args, **kwargs)
         if not block:
             result = PendingResult(self, result)
         return result
     
     def get_pending_deferred(self, deferredID, block):
         return blockingCallFromThread(self.smultiengine.get_pending_deferred, deferredID, block)
     
     def barrier(self, pendingResults):
         """Synchronize a set of `PendingResults`.
         
         This method is a synchronization primitive that waits for a set of
         `PendingResult` objects to complete.  More specifically, barier does
         the following.
         
         * The `PendingResult`s are sorted by result_id.
         * The `get_result` method is called for each `PendingResult` sequentially
           with block=True.
         * If a `PendingResult` gets a result that is an exception, it is 
           trapped and can be re-raised later by calling `get_result` again.
         * The `PendingResult`s are flushed from the controller.
                 
         After barrier has been called on a `PendingResult`, its results can 
         be retrieved by calling `get_result` again or accesing the `r` attribute
         of the instance.
         """
         
         # Convert to list for sorting and check class type 
         prList = list(pendingResults)
         for pr in prList:
             if not isinstance(pr, PendingResult):
                 raise error.NotAPendingResult("Objects passed to barrier must be PendingResult instances")
                             
         # Sort the PendingResults so they are in order
         prList.sort()
         # Block on each PendingResult object
         for pr in prList:
             try:
                 result = pr.get_result(block=True)
             except Exception:
                 pass
     
     def flush(self):
         """
         Clear all pending deferreds/results from the controller.
         
         For each `PendingResult` that is created by this client, the controller
         holds on to the result for that `PendingResult`.  This can be a problem
         if there are a large number of `PendingResult` objects that are created.
         
         Once the result of the `PendingResult` has been retrieved, the result
         is removed from the controller, but if a user doesn't get a result (
         they just ignore the `PendingResult`) the result is kept forever on the
         controller.  This method allows the user to clear out all un-retrieved
         results on the controller. 
         """
         r = blockingCallFromThread(self.smultiengine.clear_pending_deferreds)
         return r
     
     clear_pending_results = flush
     
     #---------------------------------------------------------------------------
     # IEngineMultiplexer related methods
     #---------------------------------------------------------------------------
     
     def execute(self, lines, targets=None, block=None):
         """
         Execute code on a set of engines.
         
         :Parameters:
             lines : str
                 The Python code to execute as a string
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         result = blockingCallFromThread(self.smultiengine.execute, lines,
             targets=targets, block=block)
         if block:
             result = ResultList(result)
         else:
             result = PendingResult(self, result)
             result.add_callback(wrapResultList)
         return result
     
     def push(self, namespace, targets=None, block=None):
         """
         Push a dictionary of keys and values to engines namespace.
         
         Each engine has a persistent namespace.  This method is used to push
         Python objects into that namespace.
         
         The objects in the namespace must be pickleable.
         
         :Parameters:
             namespace : dict
                 A dict that contains Python objects to be injected into
                 the engine persistent namespace.
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.push, namespace,
             targets=targets, block=block)
     
     def pull(self, keys, targets=None, block=None):
         """
         Pull Python objects by key out of engines namespaces.
         
         :Parameters:
             keys : str or list of str
                 The names of the variables to be pulled
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.pull, keys, targets=targets, block=block)
     
     def push_function(self, namespace, targets=None, block=None):
         """
         Push a Python function to an engine.
         
         This method is used to push a Python function to an engine.  This
         method can then be used in code on the engines.  Closures are not supported.
         
         :Parameters:
             namespace : dict
                 A dict whose values are the functions to be pushed.  The keys give
                 that names that the function will appear as in the engines
                 namespace.
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.push_function, namespace, targets=targets, block=block)
     
     def pull_function(self, keys, targets=None, block=None):
         """
         Pull a Python function from an engine.
         
         This method is used to pull a Python function from an engine.
         Closures are not supported.
         
         :Parameters:
             keys : str or list of str
                 The names of the functions to be pulled
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.pull_function, keys, targets=targets, block=block)
     
     def push_serialized(self, namespace, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.push_serialized, namespace, targets=targets, block=block)
     
     def pull_serialized(self, keys, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.pull_serialized, keys, targets=targets, block=block)
     
     def get_result(self, i=None, targets=None, block=None):
         """
         Get a previous result.
         
         When code is executed in an engine, a dict is created and returned.  This
         method retrieves that dict for previous commands.
         
         :Parameters:
             i : int
                 The number of the result to get
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         result = blockingCallFromThread(self.smultiengine.get_result, i, targets=targets, block=block)
         if block:
             result = ResultList(result)
         else:
             result = PendingResult(self, result)
             result.add_callback(wrapResultList)
         return result
     
     def reset(self, targets=None, block=None):
         """
         Reset an engine.
         
         This method clears out the namespace of an engine.
 
         :Parameters:
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.reset, targets=targets, block=block)
     
     def keys(self, targets=None, block=None):
         """
         Get a list of all the variables in an engine's namespace.
 
         :Parameters:
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.keys, targets=targets, block=block)
     
     def kill(self, controller=False, targets=None, block=None):
         """
         Kill the engines and controller.
         
         This method is used to stop the engine and controller by calling
         `reactor.stop`.
         
         :Parameters:
             controller : boolean
                 If True, kill the engines and controller.  If False, just the 
                 engines
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.kill, controller, targets=targets, block=block)
     
     def clear_queue(self, targets=None, block=None):
         """
         Clear out the controller's queue for an engine.
         
         The controller maintains a queue for each engine.  This clear it out.
         
         :Parameters:
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.clear_queue, targets=targets, block=block)
     
     def queue_status(self, targets=None, block=None):
         """
         Get the status of an engines queue.
         
         :Parameters:
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.queue_status, targets=targets, block=block)
     
     def set_properties(self, properties, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.set_properties, properties, targets=targets, block=block)
     
     def get_properties(self, keys=None, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.get_properties, keys, targets=targets, block=block)
     
     def has_properties(self, keys, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.has_properties, keys, targets=targets, block=block)
     
     def del_properties(self, keys, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.del_properties, keys, targets=targets, block=block)
     
     def clear_properties(self, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.clear_properties, targets=targets, block=block)
     
     #---------------------------------------------------------------------------
     # IMultiEngine related methods
     #---------------------------------------------------------------------------
     
     def get_ids(self):
         """
         Returns the ids of currently registered engines.
         """
         result = blockingCallFromThread(self.smultiengine.get_ids)
         return result
         
     #---------------------------------------------------------------------------
     # IMultiEngineCoordinator
     #---------------------------------------------------------------------------
              
     def scatter(self, key, seq, dist='b', flatten=False, targets=None, block=None):
         """
         Partition a Python sequence and send the partitions to a set of engines.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.scatter, key, seq, 
             dist, flatten, targets=targets, block=block)
     
     def gather(self, key, dist='b', targets=None, block=None):
         """
         Gather a partitioned sequence on a set of engines as a single local seq.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.gather, key, dist, 
             targets=targets, block=block)
     
     def raw_map(self, func, seq, dist='b', targets=None, block=None):
         """
         A parallelized version of Python's builtin map.
         
         This has a slightly different syntax than the builtin `map`.
         This is needed because we need to have keyword arguments and thus
         can't use *args to capture all the sequences.  Instead, they must
         be passed in a list or tuple.
         
         raw_map(func, seqs) -> map(func, seqs[0], seqs[1], ...)
         
         Most users will want to use parallel functions or the `mapper`
         and `map` methods for an API that follows that of the builtin
         `map`.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.raw_map, func, seq, 
             dist, targets=targets, block=block)
     
     def map(self, func, *sequences):
         """
         A parallel version of Python's builtin `map` function.
         
         This method applies a function to sequences of arguments.  It 
         follows the same syntax as the builtin `map`.
         
         This method creates a mapper objects by calling `self.mapper` with
         no arguments and then uses that mapper to do the mapping.  See
         the documentation of `mapper` for more details.
         """
         return self.mapper().map(func, *sequences)
     
     def mapper(self, dist='b', targets='all', block=None):
         """
         Create a mapper object that has a `map` method.
         
         This method returns an object that implements the `IMapper` 
         interface.  This method is a factory that is used to control how 
         the map happens.
         
         :Parameters:
             dist : str
                 What decomposition to use, 'b' is the only one supported
                 currently
             targets : str, int, sequence of ints
                 Which engines to use for the map
             block : boolean
                 Should calls to `map` block or not
         """
         return MultiEngineMapper(self, dist, targets, block)
     
     def parallel(self, dist='b', targets=None, block=None):
         """
         A decorator that turns a function into a parallel function.
         
         This can be used as:
         
         @parallel()
         def f(x, y)
             ...
         
         f(range(10), range(10))
         
         This causes f(0,0), f(1,1), ... to be called in parallel.
         
