def test_multiprocessing(self):
set_db_uri()
key1, key2, key3 = uuid4(), uuid4(), uuid4()
value1, value2, value3 = 11111, 22222, 33333
self.close_connections_before_forking()
pipeline_ids = [1, 2, 3]
runner = MultiprocessRunner(
system=self.system,
pipeline_ids=pipeline_ids,
infrastructure_class=self.infrastructure_class,
)
# Start running operating system processes.
with runner:
# Get local application object.
paxosapplication0 = runner.get(runner.system.process_classes["paxosapplication0"])
assert isinstance(paxosapplication0, PaxosApplication)
# Start proposing values on the different system pipelines.
paxosapplication0.change_pipeline(1)
started1 = datetime.datetime.now()
paxosapplication0.propose_value(key1, value1)
paxosapplication0.change_pipeline(2)
started2 = datetime.datetime.now()
paxosapplication0.propose_value(key2, value2)
paxosapplication0.change_pipeline(3)
started3 = datetime.datetime.now()
paxosapplication0.propose_value(key3, value3)
# Check all the process applications have expected final values.
paxosapplication1 = runner.get(runner.system.process_classes["paxosapplication1"])
paxosapplication2 = runner.get(runner.system.process_classes["paxosapplication2"])
assert isinstance(paxosapplication1, PaxosApplication)
paxosapplication0.repository.use_cache = False
paxosapplication1.repository.use_cache = False
paxosapplication2.repository.use_cache = False
self.assert_final_value(paxosapplication0, key1, value1)
self.assert_final_value(paxosapplication1, key1, value1)
self.assert_final_value(paxosapplication2, key1, value1)
duration1 = (datetime.datetime.now() - started1).total_seconds()
print("Resolved paxos 1 with multiprocessing in %ss" % duration1)
self.assert_final_value(paxosapplication0, key2, value2)
self.assert_final_value(paxosapplication1, key2, value2)
self.assert_final_value(paxosapplication2, key2, value2)
duration2 = (datetime.datetime.now() - started2).total_seconds()
print("Resolved paxos 2 with multiprocessing in %ss" % duration2)
self.assert_final_value(paxosapplication0, key3, value3)
self.assert_final_value(paxosapplication1, key3, value3)
self.assert_final_value(paxosapplication2, key3, value3)
duration3 = (datetime.datetime.now() - started3).total_seconds()
print("Resolved paxos 3 with multiprocessing in %ss" % duration3)
def test_ray_performance(self):
set_db_uri()
num_pipelines = 2
pipeline_ids = range(num_pipelines)
runner = RayRunner(
system=self.system,
pipeline_ids=pipeline_ids,
infrastructure_class=self.infrastructure_class,
setup_tables=True,
)
app_class: Type[PaxosApplication] = runner.get_class(
"paxosapplication0")
num_proposals_per_pipeline = 25
num_proposals = num_pipelines * num_proposals_per_pipeline
# Propose values.
proposals = list(((uuid4(), i) for i in range(num_proposals)))
with runner:
@retry((KeyError, AssertionError),
max_attempts=100,
wait=0.25,
stall=0)
def assert_final_value(pipeline_id, key, expected):
actual = runner.get(app_class,
pipeline_id).get_final_value(key)
self.assertEqual(actual, expected)
# Start timing (just for fun).
started = datetime.datetime.now()
for key, value in proposals:
pipeline_id = value % len(pipeline_ids)
print("Proposing key {} value {}".format(key, value))
runner.get(app_class,
pipeline_id).propose_value(key, str(value))
sleep(0.005)
# Check final values.
for key, value in proposals:
print("Asserting final value for key {} value {}".format(
key, value))
pipeline_id = value % len(pipeline_ids)
assert_final_value(pipeline_id, key, str(value))
# Print timing information (just for fun).
duration = (datetime.datetime.now() - started).total_seconds()
print("Resolved {} paxoses with ray in {:.4f}s "
"({:.1f} values/s, {:.4f}s each)".format(
num_proposals,
duration,
num_proposals / duration,
duration / num_proposals,
))
def setUp(self):
