Files
2026-07-13 13:17:40 +08:00

532 lines
15 KiB
Python

import sys
import threading
import time
from collections import defaultdict
from typing import Dict
import pytest
import ray
from ray._common.test_utils import (
PrometheusTimeseries,
SignalActor,
wait_for_condition,
)
from ray._private.state_api_test_utils import (
verify_failed_task,
)
from ray._private.test_utils import (
raw_metric_timeseries,
wait_for_aggregator_agent_if_enabled,
)
from ray._private.worker import RayContext
from ray.exceptions import RuntimeEnvSetupError
from ray.runtime_env import RuntimeEnv
from ray.util.state import list_tasks
# Run every test in this module twice: default and with core-worker to aggregator feature flag enabled
pytestmark = [
pytest.mark.parametrize(
"event_routing_config", ["default", "aggregator"], indirect=True
),
pytest.mark.usefixtures("event_routing_config"),
]
_SYSTEM_CONFIG = {
"task_events_report_interval_ms": 100,
"metrics_report_interval_ms": 200,
"enable_timeline": False,
"gcs_mark_task_failed_on_job_done_delay_ms": 1000,
}
def aggregate_task_event_metric(
info: RayContext, timeseries: PrometheusTimeseries
) -> Dict:
"""
Aggregate metrics of task events into:
{
"REPORTED": ray_gcs_task_manager_task_events_reported,
"STORED": ray_gcs_task_manager_task_events_stored,
"DROPPED_PROFILE_EVENT":
ray_gcs_task_manager_task_events_dropped PROFILE_EVENT,
"DROPPED_STATUS_EVENT":
ray_gcs_task_manager_task_events_dropped STATUS_EVENT,
}
"""
res = raw_metric_timeseries(info, timeseries)
task_events_info = defaultdict(int)
if "ray_gcs_task_manager_task_events_dropped" in res:
for sample in res["ray_gcs_task_manager_task_events_dropped"]:
if "Type" in sample.labels and sample.labels["Type"] != "":
task_events_info["DROPPED_" + sample.labels["Type"]] += sample.value
if "ray_gcs_task_manager_task_events_stored" in res:
for sample in res["ray_gcs_task_manager_task_events_stored"]:
task_events_info["STORED"] += sample.value
if "ray_gcs_task_manager_task_events_reported" in res:
for sample in res["ray_gcs_task_manager_task_events_reported"]:
task_events_info["REPORTED"] += sample.value
return task_events_info
def test_status_task_events_metrics(shutdown_only):
info = ray.init(num_cpus=1, _system_config=_SYSTEM_CONFIG)
# Start a task
@ray.remote
def f():
pass
for _ in range(10):
ray.get(f.remote())
timeseries = PrometheusTimeseries()
def verify():
metric = aggregate_task_event_metric(info, timeseries)
assert metric["REPORTED"] >= 10, (
"At least 10 tasks events should be reported. "
"Could be more than 10 with multiple flush."
)
assert metric["STORED"] == 11, "10 task + 1 driver's events should be stored."
return True
wait_for_condition(
verify,
timeout=20,
retry_interval_ms=100,
)
def test_failed_task_error(shutdown_only):
ray.init(_system_config=_SYSTEM_CONFIG)
# Test failed task with TASK_EXECUTION_EXCEPTION
error_msg_str = "fail is expected to fail"
@ray.remote
def fail(x=None):
if x is not None:
time.sleep(x)
raise ValueError(error_msg_str)
with pytest.raises(ray.exceptions.RayTaskError):
ray.get(fail.options(name="fail").remote())
wait_for_condition(
verify_failed_task,
name="fail",
error_type="TASK_EXECUTION_EXCEPTION",
error_message=error_msg_str,
)
# Test canceled tasks with TASK_CANCELLED
@ray.remote
def not_running():
raise ValueError("should not be run")
with pytest.raises(ray.exceptions.TaskCancelledError):
t = not_running.options(name="cancel-before-running").remote()
ray.cancel(t)
ray.get(t)
# Cancel task doesn't have additional error message
wait_for_condition(
verify_failed_task,
name="cancel-before-running",
error_type="TASK_CANCELLED",
error_message="",
)
# Test task failed when worker killed :WORKER_DIED
@ray.remote(max_retries=0)
def die():
exit(27)
with pytest.raises(ray.exceptions.WorkerCrashedError):
ray.get(die.options(name="die-worker").remote())
wait_for_condition(
verify_failed_task,
name="die-worker",
error_type="WORKER_DIED",
error_message="Worker exits with an exit code 27",
)
# Test actor task failed with actor dead: ACTOR_DIED
@ray.remote
class Actor:
def f(self):
time.sleep(999)
def ready(self):
pass
a = Actor.remote()
ray.get(a.ready.remote())
with pytest.raises(ray.exceptions.RayActorError):
ray.kill(a)
ray.get(a.f.options(name="actor-killed").remote())
wait_for_condition(
verify_failed_task,
name="actor-killed",
error_type="ACTOR_DIED",