         :Parameters:
             dist : str
                 What decomposition to use, 'b' is the only one supported
                 currently
             targets : str, int, sequence of ints
                 Which engines to use for the map
             block : boolean
                 Should calls to `map` block or not
         """
         targets, block = self._findTargetsAndBlock(targets, block)       
         mapper = self.mapper(dist, targets, block)
         pf = ParallelFunction(mapper)
         return pf
     
     #---------------------------------------------------------------------------
     # IMultiEngineExtras
     #---------------------------------------------------------------------------
     
     def zip_pull(self, keys, targets=None, block=None):
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.zip_pull, keys, 
             targets=targets, block=block)
     
     def run(self, filename, targets=None, block=None):
         """
         Run a Python code in a file on the engines.
         
         :Parameters:
             filename : str
                 The name of the local file to run
             targets : id or list of ids
                 The engine to use for the execution
             block : boolean
                 If False, this method will return the actual result.  If False,
                 a `PendingResult` is returned which can be used to get the result
                 at a later time.
         """
         targets, block = self._findTargetsAndBlock(targets, block)
         return self._blockFromThread(self.smultiengine.run, filename,
             targets=targets, block=block)
+    
+    def benchmark(self, push_size=10000):
+        """
+        Run performance benchmarks for the current IPython cluster.
+        
+        This method tests both the latency of sending command and data to the
+        engines as well as the throughput of sending large objects to the
+        engines using push.  The latency is measured by having one or more
+        engines execute the command 'pass'.  The throughput is measure by 
+        sending an NumPy array of size `push_size` to one or more engines.
+        
+        These benchmarks will vary widely on different hardware and networks
+        and thus can be used to get an idea of the performance characteristics
+        of a particular configuration of an IPython controller and engines.
+        
+        This function is not testable within our current testing framework.
+        """
+        import timeit, __builtin__
+        __builtin__._mec_self = self
+        benchmarks = {}
+        repeat = 3
+        count = 10
+
+        timer = timeit.Timer('_mec_self.execute("pass",0)')
+        result = 1000*min(timer.repeat(repeat,count))/count
+        benchmarks['single_engine_latency'] = (result,'msec')
+
+        timer = timeit.Timer('_mec_self.execute("pass")')
+        result = 1000*min(timer.repeat(repeat,count))/count
+        benchmarks['all_engine_latency'] = (result,'msec')
 
+        try:
+            import numpy as np
+        except:
+            pass
+        else:
+            timer = timeit.Timer(
+                "_mec_self.push(d)",
+                "import numpy as np; d = dict(a=np.zeros(%r,dtype='float64'))" % push_size
+            )
+            result = min(timer.repeat(repeat,count))/count
+            benchmarks['all_engine_push'] = (1e-6*push_size*8/result, 'MB/sec')
+
+        try:
+            import numpy as np
+        except:
+            pass
+        else:
+            timer = timeit.Timer(
+                "_mec_self.push(d,0)",
+                "import numpy as np; d = dict(a=np.zeros(%r,dtype='float64'))" % push_size
+            )
+            result = min(timer.repeat(repeat,count))/count
+            benchmarks['single_engine_push'] = (1e-6*push_size*8/result, 'MB/sec')
+
+        return benchmarks
 
 
 components.registerAdapter(FullBlockingMultiEngineClient,
             IFullSynchronousMultiEngine, IFullBlockingMultiEngineClient)
 
 
 
 

IPython/kernel/scripts/ipcluster.py

@@ -1,723 +1,783 b''
 #!/usr/bin/env python
 # encoding: utf-8
 
 """Start an IPython cluster = (controller + engines)."""
 
 #-----------------------------------------------------------------------------
 #  Copyright (C) 2008  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 signal
 import tempfile
 pjoin = os.path.join
 
 from twisted.internet import reactor, defer
 from twisted.internet.protocol import ProcessProtocol
 from twisted.internet.error import ProcessDone, ProcessTerminated
 from twisted.internet.utils import getProcessOutput
 from twisted.python import failure, log
 
 from IPython.external import argparse
 from IPython.external import Itpl
 from IPython.genutils import get_ipython_dir, num_cpus
 from IPython.kernel.fcutil import have_crypto
 from IPython.kernel.error import SecurityError
 from IPython.kernel.fcutil import have_crypto
 from IPython.kernel.twistedutil import gatherBoth
 from IPython.kernel.util import printer
 
 
 #-----------------------------------------------------------------------------
 # General process handling code
 #-----------------------------------------------------------------------------
 
 def find_exe(cmd):
     try:
         import win32api
     except ImportError:
         raise ImportError('you need to have pywin32 installed for this to work')
     else:
         try:
             (path, offest) = win32api.SearchPath(os.environ['PATH'],cmd + '.exe')
         except:
             (path, offset) = win32api.SearchPath(os.environ['PATH'],cmd + '.bat')
     return path
 
 class ProcessStateError(Exception):
     pass
 
 class UnknownStatus(Exception):
     pass
 
 class LauncherProcessProtocol(ProcessProtocol):
     """
     A ProcessProtocol to go with the ProcessLauncher.
     """
     def __init__(self, process_launcher):
         self.process_launcher = process_launcher
     
     def connectionMade(self):
         self.process_launcher.fire_start_deferred(self.transport.pid)
     
     def processEnded(self, status):
         value = status.value
         if isinstance(value, ProcessDone):
             self.process_launcher.fire_stop_deferred(0)
         elif isinstance(value, ProcessTerminated):
             self.process_launcher.fire_stop_deferred(
                 {'exit_code':value.exitCode,
                  'signal':value.signal,
                  'status':value.status
                 }
             )
         else:
             raise UnknownStatus("unknown exit status, this is probably a bug in Twisted")
     
     def outReceived(self, data):
         log.msg(data)
     
     def errReceived(self, data):
         log.err(data)
 
 class ProcessLauncher(object):
     """
     Start and stop an external process in an asynchronous manner.
     
     Currently this uses deferreds to notify other parties of process state
     changes.  This is an awkward design and should be moved to using
     a formal NotificationCenter.
     """
     def __init__(self, cmd_and_args):
         self.cmd = cmd_and_args[0]
         self.args = cmd_and_args
         self._reset()
     
     def _reset(self):
         self.process_protocol = None
         self.pid = None
         self.start_deferred = None
         self.stop_deferreds = []
         self.state = 'before' # before, running, or after
     
     @property
     def running(self):
         if self.state == 'running':
             return True
         else:
             return False
     
     def fire_start_deferred(self, pid):
         self.pid = pid
         self.state = 'running'
         log.msg('Process %r has started with pid=%i' % (self.args, pid))
         self.start_deferred.callback(pid)
     
     def start(self):
         if self.state == 'before':
             self.process_protocol = LauncherProcessProtocol(self)
             self.start_deferred = defer.Deferred()
             self.process_transport = reactor.spawnProcess(
                 self.process_protocol,
                 self.cmd,
                 self.args,
                 env=os.environ
             )
             return self.start_deferred
         else:
             s = 'the process has already been started and has state: %r' % \
                 self.state
             return defer.fail(ProcessStateError(s))
     
     def get_stop_deferred(self):
         if self.state == 'running' or self.state == 'before':
             d = defer.Deferred()
             self.stop_deferreds.append(d)
             return d
         else:
             s = 'this process is already complete'
             return defer.fail(ProcessStateError(s))
     
     def fire_stop_deferred(self, exit_code):
         log.msg('Process %r has stopped with %r' % (self.args, exit_code))
         self.state = 'after'
         for d in self.stop_deferreds:
             d.callback(exit_code)
     
     def signal(self, sig):
         """
         Send a signal to the process.
         
         The argument sig can be ('KILL','INT', etc.) or any signal number.
         """
         if self.state == 'running':
             self.process_transport.signalProcess(sig)
     
     # def __del__(self):
     #     self.signal('KILL')
     
     def interrupt_then_kill(self, delay=1.0):
         self.signal('INT')
         reactor.callLater(delay, self.signal, 'KILL')
 
 
 #-----------------------------------------------------------------------------
 # Code for launching controller and engines
 #-----------------------------------------------------------------------------
 
 
 class ControllerLauncher(ProcessLauncher):
     
     def __init__(self, extra_args=None):
         if sys.platform == 'win32':
             # This logic is needed because the ipcontroller script doesn't
             # always get installed in the same way or in the same location.
             from IPython.kernel.scripts import ipcontroller
             script_location = ipcontroller.__file__.replace('.pyc', '.py')
             # The -u option here turns on unbuffered output, which is required
             # on Win32 to prevent wierd conflict and problems with Twisted
             args = [find_exe('python'), '-u', script_location]
         else:
             args = ['ipcontroller']
         self.extra_args = extra_args
         if extra_args is not None:
             args.extend(extra_args)
         
         ProcessLauncher.__init__(self, args)
 
 
 class EngineLauncher(ProcessLauncher):
     
     def __init__(self, extra_args=None):
         if sys.platform == 'win32':
             # This logic is needed because the ipcontroller script doesn't
             # always get installed in the same way or in the same location.
             from IPython.kernel.scripts import ipengine
             script_location = ipengine.__file__.replace('.pyc', '.py')
             # The -u option here turns on unbuffered output, which is required
             # on Win32 to prevent wierd conflict and problems with Twisted
             args = [find_exe('python'), '-u', script_location]
         else:
             args = ['ipengine']
         self.extra_args = extra_args
         if extra_args is not None:
             args.extend(extra_args)
         
