.. _aem-python:

Adaptive Execution Model in Python
----------------------------------

The following example has ben generated by using the ``bbque-layapp`` command
mentioned in :ref:`aem-structure`:

.. code-block::

	#!/usr/bin/env python
	import sys
	import os
	script_path = os.path.dirname(os.path.abspath(__file__))
	barbeque_path = (script_path + "/../../lib/bbque/bindings/python" +
			str(sys.version_info[0]) + "." + str(sys.version_info[1]))
	sys.path.append(barbeque_path)
	import barbeque

	class PySample(barbeque.BbqueEXC):

	    def __init__(self, name, recipe, rtlib_services):
		super(PySample, self).__init__(name, recipe, rtlib_services)
		self.logger.Info("EXC Unique Identifier (UID): {}".format(self.GetUniqueID()))

	    def onSetup(self):
		self.logger.Notice("PySample.onSetup()")

		# Initialization code...

		return barbeque.RTLIB_ExitCode.RTLIB_OK

	    def onConfigure(self, awm_id):
		self.logger.Notice("PySample.onConfigure(): EXC [{}], AWM[{:02d}]"
				   .format(self.exc_name, awm_id))

		# Information about assigned resources
		res_w = barbeque.RTLIB_Resources_Amount_Wrapper()
		ret = self.GetAssignedResources(barbeque.RTLIB_ResourceType.PROC_ELEMENT, res_w)
		cpu_time = res_w.amount
		ret = self.GetAssignedResources(barbeque.RTLIB_ResourceType.PROC_NR, res_w)
		cpu_nr = res_w.amount
		ret = self.GetAssignedResources(barbeque.RTLIB_ResourceType.GPU, res_w)
		gpu_nr = res_w.amount
		ret = self.GetAssignedResources(barbeque.RTLIB_ResourceType.MEMORY, res_w)
		mem = res_w.amount

		self.logger.Notice("PySample.onConfigure(): [{}], AWM[{:02d}] => "
				   "cpu_time={:3d}, nr_cpus={:2d}, nr_gpus={:2d}, mem={:3d}"
				   .format(self.exc_name, awm_id, cpu_time, cpu_nr, gpu_nr, mem))

		# Configure the application w.r.t to the assigned resources
		# ...

		return barbeque.RTLIB_ExitCode.RTLIB_OK

	    def onRun(self):
		# A computation run...
		self.logger.Notice("PySample.onRun(): EXC [{}], current cycle={:d}"
				    .format(self.exc_name, self.Cycles()))

		# ...just run 30 cycles end then exit
		if self.Cycles() == 30:
		    return barbeque.RTLIB_ExitCode.RTLIB_EXC_WORKLOAD_NONE

		return barbeque.RTLIB_ExitCode.RTLIB_OK

	    def onMonitor(self):
		wmp = self.WorkingModeParams()

		# Anything to check? Performance?

		self.logger.Notice("PySample.onMonitor(): EXC [{}], AWM[{:02d}] => CPS={:.2f}"
				    .format(self.exc_name, wmp.awm_id, self.GetCPS()))
		return barbeque.RTLIB_ExitCode.RTLIB_OK

	    def onRelease(self):
		self.logger.Notice("PySample.onRelease(): terminating...")

		# Cleanup code...

		return barbeque.RTLIB_ExitCode.RTLIB_OK


	def main():

	    # Wrapper for RTLIB_Services
	    services_wrapper = barbeque.RTLIB_Services_Wrapper()

	    # Initialize rpc channel with the resource manager
	    error_check = barbeque.RTLIB_Init("PySample", services_wrapper)
	    if (error_check != barbeque.RTLIB_ExitCode.RTLIB_OK) or (not services_wrapper.services):
		print ("RTLIB initialization failed")
		return

	    # Run
	    exc = PySample("PySample", "PySample", services_wrapper.services)
	    exc.Start()
	    exc.WaitCompletion()

	if __name__ == '__main__':
	    main()


In the ``main`` function, we must initialize the RTLib as we do for the C++
version, by calling the ``barbeque.RTLIB_Init`` function, providing the
application name and a ``barbeque.RTLIB_Services_Wrapper`` object.

Then we can instantiate an object of the class derived from
``barbeque.BbqueEXC`` (in this case ``PySample``), by specifying the
application name, the recipe file name without the ``.recipe`` extension, and
reference to the  ``.services`` field of the
``barbeque.RTLIB_Services_Wrapper`` previously initialized.

At this point, we are ready to launch the actual execution by calling the
``Start()`` method and leaving the main thread waiting for the termination
(``WaitCompletion()``).

The derived class, implementing the Adaptive Execution Model, matches the same
semantic already described for the C++, we the main difference represented by
the Python syntax.

Finally, similarly to the C++ case, in order to properly install our
application under the BOSP installation path, we need to
[[docs:bosp:config|Configure BOSP]] to enable the application
building/installation under the *Sample applications* menu and launch the
build command (``make`` or ``make samples`` if the source code has been
deployed under ``bosp/samples/python``). 

By default, the application is installed under ``<BOSP_PATH>/usr/bin``.