Every user defined function or method must be synchronous (blocking). To define a function as asynchronous, use the async() decorator:
from concert.async import async @async def synchronous_function(): # long running operation return 1
Every asynchronous function returns a Future that can be used for explicit synchronization:
future = synchronous_function() print(future.done()) result = future.result()
Every future that is returned by Concert, has an additional method join that will block until execution finished and raise the exception that might have been raised in the wrapped function. It will also return the future to gather the result:
try: future = synchronous_function().join() result = future.result() except: print("synchronous_function raised an exception")
The asynchronous execution provided by Concert deals with concurrency. If the user wants to employ real parallelism they should make use of the multiprocessing module which provides functionality not limited by Python’s global interpreter lock.
When using the asynchronous getters and setters of Device and Parameter, processes can not be sure if other processes or the user manipulate the device during the execution. To lock devices or specific parameters, processes can use them as context managers:
with motor, pump['foo']: motor.position = 2 * q.mm pump.foo = 1 * q.s
Inside the with environment, the process has exclusive access to the devices and parameters.
Disable asynchronous execution¶
Testing and debugging asynchronous code can be difficult at times because the real source of an error is hidden behind calls from different places. To disable asynchronous execution (but still keeping the illusion of having Futures returned), you can import DISABLE_ASYNC and set it to True before importing anything else from Concert.
Concert already provides a Nose plugin that adds a --disable-async flag to the test runner which in turn sets DISABLE_ASYNC to True.