         ProcessLauncher.__init__(self, args)
 
 
 class LocalEngineSet(object):
     
     def __init__(self, extra_args=None):
         self.extra_args = extra_args
         self.launchers = []
     
     def start(self, n):
         dlist = []
         for i in range(n):
             el = EngineLauncher(extra_args=self.extra_args)
             d = el.start()
             self.launchers.append(el)
             dlist.append(d)
         dfinal = gatherBoth(dlist, consumeErrors=True)
         dfinal.addCallback(self._handle_start)
         return dfinal
     
     def _handle_start(self, r):
         log.msg('Engines started with pids: %r' % r)
         return r
     
     def _handle_stop(self, r):
         log.msg('Engines received signal: %r' % r)
         return r
     
     def signal(self, sig):
         dlist = []
         for el in self.launchers:
             d = el.get_stop_deferred()
             dlist.append(d)
             el.signal(sig)
         dfinal = gatherBoth(dlist, consumeErrors=True)
         dfinal.addCallback(self._handle_stop)
         return dfinal
     
     def interrupt_then_kill(self, delay=1.0):
         dlist = []
         for el in self.launchers:
             d = el.get_stop_deferred()
             dlist.append(d)
             el.interrupt_then_kill(delay)
         dfinal = gatherBoth(dlist, consumeErrors=True)
         dfinal.addCallback(self._handle_stop)
         return dfinal
 
 
 class BatchEngineSet(object):
     
     # Subclasses must fill these in.  See PBSEngineSet
     submit_command = ''
     delete_command = ''
     job_id_regexp = ''
     
     def __init__(self, template_file, **kwargs):
         self.template_file = template_file
         self.context = {}
         self.context.update(kwargs)
         self.batch_file = self.template_file+'-run'
     
     def parse_job_id(self, output):
         m = re.match(self.job_id_regexp, output)
         if m is not None:
             job_id = m.group()
         else:
             raise Exception("job id couldn't be determined: %s" % output)
         self.job_id = job_id
         log.msg('Job started with job id: %r' % job_id)
         return job_id
     
     def write_batch_script(self, n):
         self.context['n'] = n
         template = open(self.template_file, 'r').read()
         log.msg('Using template for batch script: %s' % self.template_file)
         script_as_string = Itpl.itplns(template, self.context)
         log.msg('Writing instantiated batch script: %s' % self.batch_file)
         f = open(self.batch_file,'w')
         f.write(script_as_string)
         f.close()
     
     def handle_error(self, f):
         f.printTraceback()
         f.raiseException()
     
     def start(self, n):
         self.write_batch_script(n)
         d = getProcessOutput(self.submit_command,
             [self.batch_file],env=os.environ)
         d.addCallback(self.parse_job_id)
         d.addErrback(self.handle_error)
         return d
     
     def kill(self):
         d = getProcessOutput(self.delete_command,
             [self.job_id],env=os.environ)
         return d
 
 class PBSEngineSet(BatchEngineSet):
     
     submit_command = 'qsub'
     delete_command = 'qdel'
     job_id_regexp = '\d+'
     
     def __init__(self, template_file, **kwargs):
         BatchEngineSet.__init__(self, template_file, **kwargs)
 
 
 sshx_template="""#!/bin/sh
 "$@" &> /dev/null &
 echo $!
 """
 
 engine_killer_template="""#!/bin/sh
 ps -fu `whoami` | grep '[i]pengine' | awk '{print $2}' | xargs kill -TERM
 """
 
 class SSHEngineSet(object):
     sshx_template=sshx_template
     engine_killer_template=engine_killer_template
     
     def __init__(self, engine_hosts, sshx=None, ipengine="ipengine"):
         """Start a controller on localhost and engines using ssh.
         
         The engine_hosts argument is a dict with hostnames as keys and
         the number of engine (int) as values.  sshx is the name of a local
         file that will be used to run remote commands.  This file is used
         to setup the environment properly.
         """
         
         self.temp_dir = tempfile.gettempdir()
         if sshx is not None:
             self.sshx = sshx
         else:
             # Write the sshx.sh file locally from our template.
             self.sshx = os.path.join(
                 self.temp_dir,
                 '%s-main-sshx.sh' % os.environ['USER']
             )
             f = open(self.sshx, 'w')
             f.writelines(self.sshx_template)
             f.close()
         self.engine_command = ipengine
         self.engine_hosts = engine_hosts
         # Write the engine killer script file locally from our template.
         self.engine_killer = os.path.join(
             self.temp_dir,  
             '%s-local-engine_killer.sh' % os.environ['USER']
         )
         f = open(self.engine_killer, 'w')
         f.writelines(self.engine_killer_template)
         f.close()
     
     def start(self, send_furl=False):
         dlist = []
         for host in self.engine_hosts.keys():
             count = self.engine_hosts[host]
             d = self._start(host, count, send_furl)
             dlist.append(d)
         return gatherBoth(dlist, consumeErrors=True)
     
     def _start(self, hostname, count=1, send_furl=False):
         if send_furl:
             d = self._scp_furl(hostname)
         else:
             d = defer.succeed(None)
         d.addCallback(lambda r: self._scp_sshx(hostname))
         d.addCallback(lambda r: self._ssh_engine(hostname, count))
         return d
         
     def _scp_furl(self, hostname):
         scp_cmd = "scp ~/.ipython/security/ipcontroller-engine.furl %s:.ipython/security/" % (hostname)
         cmd_list = scp_cmd.split()
         cmd_list[1] = os.path.expanduser(cmd_list[1])
         log.msg('Copying furl file: %s' % scp_cmd)
         d = getProcessOutput(cmd_list[0], cmd_list[1:], env=os.environ) 
         return d
     
     def _scp_sshx(self, hostname):
         scp_cmd = "scp %s %s:%s/%s-sshx.sh" % (
             self.sshx, hostname, 
             self.temp_dir, os.environ['USER']
         )
         print
         log.msg("Copying sshx: %s" % scp_cmd)
         sshx_scp = scp_cmd.split()
         d = getProcessOutput(sshx_scp[0], sshx_scp[1:], env=os.environ)
         return d
     
     def _ssh_engine(self, hostname, count):
         exec_engine = "ssh %s sh %s/%s-sshx.sh %s" % (
             hostname, self.temp_dir, 
             os.environ['USER'], self.engine_command
         )
         cmds = exec_engine.split()
         dlist = []
         log.msg("about to start engines...")
         for i in range(count):
             log.msg('Starting engines: %s' % exec_engine)
             d = getProcessOutput(cmds[0], cmds[1:], env=os.environ)
             dlist.append(d)
         return gatherBoth(dlist, consumeErrors=True)
     
     def kill(self):
         dlist = []
         for host in self.engine_hosts.keys():
             d = self._killall(host)
             dlist.append(d)
         return gatherBoth(dlist, consumeErrors=True)
     
     def _killall(self, hostname):
         d = self._scp_engine_killer(hostname)
         d.addCallback(lambda r: self._ssh_kill(hostname))
         # d.addErrback(self._exec_err)
         return d
 
     def _scp_engine_killer(self, hostname):
         scp_cmd = "scp %s %s:%s/%s-engine_killer.sh" % (
             self.engine_killer, 
             hostname, 
             self.temp_dir, 
             os.environ['USER']
         )
         cmds = scp_cmd.split()
         log.msg('Copying engine_killer: %s' % scp_cmd)
         d = getProcessOutput(cmds[0], cmds[1:], env=os.environ)
         return d
     
     def _ssh_kill(self, hostname):
         kill_cmd = "ssh %s sh %s/%s-engine_killer.sh" % (
             hostname,
             self.temp_dir, 
             os.environ['USER']
         )
         log.msg('Killing engine: %s' % kill_cmd)
         kill_cmd = kill_cmd.split()
         d = getProcessOutput(kill_cmd[0], kill_cmd[1:], env=os.environ)
         return d
 
     def _exec_err(self, r):
         log.msg(r)
 
 #-----------------------------------------------------------------------------
 # Main functions for the different types of clusters
 #-----------------------------------------------------------------------------
 
 # TODO:
 # The logic in these codes should be moved into classes like LocalCluster
 # MpirunCluster, PBSCluster, etc.  This would remove alot of the duplications.
 # The main functions should then just parse the command line arguments, create
 # the appropriate class and call a 'start' method.
 
 def check_security(args, cont_args):
     if (not args.x or not args.y) and not have_crypto:
         log.err("""
 OpenSSL/pyOpenSSL is not available, so we can't run in secure mode.
 Try running ipcluster with the -xy flags:  ipcluster local -xy -n 4""")
         reactor.stop()
         return False
     if args.x:
         cont_args.append('-x')
     if args.y:
         cont_args.append('-y')
     return True
 
