def _python_function_to_java(function):
"""Create the serializable Java object for a Python function."""
wrapped_func = PythonFunctionWrapper(function)
if running_mode == 'local':
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.LOCAL)
else:
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.HADOOP)
return wrapped_func
def _python_function_to_java(function):
"""Create the serializable Java object for a Python function."""
import inspect, py_compile
from org.python.core import BytecodeLoader
source = inspect.getsource(function)
function_name = function.func_name
print 'source:', source
print 'name:', function_name
exec(source)
exec('print ' + function_name + '("hi there")')
wrapped_func = PythonFunctionWrapper(function)
if config['running.mode'] == 'local':
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.LOCAL)
else:
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.HADOOP)
return wrapped_func
def _wrap_function(function, casc_function_type):
"""Wraps a Python function into a Serializable and callable Java object.
This wrapping is necessary as Cascading serializes the job pipeline before
it sends the job to the workers. We need to in essence reconstruct the
Python function from source on the receiving end when we deserialize the
function, as Python is an interpreted language.
Arguments:
function -- either a Cascading Operation, a PyCascading-decorated Python
function, or a native Python function
casc_function_type -- the Cascading Operation that this Python function
will be called by in its operate method
"""
if isinstance(function, cascading.operation.Operation):
return function
if isinstance(function, DecoratedFunction):
# Build the arguments for the constructor
args = []
decorators = function.decorators
if 'numargs_expected' in decorators:
args.append(decorators['numargs_expected'])
if 'produces' in decorators and decorators['produces']:
args.append(coerce_to_fields(decorators['produces']))
# Create the appropriate type (function or filter)
fw = casc_function_type(*args)
function = decorators['function']
fw.setConvertInputTuples(decorators['input_conversion'])
if decorators['type'] in set(['map', 'reduce']):
fw.setOutputMethod(decorators['output_method'])
fw.setOutputType(decorators['output_type'])
fw.setFlowProcessPassIn(decorators['flow_process_pass_in'])
fw.setContextArgs(decorators['args'])
fw.setContextKwArgs(decorators['kwargs'])
else:
# When function is a pure Python function, declared without decorators
fw = casc_function_type()
wrapped_func = PythonFunctionWrapper(function)
if running_mode == 'local':
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.LOCAL)
else:
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.HADOOP)
fw.setFunction(wrapped_func)
return fw
def _wrap_function(function, casc_function_type):
"""Wraps a Python function into a Serializable and callable Java object.
This wrapping is necessary as Cascading serializes the job pipeline before
it sends the job to the workers. We need to in essence reconstruct the
Python function from source on the receiving end when we deserialize the
function, as Python is an interpreted language.
Arguments:
function -- either a Cascading Operation, a PyCascading-decorated Python
function, or a native Python function
casc_function_type -- the Cascading Operation that this Python function
will be called by in its operate method
"""
if isinstance(function, cascading.operation.Operation):
return function
if isinstance(function, DecoratedFunction):
# Build the arguments for the constructor
args = []
decorators = function.decorators
if 'numargs_expected' in decorators:
args.append(decorators['numargs_expected'])
if 'produces' in decorators and decorators['produces']:
args.append(coerce_to_fields(decorators['produces']))
# Create the appropriate type (function or filter)
fw = casc_function_type(*args)
function = decorators['function']
fw.setConvertInputTuples(decorators['input_conversion'])
if decorators['type'] in set(['map', 'reduce']):
fw.setOutputMethod(decorators['output_method'])
fw.setOutputType(decorators['output_type'])
fw.setFlowProcessPassIn(decorators['flow_process_pass_in'])
fw.setContextArgs(decorators['args'])
fw.setContextKwArgs(decorators['kwargs'])
else:
# When function is a pure Python function, declared without decorators
fw = casc_function_type()
wrapped_func = PythonFunctionWrapper(function)
if running_mode == 'local':
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.LOCAL)
else:
wrapped_func.setRunningMode(PythonFunctionWrapper.RunningMode.HADOOP)
fw.setFunction(wrapped_func)
return fw