# Define system.
self.system = System(
Orders | Reservations | Orders | Payments | Orders,
infrastructure_class=self.infrastructure_class,
)
# Set environment.
set_db_uri()
def test_multiprocessing_performance(self):
set_db_uri()
num_pipelines = 2
pipeline_ids = range(num_pipelines)
runner = MultiprocessRunner(
system=self.system,
pipeline_ids=pipeline_ids,
infrastructure_class=self.infrastructure_class,
setup_tables=True,
)
num_proposals = 50
self.close_connections_before_forking()
with runner:
sleep(1)
# Construct an application instance in this process.
paxosapplication0 = runner.get(
runner.system.process_classes["paxosapplication0"])
assert isinstance(paxosapplication0, PaxosApplication)
# Don't use the cache, so as to keep checking actual database.
paxosapplication0.repository.use_cache = False
# Start timing (just for fun).
started = datetime.datetime.now()
# Propose values.
proposals = list(((uuid4(), i) for i in range(num_proposals)))
for key, value in proposals:
paxosapplication0.change_pipeline((value % len(pipeline_ids)))
print("Proposing key {} value {}".format(key, value))
paxosapplication0.propose_value(key, str(value))
sleep(0.0)
# Check final values.
for key, value in proposals:
print("Asserting final value for key {} value {}".format(
key, value))
self.assert_final_value(paxosapplication0, key, str(value))
# Print timing information (just for fun).
duration = (datetime.datetime.now() - started).total_seconds()
print(
"Resolved {} paxoses with multiprocessing in {:.4f}s ({:.4f}s "
"each)".format(num_proposals, duration,
duration / num_proposals))
def test_multi_threaded(self):
if "TRAVIS_PYTHON_VERSION" in os.environ:
if self.infrastructure_class is SQLAlchemyApplication:
self.skipTest(
"There's an intermittent problem with the multi-threaded"
"runner with SQLAlchemy and Python 3.7 on Travis. Fix me :)."
)
set_db_uri()
key1, key2, key3 = uuid4(), uuid4(), uuid4()
value1, value2, value3 = 11111, 22222, 33333
runner = MultiThreadedRunner(
system=self.system, infrastructure_class=self.infrastructure_class
)
with runner:
paxosapplication0 = runner.processes["paxosapplication0"]
paxosapplication1 = runner.processes["paxosapplication1"]
paxosapplication2 = runner.processes["paxosapplication2"]
started1 = datetime.datetime.now()
assert isinstance(paxosapplication0, PaxosApplication)
paxosapplication0.propose_value(key1, value1)
ended1 = (datetime.datetime.now() - started1).total_seconds()
# Check each process has expected final value.
self.assert_final_value(paxosapplication0, key1, value1)
self.assert_final_value(paxosapplication1, key1, value1)
self.assert_final_value(paxosapplication2, key1, value1)
print("Resolved paxos 1 with multi threads in %ss" % ended1)
started2 = datetime.datetime.now()
paxosapplication1.propose_value(key2, value2)
ended2 = (datetime.datetime.now() - started2).total_seconds()
# Check each process has a resolution.
self.assert_final_value(paxosapplication0, key2, value2)
self.assert_final_value(paxosapplication1, key2, value2)
self.assert_final_value(paxosapplication2, key2, value2)
print("Resolved paxos 2 with multi threads in %ss" % ended2)
started3 = datetime.datetime.now()
paxosapplication2.propose_value(key3, value3)
ended3 = (datetime.datetime.now() - started3).total_seconds()
# Check each process has a resolution.
self.assert_final_value(paxosapplication0, key3, value3)
self.assert_final_value(paxosapplication1, key3, value3)
self.assert_final_value(paxosapplication2, key3, value3)
print("Resolved paxos 3 with multi threads in %ss" % ended3)
def set_db_uri(self):
set_db_uri()