+def check_reuse(args, cont_args):
+    if args.r:
+        cont_args.append('-r')
+        if args.client_port == 0 or args.engine_port == 0:
+            log.err("""
+To reuse FURL files, you must also set the client and engine ports using
+the --client-port and --engine-port options.""")
+            reactor.stop()
+            return False
+        cont_args.append('--client-port=%i' % args.client_port)
+        cont_args.append('--engine-port=%i' % args.engine_port)
+    return True
 
 def main_local(args):
     cont_args = []
     cont_args.append('--logfile=%s' % pjoin(args.logdir,'ipcontroller'))
-
+    
     # Check security settings before proceeding
     if not check_security(args, cont_args):
         return
-
+    
+    # See if we are reusing FURL files
+    if not check_reuse(args, cont_args):
+        return
+    
     cl = ControllerLauncher(extra_args=cont_args)
     dstart = cl.start()
     def start_engines(cont_pid):
         engine_args = []
         engine_args.append('--logfile=%s' % \
             pjoin(args.logdir,'ipengine%s-' % cont_pid))
         eset = LocalEngineSet(extra_args=engine_args)
         def shutdown(signum, frame):
             log.msg('Stopping local cluster')
             # We are still playing with the times here, but these seem
             # to be reliable in allowing everything to exit cleanly.
             eset.interrupt_then_kill(0.5)
             cl.interrupt_then_kill(0.5)
             reactor.callLater(1.0, reactor.stop)
         signal.signal(signal.SIGINT,shutdown)
         d = eset.start(args.n)
         return d
     def delay_start(cont_pid):
         # This is needed because the controller doesn't start listening
         # right when it starts and the controller needs to write
         # furl files for the engine to pick up
         reactor.callLater(1.0, start_engines, cont_pid)
     dstart.addCallback(delay_start)
     dstart.addErrback(lambda f: f.raiseException())
 
 
-def main_mpirun(args):
+def main_mpi(args):
     cont_args = []
     cont_args.append('--logfile=%s' % pjoin(args.logdir,'ipcontroller'))
-
+    
     # Check security settings before proceeding
     if not check_security(args, cont_args):
         return
-
+    
+    # See if we are reusing FURL files
+    if not check_reuse(args, cont_args):
+        return
+    
     cl = ControllerLauncher(extra_args=cont_args)
     dstart = cl.start()
     def start_engines(cont_pid):
-        raw_args = ['mpirun']
+        raw_args = [args.cmd]
         raw_args.extend(['-n',str(args.n)])
         raw_args.append('ipengine')
         raw_args.append('-l')
         raw_args.append(pjoin(args.logdir,'ipengine%s-' % cont_pid))
         if args.mpi:
             raw_args.append('--mpi=%s' % args.mpi)
         eset = ProcessLauncher(raw_args)
         def shutdown(signum, frame):
             log.msg('Stopping local cluster')
             # We are still playing with the times here, but these seem
             # to be reliable in allowing everything to exit cleanly.
             eset.interrupt_then_kill(1.0)
             cl.interrupt_then_kill(1.0)
             reactor.callLater(2.0, reactor.stop)
         signal.signal(signal.SIGINT,shutdown)
         d = eset.start()
         return d
     def delay_start(cont_pid):
         # This is needed because the controller doesn't start listening
         # right when it starts and the controller needs to write
         # furl files for the engine to pick up
         reactor.callLater(1.0, start_engines, cont_pid)
     dstart.addCallback(delay_start)
     dstart.addErrback(lambda f: f.raiseException())
 
 
 def main_pbs(args):
     cont_args = []
     cont_args.append('--logfile=%s' % pjoin(args.logdir,'ipcontroller'))
-
+    
     # Check security settings before proceeding
     if not check_security(args, cont_args):
         return
-
+    
+    # See if we are reusing FURL files
+    if not check_reuse(args, cont_args):
+        return
+    
     cl = ControllerLauncher(extra_args=cont_args)
     dstart = cl.start()
     def start_engines(r):
         pbs_set =  PBSEngineSet(args.pbsscript)
         def shutdown(signum, frame):
             log.msg('Stopping pbs cluster')
             d = pbs_set.kill()
             d.addBoth(lambda _: cl.interrupt_then_kill(1.0))
             d.addBoth(lambda _: reactor.callLater(2.0, reactor.stop))
         signal.signal(signal.SIGINT,shutdown)
         d = pbs_set.start(args.n)
         return d
     dstart.addCallback(start_engines)
     dstart.addErrback(lambda f: f.raiseException())
 
 
 def main_ssh(args):
     """Start a controller on localhost and engines using ssh.
     
     Your clusterfile should look like::
     
         send_furl = False # True, if you want 
         engines = {
             'engine_host1' : engine_count, 
             'engine_host2' : engine_count2
         } 
     """
     clusterfile = {}
     execfile(args.clusterfile, clusterfile)
     if not clusterfile.has_key('send_furl'):
         clusterfile['send_furl'] = False
         
     cont_args = []
     cont_args.append('--logfile=%s' % pjoin(args.logdir,'ipcontroller'))
     
     # Check security settings before proceeding
     if not check_security(args, cont_args):
         return
     
+    # See if we are reusing FURL files
+    if not check_reuse(args, cont_args):
+        return
+    
     cl = ControllerLauncher(extra_args=cont_args)
     dstart = cl.start()
     def start_engines(cont_pid):
         ssh_set = SSHEngineSet(clusterfile['engines'], sshx=args.sshx)
         def shutdown(signum, frame):
             d = ssh_set.kill()
-            # d.addErrback(log.err)
             cl.interrupt_then_kill(1.0)
             reactor.callLater(2.0, reactor.stop)
         signal.signal(signal.SIGINT,shutdown)
         d = ssh_set.start(clusterfile['send_furl'])
         return d
     
     def delay_start(cont_pid):
         reactor.callLater(1.0, start_engines, cont_pid)
         
     dstart.addCallback(delay_start)
     dstart.addErrback(lambda f: f.raiseException())
 
 
 def get_args():
     base_parser = argparse.ArgumentParser(add_help=False)
     base_parser.add_argument(
+        '-r',
+        action='store_true',
+        dest='r',
+        help='try to reuse FURL files.  Use with --client-port and --engine-port'
+    )
+    base_parser.add_argument(
+        '--client-port',
+        type=int,
+        dest='client_port',
+        help='the port the controller will listen on for client connections',
+        default=0
+    )
+    base_parser.add_argument(
+        '--engine-port',
+        type=int,
+        dest='engine_port',
+        help='the port the controller will listen on for engine connections',
+        default=0
+    )
+    base_parser.add_argument(
         '-x',
         action='store_true',
         dest='x',
         help='turn off client security'
     )
     base_parser.add_argument(
         '-y',
         action='store_true',
         dest='y',
         help='turn off engine security'
     )
     base_parser.add_argument(
         "--logdir", 
         type=str,
         dest="logdir",
         help="directory to put log files (default=$IPYTHONDIR/log)",
         default=pjoin(get_ipython_dir(),'log')
     )
     base_parser.add_argument(
         "-n",
         "--num", 
         type=int,
         dest="n",
         default=2,
         help="the number of engines to start"
     )
     
     parser = argparse.ArgumentParser(
         description='IPython cluster startup.  This starts a controller and\
         engines using various approaches.  THIS IS A TECHNOLOGY PREVIEW AND\
         THE API WILL CHANGE SIGNIFICANTLY BEFORE THE FINAL RELEASE.'
     )
     subparsers = parser.add_subparsers(
         help='available cluster types.  For help, do "ipcluster TYPE --help"')
     
     parser_local = subparsers.add_parser(
         'local',
         help='run a local cluster',
         parents=[base_parser]
     )
     parser_local.set_defaults(func=main_local)
     
     parser_mpirun = subparsers.add_parser(
         'mpirun',
-        help='run a cluster using mpirun',
+        help='run a cluster using mpirun (mpiexec also works)',
         parents=[base_parser]
     )
     parser_mpirun.add_argument(
         "--mpi",
         type=str,
         dest="mpi", # Don't put a default here to allow no MPI support
         help="how to call MPI_Init (default=mpi4py)"
     )
-    parser_mpirun.set_defaults(func=main_mpirun)
+    parser_mpirun.set_defaults(func=main_mpi, cmd='mpirun')
+
+    parser_mpiexec = subparsers.add_parser(
+        'mpiexec',
+        help='run a cluster using mpiexec (mpirun also works)',
+        parents=[base_parser]
+    )
+    parser_mpiexec.add_argument(
+        "--mpi",
+        type=str,
+        dest="mpi", # Don't put a default here to allow no MPI support
+        help="how to call MPI_Init (default=mpi4py)"
+    )
+    parser_mpiexec.set_defaults(func=main_mpi, cmd='mpiexec')
     
     parser_pbs = subparsers.add_parser(
         'pbs', 
         help='run a pbs cluster',
         parents=[base_parser]
     )
     parser_pbs.add_argument(
         '--pbs-script',
         type=str, 
         dest='pbsscript',
         help='PBS script template',
         default='pbs.template'
     )
     parser_pbs.set_defaults(func=main_pbs)
     
     parser_ssh = subparsers.add_parser(
         'ssh',
         help='run a cluster using ssh, should have ssh-keys setup',
         parents=[base_parser]
     )
     parser_ssh.add_argument(
         '--clusterfile', 
         type=str,
         dest='clusterfile',
         help='python file describing the cluster',
         default='clusterfile.py',
     )
     parser_ssh.add_argument(
         '--sshx', 
         type=str,
         dest='sshx',
         help='sshx launcher helper'
     )
     parser_ssh.set_defaults(func=main_ssh)
     
     args = parser.parse_args()
     return args
 
 def main():
     args = get_args()
     reactor.callWhenRunning(args.func, args)
     log.startLogging(sys.stdout)
     reactor.run()
 
 if __name__ == '__main__':
     main()

docs/source/changes.txt

@@ -1,398 +1,401 b''
 .. _changes:
 
 ==========
 What's new
 ==========
 
 .. contents::
 ..
     1  Release 0.9.1
     2  Release 0.9
       2.1  New features
       2.2  Bug fixes
       2.3  Backwards incompatible changes
       2.4  Changes merged in from IPython1
         2.4.1  New features
         2.4.2  Bug fixes
         2.4.3  Backwards incompatible changes
     3  Release 0.8.4
     4  Release 0.8.3
     5  Release 0.8.2
     6  Older releases
 ..
 
