def test_shuffle_hang():
try:
shuffle.run(object_store_memory=1e9,
num_partitions=200,
partition_size=10e6)
finally:
ray.shutdown()
def test_shuffle_multi_node_no_streaming(ray_start_cluster):
cluster = ray_start_cluster
for _ in range(4):
cluster.add_node(num_cpus=2, object_store_memory=1e9)
shuffle.run(ray_address="auto",
num_partitions=200,
partition_size=10e6,
no_streaming=True)
def test_streaming_shuffle(set_kill_interval):
lineage_reconstruction_enabled, kill_interval, _ = set_kill_interval
try:
# Create our own tracker so that it gets scheduled onto the head node.
tracker = ShuffleStatusTracker.remote()
ray.get(tracker.get_progress.remote())
assert len(ray.nodes()) == 1, (
"Tracker actor may have been scheduled to remote node "
"and may get killed during the test")
shuffle.run(ray_address="auto",
no_streaming=False,
num_partitions=200,
partition_size=1e6,
tracker=tracker)
except (RayTaskError, ObjectLostError):
assert kill_interval is not None
def test_shuffle():
try:
shuffle.run()
finally:
ray.shutdown()
def test_shuffle_no_streaming():
try:
shuffle.run(no_streaming=True)
finally:
ray.shutdown()