mirror of
https://github.com/ets-labs/python-dependency-injector.git
synced 2024-11-22 17:47:02 +03:00
52 lines
1.6 KiB
Python
52 lines
1.6 KiB
Python
"""`ThreadLocalSingleton` providers example."""
|
|
|
|
import threading
|
|
import Queue
|
|
|
|
import dependency_injector.providers as providers
|
|
|
|
|
|
def example(example_object, queue):
|
|
"""Example function that puts provided object in the provided queue."""
|
|
queue.put(example_object)
|
|
|
|
|
|
# Create thread-local singleton provider for some object (main thread):
|
|
thread_local_object = providers.ThreadLocalSingleton(object)
|
|
|
|
# Create singleton provider for thread-safe queue:
|
|
queue = providers.Singleton(Queue.Queue)
|
|
|
|
# Create callable provider for example(), inject dependencies:
|
|
example = providers.DelegatedCallable(example,
|
|
example_object=thread_local_object,
|
|
queue=queue)
|
|
|
|
# Create factory provider for threads that are targeted to execute example():
|
|
thread_factory = providers.Factory(threading.Thread,
|
|
target=example)
|
|
|
|
if __name__ == '__main__':
|
|
# Create 10 threads for concurrent execution of example():
|
|
threads = []
|
|
for thread_number in range(10):
|
|
threads.append(thread_factory(name='Thread{0}'.format(thread_number)))
|
|
|
|
# Start execution of all created threads:
|
|
for thread in threads:
|
|
thread.start()
|
|
|
|
# Wait while threads would complete their work:
|
|
for thread in threads:
|
|
thread.join()
|
|
|
|
# Making some asserts (main thread):
|
|
all_objects = set()
|
|
|
|
while not queue().empty():
|
|
all_objects.add(queue().get())
|
|
|
|
assert len(all_objects) == len(threads)
|
|
# Queue contains same number of objects as number of threads where
|
|
# thread-local singleton provider was used.
|