 Release dev
 ===========
 
 New features
 ------------
 
 * The new ipcluster now has a fully working ssh mode that should work on
   Linux, Unix and OS X.  Thanks to Vishal Vatsa for implementing this!
 
 * The wonderful TextMate editor can now be used with %edit on OS X.  Thanks
   to Matt Foster for this patch.
 
 * Fully refactored :command:`ipcluster` command line program for starting
   IPython clusters.  This new version is a complete rewrite and 1) is fully
   cross platform (we now use Twisted's process management), 2) has much
   improved performance, 3) uses subcommands for different types of clusters, 
   4) uses argparse for parsing command line options, 5) has better support
   for starting clusters using :command:`mpirun`, 6) has experimental support
   for starting engines using PBS.  However, this new version of ipcluster
   should be considered a technology preview.  We plan on changing the API
   in significant ways before it is final.
 
 * The :mod:`argparse` module has been added to :mod:`IPython.external`.
 
 * Fully description of the security model added to the docs.
 
 * cd completer: show bookmarks if no other completions are available.
 
 * sh profile: easy way to give 'title' to prompt: assign to variable
   '_prompt_title'. It looks like this::
       
         [~]|1> _prompt_title = 'sudo!'
         sudo![~]|2>
 
 * %edit: If you do '%edit pasted_block', pasted_block
   variable gets updated with new data (so repeated 
   editing makes sense)
 
 Bug fixes
 ---------
 
 * Numerous bugs on Windows with the new ipcluster have been fixed.
 
 * The ipengine and ipcontroller scripts now handle missing furl files
   more gracefully by giving better error messages.
 
 * %rehashx: Aliases no longer contain dots. python3.0 binary
   will create alias python30. Fixes:
   #259716 "commands with dots in them don't work"
 
 * %cpaste: %cpaste -r repeats the last pasted block.
   The block is assigned to pasted_block even if code
   raises exception.
 
+* Bug #274067 'The code in get_home_dir is broken for py2exe' was
+  fixed.
+
 Backwards incompatible changes
 ------------------------------
 
 * The controller now has a ``-r`` flag that needs to be used if you want to
   reuse existing furl files.  Otherwise they are deleted (the default).
 
 * Remove ipy_leo.py. "easy_install ipython-extension" to get it.
   (done to decouple it from ipython release cycle)
 
 
 
 Release 0.9.1
 =============
 
 This release was quickly made to restore compatibility with Python 2.4, which
 version 0.9 accidentally broke.  No new features were introduced, other than
 some additional testing support for internal use.
 
 
 Release 0.9
 ===========
 
 New features
 ------------
 
 * All furl files and security certificates are now put in a read-only
   directory named ~./ipython/security.
 
 * A single function :func:`get_ipython_dir`, in :mod:`IPython.genutils` that
   determines the user's IPython directory in a robust manner.
 
 * Laurent's WX application has been given a top-level script called
   ipython-wx, and it has received numerous fixes. We expect this code to be
   architecturally better integrated with Gael's WX 'ipython widget' over the
   next few releases.
 
 * The Editor synchronization work by Vivian De Smedt has been merged in.  This
   code adds a number of new editor hooks to synchronize with editors under
   Windows.
 
 * A new, still experimental but highly functional, WX shell by Gael Varoquaux.
   This work was sponsored by Enthought, and while it's still very new, it is
   based on a more cleanly organized arhictecture of the various IPython
   components. We will continue to develop this over the next few releases as a
   model for GUI components that use IPython.
 
 * Another GUI frontend, Cocoa based (Cocoa is the OSX native GUI framework),
   authored by Barry Wark.  Currently the WX and the Cocoa ones have slightly
   different internal organizations, but the whole team is working on finding
   what the right abstraction points are for a unified codebase.
 
 * As part of the frontend work, Barry Wark also implemented an experimental
   event notification system that various ipython components can use.  In the
   next release the implications and use patterns of this system regarding the
   various GUI options will be worked out.
 
 * IPython finally has a full test system, that can test docstrings with
   IPython-specific functionality.  There are still a few pieces missing for it
   to be widely accessible to all users (so they can run the test suite at any
   time and report problems), but it now works for the developers.  We are
   working hard on continuing to improve it, as this was probably IPython's
   major Achilles heel (the lack of proper test coverage made it effectively
   impossible to do large-scale refactoring).  The full test suite can now
   be run using the :command:`iptest` command line program.
 
 * The notion of a task has been completely reworked.  An `ITask` interface has
   been created.  This interface defines the methods that tasks need to
   implement.  These methods are now responsible for things like submitting
   tasks and processing results.  There are two basic task types:
   :class:`IPython.kernel.task.StringTask` (this is the old `Task` object, but
   renamed) and the new :class:`IPython.kernel.task.MapTask`, which is based on
   a function.
 
 * A new interface, :class:`IPython.kernel.mapper.IMapper` has been defined to
   standardize the idea of a `map` method.  This interface has a single `map`
   method that has the same syntax as the built-in `map`.  We have also defined
   a `mapper` factory interface that creates objects that implement
   :class:`IPython.kernel.mapper.IMapper` for different controllers.  Both the
   multiengine and task controller now have mapping capabilties.
 
 * The parallel function capabilities have been reworks.  The major changes are
   that i) there is now an `@parallel` magic that creates parallel functions,
   ii) the syntax for mulitple variable follows that of `map`, iii) both the
   multiengine and task controller now have a parallel function implementation.
 
 * All of the parallel computing capabilities from `ipython1-dev` have been
   merged into IPython proper.  This resulted in the following new subpackages:
   :mod:`IPython.kernel`, :mod:`IPython.kernel.core`, :mod:`IPython.config`,
   :mod:`IPython.tools` and :mod:`IPython.testing`.
 
 * As part of merging in the `ipython1-dev` stuff, the `setup.py` script and
   friends have been completely refactored.  Now we are checking for
   dependencies using the approach that matplotlib uses.
 
 * The documentation has been completely reorganized to accept the 
   documentation from `ipython1-dev`.
 
 * We have switched to using Foolscap for all of our network protocols in
   :mod:`IPython.kernel`.  This gives us secure connections that are both
   encrypted and authenticated.
 
 * We have a brand new `COPYING.txt` files that describes the IPython license
   and copyright. The biggest change is that we are putting "The IPython
   Development Team" as the copyright holder. We give more details about
   exactly what this means in this file. All developer should read this and use
   the new banner in all IPython source code files.
 
 * sh profile: ./foo runs foo as system command, no need to do !./foo anymore
 
 * String lists now support ``sort(field, nums = True)`` method (to easily sort
   system command output). Try it with ``a = !ls -l ; a.sort(1, nums=1)``.
 
 * '%cpaste foo' now assigns the pasted block as string list, instead of string
 
 * The ipcluster script now run by default with no security.  This is done
   because the main usage of the script is for starting things on localhost.
   Eventually when ipcluster is able to start things on other hosts, we will put
   security back.
 
 * 'cd --foo' searches directory history for string foo, and jumps to that dir.
   Last part of dir name is checked first. If no matches for that are found,
   look at the whole path.
 
   
 Bug fixes
 ---------
 
 * The Windows installer has been fixed.  Now all IPython scripts have ``.bat``
   versions created.  Also, the Start Menu shortcuts have been updated.
 
 * The colors escapes in the multiengine client are now turned off on win32 as
   they don't print correctly.
 
 * The :mod:`IPython.kernel.scripts.ipengine` script was exec'ing
   mpi_import_statement incorrectly, which was leading the engine to crash when
   mpi was enabled.
 
 * A few subpackages had missing ``__init__.py`` files.
 
 * The documentation is only created if Sphinx is found.  Previously, the
   ``setup.py`` script would fail if it was missing.
 
