def test_cancel_waiting_workers(join_worker_pool):
"""
If we have a small pool and many workers, it is possible for workers to be enqueued
one after another in one thread.
We test that if we call `cancel()`, these enqueued workers are cancelled too.
"""
outcomes, worker = generate_workers([
(WorkerRule(timeout_min=1, timeout_max=1), 100),
],
seed=123)
factory = AllAtOnceFactory(list(outcomes))
pool = WorkerPool(worker,
factory,
target_successes=10,
timeout=10,
threadpool_size=10)
join_worker_pool(pool)
t_start = time.monotonic()
pool.start()
pool.block_until_target_successes()
pool.cancel()
pool.join()
t_end = time.monotonic()
# We have 10 threads in the pool and 100 workers that are all enqueued at once at the start.
# If we didn't check for the cancel condition, we would have to wait for 10 seconds.
# We get 10 successes after 1s and cancel the workers,
# but the next workers in each thread have already started, so we have to wait for another 1s.
assert t_end - t_start < 2.5
def test_join(join_worker_pool):
"""
Test joining the pool.
"""
outcomes, worker = generate_workers([
(WorkerRule(timeout_min=0.5, timeout_max=1.5), 9),
(WorkerRule(fails=True, timeout_min=0.5, timeout_max=1.5), 20),
],
seed=123)
factory = AllAtOnceFactory(list(outcomes))
pool = WorkerPool(worker,
factory,
target_successes=10,
timeout=1,
threadpool_size=30)
join_worker_pool(pool)
t_start = time.monotonic()
pool.start()
pool.join()
t_end = time.monotonic()
pool.join() # should work the second time too
# Even though timeout is 1, there are long-running workers which we can't interupt.
assert t_end - t_start < 3
def test_wait_for_successes_timed_out(join_worker_pool):
"""
Checks that if enough successful workers can't finish before the timeout, we get an exception.
"""
outcomes, worker = generate_workers([
(WorkerRule(timeout_min=0, timeout_max=0.5), 9),
(WorkerRule(timeout_min=1.5, timeout_max=2.5), 1),
(WorkerRule(fails=True, timeout_min=1.5, timeout_max=2.5), 20),
],
seed=123)
factory = AllAtOnceFactory(list(outcomes))
pool = WorkerPool(worker,
factory,
target_successes=10,
timeout=1,
threadpool_size=30)
join_worker_pool(pool)
t_start = time.monotonic()
pool.start()
with pytest.raises(WorkerPool.TimedOut):
successes = pool.block_until_target_successes()
t_end = time.monotonic()
# Even though timeout is 1, there are long-running workers which we can't interupt.
assert t_end - t_start < 3
def test_wait_for_successes_out_of_values(join_worker_pool):
"""
Checks that if there weren't enough successful workers, `block_until_target_successes()`
raises an exception when the value factory is exhausted.
"""
outcomes, worker = generate_workers([
(WorkerRule(timeout_min=0.5, timeout_max=1.5), 9),
(WorkerRule(fails=True, timeout_min=0.5, timeout_max=1.5), 20),
],
seed=123)
factory = AllAtOnceFactory(list(outcomes))
pool = WorkerPool(worker,
factory,
target_successes=10,
timeout=10,
threadpool_size=15)
join_worker_pool(pool)
t_start = time.monotonic()
pool.start()
with pytest.raises(WorkerPool.OutOfValues):
successes = pool.block_until_target_successes()
t_end = time.monotonic()
# We have roughly 2 workers per thread, so it shouldn't take longer than 1.5s (max timeout) * 2
assert t_end - t_start < 4
def test_wait_for_successes(join_worker_pool):
"""
Checks that `block_until_target_successes()` returns in time and gives all the successes,
if there were enough of them.
"""
outcomes, worker = generate_workers(
[
(WorkerRule(timeout_min=0.5, timeout_max=1.5), 10),
(WorkerRule(fails=True, timeout_min=1, timeout_max=3), 20),
],
seed=123)
factory = AllAtOnceFactory(list(outcomes))
pool = WorkerPool(worker, factory, target_successes=10, timeout=10, threadpool_size=30)
join_worker_pool(pool)
t_start = time.monotonic()
pool.start()
successes = pool.block_until_target_successes()
t_end = time.monotonic()
failures = pool.get_failures()
assert all(outcomes[value].fails for value in failures)
assert len(successes) == 10
# We have more threads in the pool than the workers,
# so all the successful ones should be able to finish right away.
assert t_end - t_start < 2
# Should be able to do it several times
successes = pool.block_until_target_successes()
assert len(successes) == 10
def __init__(self,
worker,
nodes,
percent_to_complete_before_release=5,
threadpool_size=120,
timeout=20):
self._total = len(nodes)
self._block_until_this_many_are_complete = math.ceil(len(nodes) * percent_to_complete_before_release / 100)
self._worker_pool = WorkerPool(worker=worker,
value_factory=AllAtOnceFactory(nodes),
target_successes=self._block_until_this_many_are_complete,
timeout=timeout,
stagger_timeout=0,
threadpool_size=threadpool_size)
def _enact_arrangements(
self,
network_middleware: RestMiddleware,
arrangements: Dict[Ursula, Arrangement],
publication_transaction: Optional[HexBytes] = None,
publish_treasure_map: bool = True,
timeout: int = 10,
):
"""
Attempts to distribute kfrags to Ursulas that accepted arrangements earlier.
"""
def worker(ursula_and_kfrag):
ursula, kfrag = ursula_and_kfrag
arrangement = arrangements[ursula]
# TODO: seems like it would be enough to just encrypt this with Ursula's public key,
# and not create a whole capsule.
# Can't change for now since it's node protocol.
payload = self._make_enactment_payload(publication_transaction,
kfrag)
message_kit, _signature = self.alice.encrypt_for(ursula, payload)
try:
# TODO: Concurrency
response = network_middleware.enact_policy(
ursula, arrangement.id, message_kit.to_bytes())
except network_middleware.UnexpectedResponse as e:
status = e.status
else:
status = response.status_code
return status
value_factory = AllAtOnceFactory(list(zip(arrangements, self.kfrags)))
worker_pool = WorkerPool(worker=worker,
value_factory=value_factory,
target_successes=self.n,
timeout=timeout,
threadpool_size=self.n)
worker_pool.start()
# Block until everything is complete. We need all the workers to finish.
worker_pool.join()
successes = worker_pool.get_successes()
if len(successes) != self.n:
raise Policy.EnactmentError()
# TODO: Enable re-tries?
statuses = {
ursula_and_kfrag[0].checksum_address: status
for ursula_and_kfrag, status in successes.items()
}
if not all(status == 200 for status in statuses.values()):
report = "\n".join(f"{address}: {status}"
for address, status in statuses.items())
self.log.debug(
f"Policy enactment failed. Request statuses:\n{report}")
# OK, let's check: if two or more Ursulas claimed we didn't pay,
# we need to re-evaulate our situation here.
number_of_claims_of_freeloading = sum(
status == 402 for status in statuses.values())
# TODO: a better exception here?
if number_of_claims_of_freeloading > 2:
raise self.alice.NotEnoughNodes
# otherwise just raise a more generic error
raise Policy.EnactmentError()