 * Greedy ``cd`` completion has been disabled again (it was enabled in 0.8.4) as
   it caused problems on certain platforms.
           
 
 Backwards incompatible changes
 ------------------------------
 
 * The ``clusterfile`` options of the :command:`ipcluster` command has been
   removed as it was not working and it will be replaced soon by something much
   more robust.
 
 * The :mod:`IPython.kernel` configuration now properly find the user's
   IPython directory.
 
 * In ipapi, the :func:`make_user_ns` function has been replaced with
   :func:`make_user_namespaces`, to support dict subclasses in namespace
   creation.
 
 * :class:`IPython.kernel.client.Task` has been renamed
   :class:`IPython.kernel.client.StringTask` to make way for new task types.
 
 * The keyword argument `style` has been renamed `dist` in `scatter`, `gather`
   and `map`.
 
 * Renamed the values that the rename `dist` keyword argument can have from
   `'basic'` to `'b'`.
 
 * IPython has a larger set of dependencies if you want all of its capabilities.
   See the `setup.py` script for details.
 
 * The constructors for :class:`IPython.kernel.client.MultiEngineClient` and 
   :class:`IPython.kernel.client.TaskClient` no longer take the (ip,port) tuple.
   Instead they take the filename of a file that contains the FURL for that
   client.  If the FURL file is in your IPYTHONDIR, it will be found automatically
   and the constructor can be left empty.
 
 * The asynchronous clients in :mod:`IPython.kernel.asyncclient` are now created 
   using the factory functions :func:`get_multiengine_client` and 
   :func:`get_task_client`.  These return a `Deferred` to the actual client.
 
 * The command line options to `ipcontroller` and `ipengine` have changed to
   reflect the new Foolscap network protocol and the FURL files.  Please see the
   help for these scripts for details.
 
 * The configuration files for the kernel have changed because of the Foolscap
   stuff.  If you were using custom config files before, you should delete them
   and regenerate new ones.
 
 Changes merged in from IPython1
 -------------------------------
 
 New features
 ............
 
 * Much improved ``setup.py`` and ``setupegg.py`` scripts.  Because Twisted and
   zope.interface are now easy installable, we can declare them as dependencies
   in our setupegg.py script.
 
 * IPython is now compatible with Twisted 2.5.0 and 8.x.
 
 * Added a new example of how to use :mod:`ipython1.kernel.asynclient`.
 
 * Initial draft of a process daemon in :mod:`ipython1.daemon`.  This has not
   been merged into IPython and is still in `ipython1-dev`.
   
 * The ``TaskController`` now has methods for getting the queue status.
 
 * The ``TaskResult`` objects not have information about how long the task
   took to run.
   
 * We are attaching additional attributes to exceptions ``(_ipython_*)`` that 
   we use to carry additional info around.
   
 * New top-level module :mod:`asyncclient` that has asynchronous versions (that
   return deferreds) of the client classes.  This is designed to users who want
   to run their own Twisted reactor.
   
 * All the clients in :mod:`client` are now based on Twisted.  This is done by 
   running the Twisted reactor in a separate thread and using the
   :func:`blockingCallFromThread` function that is in recent versions of Twisted.
 
 * Functions can now be pushed/pulled to/from engines using
   :meth:`MultiEngineClient.push_function` and
   :meth:`MultiEngineClient.pull_function`.
 
 * Gather/scatter are now implemented in the client to reduce the work load
   of the controller and improve performance.
 
 * Complete rewrite of the IPython docuementation.  All of the documentation
   from the IPython website has been moved into docs/source as restructured
   text documents.  PDF and HTML documentation are being generated using 
   Sphinx.
 
 * New developer oriented documentation: development guidelines and roadmap. 
 
 * Traditional ``ChangeLog`` has been changed to a more useful ``changes.txt``
   file that is organized by release and is meant to provide something more
   relevant for users.
 
 Bug fixes
 .........
 
 * Created a proper ``MANIFEST.in`` file to create source distributions.
 
 * Fixed a bug in the ``MultiEngine`` interface.  Previously, multi-engine 
   actions were being collected with a :class:`DeferredList` with 
   ``fireononeerrback=1``.  This meant that methods were returning 
   before all engines had given their results.  This was causing extremely odd 
   bugs in certain cases. To fix this problem, we have 1) set 
   ``fireononeerrback=0`` to make sure all results (or exceptions) are in 
   before returning and 2) introduced a :exc:`CompositeError` exception 
   that wraps all of the engine exceptions.  This is a huge change as it means 
   that users will have to catch :exc:`CompositeError` rather than the actual
   exception.
 
 Backwards incompatible changes
 ..............................
 
 * All names have been renamed to conform to the lowercase_with_underscore
   convention.  This will require users to change references to all names like
   ``queueStatus`` to ``queue_status``.
 
 * Previously, methods like :meth:`MultiEngineClient.push` and
   :meth:`MultiEngineClient.push` used ``*args`` and ``**kwargs``.  This was
   becoming a problem as we weren't able to introduce new keyword arguments into
   the API.  Now these methods simple take a dict or sequence.  This has also
   allowed us to get rid of the ``*All`` methods like :meth:`pushAll` and
   :meth:`pullAll`.  These things are now handled with the ``targets`` keyword
   argument that defaults to ``'all'``.
 
 * The :attr:`MultiEngineClient.magicTargets` has been renamed to
   :attr:`MultiEngineClient.targets`. 
 
 * All methods in the MultiEngine interface now accept the optional keyword
   argument ``block``.
 
 * Renamed :class:`RemoteController` to :class:`MultiEngineClient` and 
   :class:`TaskController` to :class:`TaskClient`.
 
 * Renamed the top-level module from :mod:`api` to :mod:`client`.
 
 * Most methods in the multiengine interface now raise a :exc:`CompositeError`
   exception that wraps the user's exceptions, rather than just raising the raw
   user's exception.
 
 * Changed the ``setupNS`` and ``resultNames`` in the ``Task`` class to ``push`` 
   and ``pull``.
 
   
 Release 0.8.4
 =============
 
 This was a quick release to fix an unfortunate bug that slipped into the 0.8.3
 release.  The ``--twisted`` option was disabled, as it turned out to be broken
 across several platforms.
 
 
 Release 0.8.3
 =============
   
 * pydb is now disabled by default (due to %run -d problems). You can enable
   it by passing -pydb command line argument to IPython. Note that setting
   it in config file won't work.
 
   
 Release 0.8.2
 =============
 
 * %pushd/%popd behave differently; now "pushd /foo" pushes CURRENT directory 
   and jumps to /foo. The current behaviour is closer to the documented 
   behaviour, and should not trip anyone.
 
   
 Older releases
 ==============
 
 Changes in earlier releases of IPython are described in the older file
 ``ChangeLog``.  Please refer to this document for details.
 

docs/source/parallel/parallel_mpi.txt

@@ -1,157 +1,157 b''
 .. _parallelmpi:
 
 =======================
 Using MPI with IPython
 =======================
 
 Often, a parallel algorithm will require moving data between the engines.  One way of accomplishing this is by doing a pull and then a push using the multiengine client.  However, this will be slow as all the data has to go through the controller to the client and then back through the controller, to its final destination.
 
 A much better way of moving data between engines is to use a message passing library, such as the Message Passing Interface (MPI) [MPI]_.  IPython's parallel computing architecture has been designed from the ground up to integrate with MPI.  This document describes how to use MPI with IPython.
 
 Additional installation requirements
 ====================================
 
 If you want to use MPI with IPython, you will need to install:
 
 * A standard MPI implementation such as OpenMPI [OpenMPI]_ or MPICH.
 * The mpi4py [mpi4py]_ package.
 
 .. note::
 
     The mpi4py package is not a strict requirement. However, you need to
     have *some* way of calling MPI from Python. You also need some way of
     making sure that :func:`MPI_Init` is called when the IPython engines start
     up. There are a number of ways of doing this and a good number of
     associated subtleties. We highly recommend just using mpi4py as it
     takes care of most of these problems. If you want to do something
     different, let us know and we can help you get started.
 
 Starting the engines with MPI enabled
 =====================================
 
 To use code that calls MPI, there are typically two things that MPI requires.
 
 1. The process that wants to call MPI must be started using
-   :command:`mpirun` or a batch system (like PBS) that has MPI support.
+   :command:`mpiexec` or a batch system (like PBS) that has MPI support.
 2. Once the process starts, it must call :func:`MPI_Init`.
 
 There are a couple of ways that you can start the IPython engines and get these things to happen.
 
-Automatic starting using :command:`mpirun` and :command:`ipcluster`
+Automatic starting using :command:`mpiexec` and :command:`ipcluster`
 -------------------------------------------------------------------
 
-The easiest approach is to use the `mpirun` mode of :command:`ipcluster`, which will first start a controller and then a set of engines using :command:`mpirun`::
+The easiest approach is to use the `mpiexec` mode of :command:`ipcluster`, which will first start a controller and then a set of engines using :command:`mpiexec`::
 
-    $ ipcluster mpirun -n 4
+    $ ipcluster mpiexec -n 4
 
 This approach is best as interrupting :command:`ipcluster` will automatically
 stop and clean up the controller and engines.
 
-Manual starting using :command:`mpirun`
+Manual starting using :command:`mpiexec`
 ---------------------------------------
 
-If you want to start the IPython engines using the :command:`mpirun`, just do::
+If you want to start the IPython engines using the :command:`mpiexec`, just do::
 
-    $ mpirun -n 4 ipengine --mpi=mpi4py
+    $ mpiexec -n 4 ipengine --mpi=mpi4py
 
 This requires that you already have a controller running and that the FURL
 files for the engines are in place. We also have built in support for
 PyTrilinos [PyTrilinos]_, which can be used (assuming is installed) by
 starting the engines with::
 
-    	mpirun -n 4 ipengine --mpi=pytrilinos
+    	mpiexec -n 4 ipengine --mpi=pytrilinos
 
 Automatic starting using PBS and :command:`ipcluster`
 -----------------------------------------------------
 
 The :command:`ipcluster` command also has built-in integration with PBS. For more information on this approach, see our documentation on :ref:`ipcluster <parallel_process>`.
 
 Actually using MPI
 ==================
 
 Once the engines are running with MPI enabled, you are ready to go.  You can now call any code that uses MPI in the IPython engines.  And, all of this can be done interactively.  Here we show a simple example that uses mpi4py [mpi4py]_.
 
 First, lets define a simply function that uses MPI to calculate the sum of a distributed array.  Save the following text in a file called :file:`psum.py`:
 
 .. sourcecode:: python
 
     from mpi4py import MPI
     import numpy as np
     
     def psum(a):
         s = np.sum(a)
         return MPI.COMM_WORLD.Allreduce(s,MPI.SUM)
 
 Now, start an IPython cluster in the same directory as :file:`psum.py`::
 
-    $ ipcluster mpirun -n 4
+    $ ipcluster mpiexec -n 4
 
 Finally, connect to the cluster and use this function interactively.  In this case, we create a random array on each engine and sum up all the random arrays using our :func:`psum` function:
 
 .. sourcecode:: ipython
 
     In [1]: from IPython.kernel import client
     
     In [2]: mec = client.MultiEngineClient()
     
     In [3]: mec.activate()
     
     In [4]: px import numpy as np
     Parallel execution on engines: all
     Out[4]: 
     <Results List>
     [0] In [13]: import numpy as np
     [1] In [13]: import numpy as np
     [2] In [13]: import numpy as np
     [3] In [13]: import numpy as np
     
     In [6]: px a = np.random.rand(100)
     Parallel execution on engines: all
     Out[6]: 
     <Results List>
     [0] In [15]: a = np.random.rand(100)
     [1] In [15]: a = np.random.rand(100)
     [2] In [15]: a = np.random.rand(100)
     [3] In [15]: a = np.random.rand(100)
     
     In [7]: px from psum import psum
     Parallel execution on engines: all
     Out[7]: 
     <Results List>
     [0] In [16]: from psum import psum
     [1] In [16]: from psum import psum
     [2] In [16]: from psum import psum
     [3] In [16]: from psum import psum
     
     In [8]: px s = psum(a)
     Parallel execution on engines: all
     Out[8]: 
     <Results List>
     [0] In [17]: s = psum(a)
     [1] In [17]: s = psum(a)
     [2] In [17]: s = psum(a)
     [3] In [17]: s = psum(a)
     
     In [9]: px print s
     Parallel execution on engines: all
     Out[9]: 
     <Results List>
     [0] In [18]: print s
     [0] Out[18]: 187.451545803
     
     [1] In [18]: print s
     [1] Out[18]: 187.451545803
     
     [2] In [18]: print s
     [2] Out[18]: 187.451545803
 
     [3] In [18]: print s
     [3] Out[18]: 187.451545803
 
 Any Python code that makes calls to MPI can be used in this manner, including
 compiled C, C++ and Fortran libraries that have been exposed to Python.
 
 .. [MPI] Message Passing Interface.  http://www-unix.mcs.anl.gov/mpi/
 .. [mpi4py] MPI for Python. mpi4py: http://mpi4py.scipy.org/
 .. [OpenMPI] Open MPI. http://www.open-mpi.org/
 .. [PyTrilinos] PyTrilinos. http://trilinos.sandia.gov/packages/pytrilinos/
  No newline at end of file

docs/source/parallel/parallel_process.txt

@@ -1,324 +1,336 b''
 .. _parallel_process:
 
 ===========================================
 Starting the IPython controller and engines
 ===========================================
 
 To use IPython for parallel computing, you need to start one instance of
 the controller and one or more instances of the engine. The controller
 and each engine can run on different machines or on the same machine.
 Because of this, there are many different possibilities.
 
 Broadly speaking, there are two ways of going about starting a controller and engines:
 
 * In an automated manner using the :command:`ipcluster` command.
 * In a more manual way using the :command:`ipcontroller` and 
   :command:`ipengine` commands.
 
 This document describes both of these methods.  We recommend that new users start with the :command:`ipcluster` command as it simplifies many common usage cases.
 
 General considerations
 ======================
 
 Before delving into the details about how you can start a controller and engines using the various methods, we outline some of the general issues that come up when starting the controller and engines.  These things come up no matter which method you use to start your IPython cluster.
 
 Let's say that you want to start the controller on ``host0`` and engines on hosts ``host1``-``hostn``.  The following steps are then required:
 
 1. Start the controller on ``host0`` by running :command:`ipcontroller` on
    ``host0``.
 2. Move the FURL file (:file:`ipcontroller-engine.furl`) created by the 
    controller from ``host0`` to hosts ``host1``-``hostn``.
 3. Start the engines on hosts ``host1``-``hostn`` by running 
    :command:`ipengine`.  This command has to be told where the FURL file 
    (:file:`ipcontroller-engine.furl`) is located.
 
 At this point, the controller and engines will be connected. By default, the
 FURL files created by the controller are put into the
 :file:`~/.ipython/security` directory. If the engines share a filesystem with
 the controller, step 2 can be skipped as the engines will automatically look
 at that location.
 
 The final step required required to actually use the running controller from a
 client is to move the FURL files :file:`ipcontroller-mec.furl` and
 :file:`ipcontroller-tc.furl` from ``host0`` to the host where the clients will
 be run.  If these file are put into the :file:`~/.ipython/security` directory of the client's host, they will be found automatically.  Otherwise, the full path to them has to be passed to the client's constructor.
 
 Using :command:`ipcluster`
 ==========================
 
 The :command:`ipcluster` command provides a simple way of starting a controller and engines in the following situations:
 
 1. When the controller and engines are all run on localhost. This is useful
    for testing or running on a multicore computer.
 2. When engines are started using the :command:`mpirun` command that comes 
    with most MPI [MPI]_ implementations
 3. When engines are started using the PBS [PBS]_ batch system.
 4. When the controller is started on localhost and the engines are started on 
    remote nodes using :command:`ssh`.
 
 .. note::
 
     It is also possible for advanced users to add support to
     :command:`ipcluster` for starting controllers and engines using other
     methods (like Sun's Grid Engine for example).
 
 .. note::
 
     Currently :command:`ipcluster` requires that the
     :file:`~/.ipython/security` directory live on a shared filesystem that is
     seen by both the controller and engines. If you don't have a shared file
     system you will need to use :command:`ipcontroller` and
     :command:`ipengine` directly.  This constraint can be relaxed if you are 
     using the :command:`ssh` method to start the cluster.
 
 Underneath the hood, :command:`ipcluster` just uses :command:`ipcontroller`
 and :command:`ipengine` to perform the steps described above.
 
 Using :command:`ipcluster` in local mode
 ----------------------------------------
 
 To start one controller and 4 engines on localhost, just do::
 
     $ ipcluster local -n 4
 
 To see other command line options for the local mode, do::
 
     $ ipcluster local -h
 
-Using :command:`ipcluster` in mpirun mode
------------------------------------------
+Using :command:`ipcluster` in mpiexec/mpirun mode
+-------------------------------------------------
 
-The mpirun mode is useful if you:
+The mpiexec/mpirun mode is useful if you:
 
 1. Have MPI installed.
-2. Your systems are configured to use the :command:`mpirun` command to start 
-   processes.
+2. Your systems are configured to use the :command:`mpiexec` or
+   :command:`mpirun` commands to start MPI processes.
+
+.. note::
+
+    The preferred command to use is :command:`mpiexec`.  However, we also 
+    support :command:`mpirun` for backwards compatibility.  The underlying
+    logic used is exactly the same, the only difference being the name of the 
+    command line program that is called.
 
 If these are satisfied, you can start an IPython cluster using::
 
-    $ ipcluster mpirun -n 4
+    $ ipcluster mpiexec -n 4
 
 This does the following:
 
 1. Starts the IPython controller on current host.
-2. Uses :command:`mpirun` to start 4 engines.
+2. Uses :command:`mpiexec` to start 4 engines.
 
 On newer MPI implementations (such as OpenMPI), this will work even if you don't make any calls to MPI or call :func:`MPI_Init`.  However, older MPI implementations actually require each process to call :func:`MPI_Init` upon starting.  The easiest way of having this done is to install the mpi4py [mpi4py]_ package and then call ipcluster with the ``--mpi`` option::
 
-    $ ipcluster mpirun -n 4 --mpi=mpi4py
+    $ ipcluster mpiexec -n 4 --mpi=mpi4py
 
 Unfortunately, even this won't work for some MPI implementations.  If you are having problems with this, you will likely have to use a custom Python executable that itself calls :func:`MPI_Init` at the appropriate time.  Fortunately, mpi4py comes with such a custom Python executable that is easy to install and use.  However, this custom Python executable approach will not work with :command:`ipcluster` currently.
 
 Additional command line options for this mode can be found by doing::
 
-    $ ipcluster mpirun -h
+    $ ipcluster mpiexec -h
 
 More details on using MPI with IPython can be found :ref:`here <parallelmpi>`.
 
 
 Using :command:`ipcluster` in PBS mode
 --------------------------------------
 
 The PBS mode uses the Portable Batch System [PBS]_ to start the engines.  To use this mode, you first need to create a PBS script template that will be used to start the engines.  Here is a sample PBS script template:
 
 .. sourcecode:: bash
 
     #PBS -N ipython
     #PBS -j oe
     #PBS -l walltime=00:10:00
     #PBS -l nodes=${n/4}:ppn=4
     #PBS -q parallel
 
     cd $$PBS_O_WORKDIR
     export PATH=$$HOME/usr/local/bin
     export PYTHONPATH=$$HOME/usr/local/lib/python2.4/site-packages
     /usr/local/bin/mpiexec -n ${n} ipengine --logfile=$$PBS_O_WORKDIR/ipengine
 
 There are a few important points about this template:
 
 1. This template will be rendered at runtime using IPython's :mod:`Itpl` 
    template engine.
 
 2. Instead of putting in the actual number of engines, use the notation
    ``${n}`` to indicate the number of engines to be started. You can also uses
    expressions like ``${n/4}`` in the template to indicate the number of 
    nodes.
 
 3. Because ``$`` is a special character used by the template engine, you must  
    escape any ``$`` by using ``$$``.  This is important when referring to 
    environment variables in the template.
 
 4. Any options to :command:`ipengine` should be given in the batch script 
    template.
 
 5. Depending on the configuration of you system, you may have to set 
    environment variables in the script template.
 
 Once you have created such a script, save it with a name like :file:`pbs.template`.  Now you are ready to start your job::
 
     $ ipcluster pbs -n 128 --pbs-script=pbs.template
 
 Additional command line options for this mode can be found by doing::
 
     $ ipcluster pbs -h
 
 Using :command:`ipcluster` in SSH mode
 --------------------------------------
 
 The SSH mode uses :command:`ssh` to execute :command:`ipengine` on remote
 nodes and the :command:`ipcontroller` on localhost.
 
 When using using this mode it highly recommended that you have set up SSH keys and are using ssh-agent [SSH]_ for password-less logins.
 
 To use this mode you need a python file describing the cluster, here is an example of such a "clusterfile":
 
 .. sourcecode:: python
     
     send_furl = True
     engines = { 'host1.example.com' : 2,
                 'host2.example.com' : 5,
                 'host3.example.com' : 1,
                 'host4.example.com' : 8 }
 
 Since this is a regular python file usual python syntax applies. Things to note:
 
 * The `engines` dict, where the keys is the host we want to run engines on and
   the value is the number of engines to run on that host.
 * send_furl can either be `True` or `False`, if `True` it will copy over the
   furl needed for :command:`ipengine` to each host.
 
 The ``--clusterfile`` command line option lets you specify the file to use for
 the cluster definition. Once you have your cluster file and you can
 :command:`ssh` into the remote hosts with out an password you are ready to
 start your cluster like so:
 
 .. sourcecode:: bash 
 
    $ ipcluster ssh --clusterfile /path/to/my/clusterfile.py
 
 
 Two helper shell scripts are used to start and stop :command:`ipengine` on remote hosts:
 
 * sshx.sh 
 * engine_killer.sh
 
 Defaults for both of these are contained in the source code for :command:`ipcluster`. The default scripts are written to a local file in a tmep directory and then copied to a temp directory on the remote host and executed from there.  On most Unix, Linux and OS X systems this is /tmp.
 
 The default sshx.sh is the following:
 
 .. sourcecode:: bash
 
    #!/bin/sh
    "$@" &> /dev/null &
    echo $!
 
 If you want to use a custom sshx.sh script you need to use the ``--sshx``
 option and specify the file to use. Using a custom sshx.sh file could be
 helpful when you need to setup the environment on the remote host before
 executing :command:`ipengine`.
 
 For a detailed options list:
 
 .. sourcecode:: bash
 
    $ ipcluster ssh -h
 
 Current limitations of the SSH mode of :command:`ipcluster` are:
 
 * Untested on Windows. Would require a working :command:`ssh` on Windows.
   Also, we are using shell scripts to setup and execute commands on remote
   hosts.
 * :command:`ipcontroller` is started on localhost, with no option to start it
   on a remote node.
 
 Using the :command:`ipcontroller` and :command:`ipengine` commands
 ==================================================================
 
 It is also possible to use the :command:`ipcontroller` and :command:`ipengine` commands to start your controller and engines.  This approach gives you full control over all aspects of the startup process.
 
 Starting the controller and engine on your local machine
 --------------------------------------------------------
 
 To use :command:`ipcontroller` and :command:`ipengine` to start things on your
 local machine, do the following.
 
 First start the controller::
 
 	$ ipcontroller
 	
 Next, start however many instances of the engine you want using (repeatedly) the command::
 
 	$ ipengine
 
 The engines should start and automatically connect to the controller using the FURL files in :file:`~./ipython/security`. You are now ready to use the controller and engines from IPython.
 
 .. warning:: 
 	
 	The order of the above operations is very important.  You *must*
  	start the controller before the engines, since the engines connect 
 	to the controller as they get started.
 
 .. note::
 
     On some platforms (OS X), to put the controller and engine into the
     background you may need to give these commands in the form ``(ipcontroller
     &)`` and ``(ipengine &)`` (with the parentheses) for them to work
     properly.
 
 Starting the controller and engines on different hosts
 ------------------------------------------------------
 
 When the controller and engines are running on different hosts, things are
 slightly more complicated, but the underlying ideas are the same:
 
 1. Start the controller on a host using :command:`ipcontroller`.
 2. Copy :file:`ipcontroller-engine.furl` from :file:`~./ipython/security` on the controller's host to the host where the engines will run.
 3. Use :command:`ipengine` on the engine's hosts to start the engines.
 
 The only thing you have to be careful of is to tell :command:`ipengine` where the :file:`ipcontroller-engine.furl` file is located.  There are two ways you can do this:
 
 * Put :file:`ipcontroller-engine.furl` in the :file:`~./ipython/security`
   directory on the engine's host, where it will be found automatically.
 * Call :command:`ipengine` with the ``--furl-file=full_path_to_the_file``
   flag.
 
 The ``--furl-file`` flag works like this::
 
     $ ipengine --furl-file=/path/to/my/ipcontroller-engine.furl
 
 .. note::
     
     If the controller's and engine's hosts all have a shared file system  
     (:file:`~./ipython/security` is the same on all of them), then things
     will just work!
 
 Make FURL files persistent
 ---------------------------
 
 At fist glance it may seem that that managing the FURL files is a bit annoying.  Going back to the house and key analogy, copying the FURL around each time you start the controller is like having to make a new key every time you want to unlock the door and enter your house.  As with your house, you want to be able to create the key (or FURL file) once, and then simply use it at any point in the future.
 
 This is possible.  The only thing you have to do is decide what ports the controller will listen on for the engines and clients.  This is done as follows::
 
     $ ipcontroller -r --client-port=10101 --engine-port=10102
 
+These options also work with all of the various modes of 
+:command:`ipcluster`::
+
+    $ ipcluster local -n 2 -r --client-port=10101 --engine-port=10102
+
 Then, just copy the furl files over the first time and you are set.  You can start and stop the controller and engines any many times as you want in the future, just make sure to tell the controller to use the *same* ports.
 
 .. note::
 
     You may ask the question: what ports does the controller listen on if you
     don't tell is to use specific ones? The default is to use high random port
     numbers. We do this for two reasons: i) to increase security through
     obscurity and ii) to multiple controllers on a given host to start and
     automatically use different ports.
 
 Log files
 ---------
 
 All of the components of IPython have log files associated with them.
 These log files can be extremely useful in debugging problems with
 IPython and can be found in the directory :file:`~/.ipython/log`.  Sending
 the log files to us will often help us to debug any problems.
 
 
 .. [PBS] Portable Batch System.  http://www.openpbs.org/
 .. [SSH] SSH-Agent http://en.wikipedia.org/wiki/Ssh-agent