error_message="The actor is dead because it was killed by `ray.kill`",
)
def test_failed_task_failed_due_to_node_failure(ray_start_cluster):
cluster = ray_start_cluster
head_node_id = cluster.add_node(num_cpus=1).node_id
ray.init(address=cluster.address)
node = cluster.add_node(num_cpus=2)
signal = SignalActor.options(
label_selector={"ray.io/node-id": head_node_id}
).remote()
@ray.remote(num_cpus=2, max_retries=0)
def sleep():
ray.get(signal.send.remote())
while True:
time.sleep(1)
obj_ref = sleep.options(name="node-killed").remote()
ray.get(signal.wait.remote())
cluster.remove_node(node)
with pytest.raises(ray.exceptions.NodeDiedError):
ray.get(obj_ref)
wait_for_condition(
verify_failed_task,
name="node-killed",
error_type="NODE_DIED",
error_message="Task failed because the node it was running on is dead or unavailable",
)
def test_failed_task_unschedulable(shutdown_only):
ray.init(num_cpus=1, _system_config=_SYSTEM_CONFIG)
node_id = ray.get_runtime_context().get_node_id()
policy = ray.util.scheduling_strategies.NodeAffinitySchedulingStrategy(
node_id=node_id,
soft=False,
)
@ray.remote
def task():
pass
task.options(
scheduling_strategy=policy,
name="task-unschedulable",
num_cpus=2,
).remote()
wait_for_condition(
verify_failed_task,
name="task-unschedulable",
error_type="TASK_UNSCHEDULABLE_ERROR",
error_message=(
"The node specified via NodeAffinitySchedulingStrategy"
" doesn't exist any more or is infeasible"
),
)
def test_failed_task_runtime_env_setup(shutdown_only):
@ray.remote
def f():
pass
bad_env = RuntimeEnv(
pip=["nonexistent-package"],
)
with pytest.raises(
RuntimeEnvSetupError,
):
ray.get(f.options(runtime_env=bad_env, name="task-runtime-env-failed").remote())
wait_for_condition(
verify_failed_task,
name="task-runtime-env-failed",
error_type="RUNTIME_ENV_SETUP_FAILED",
# Verify that the pip error content is propagated to task event.
error_message="nonexistent-package",
)
def test_parent_task_id_threaded_task(shutdown_only):
ray.init(_system_config=_SYSTEM_CONFIG)
# Task starts a thread
@ray.remote
def main_task():
def thd_task():
@ray.remote
def thd_task():
pass
ray.get(thd_task.remote())
thd = threading.Thread(target=thd_task)
thd.start()
thd.join()
ray.get(main_task.remote())
def verify():
tasks = list_tasks()
assert len(tasks) == 2
expect_parent_task_id = None
actual_parent_task_id = None
for task in tasks:
if task["name"] == "main_task":
expect_parent_task_id = task["task_id"]
elif task["name"] == "thd_task":
actual_parent_task_id = task["parent_task_id"]
assert actual_parent_task_id is not None
assert expect_parent_task_id == actual_parent_task_id
return True
wait_for_condition(verify)
def test_parent_task_id_non_concurrent_actor(shutdown_only):
ray.init(_system_config=_SYSTEM_CONFIG)
def run_task_in_thread():
def thd_task():
@ray.remote
def thd_task():
pass
ray.get(thd_task.remote())
thd = threading.Thread(target=thd_task)
thd.start()
thd.join()
@ray.remote
class Actor:
def main_task(self):
run_task_in_thread()
a = Actor.remote()
ray.get(a.main_task.remote())
def verify():
tasks = list_tasks()
expect_parent_task_id = None
actual_parent_task_id = None
for task in tasks:
if "main_task" in task["name"]:
expect_parent_task_id = task["task_id"]
elif "thd_task" in task["name"]:
actual_parent_task_id = task["parent_task_id"]
print(tasks)
assert actual_parent_task_id is not None
assert expect_parent_task_id == actual_parent_task_id
return True
wait_for_condition(verify)
@pytest.mark.parametrize("actor_concurrency", [3, 10])
def test_parent_task_id_concurrent_actor(shutdown_only, actor_concurrency):
# Test tasks runs in user started thread from actors have a parent_task_id
# as the actor's creation task.
ray.init(_system_config=_SYSTEM_CONFIG)
def run_task_in_thread(name, i):
def thd_task():
@ray.remote
def thd_task():
pass
ray.get(thd_task.options(name=f"{name}_{i}").remote())
thd = threading.Thread(target=thd_task)
thd.start()
thd.join()
@ray.remote
class AsyncActor:
async def main_task(self, i):
run_task_in_thread("async_thd_task", i)
@ray.remote
class ThreadedActor:
def main_task(self, i):
run_task_in_thread("threaded_thd_task", i)
def verify(actor_method_name, actor_class_name):
tasks = list_tasks()
print(tasks)
expect_parent_task_id = None
actual_parent_task_id = None
for task in tasks:
if f"{actor_class_name}.__init__" in task["name"]:
expect_parent_task_id = task["task_id"]
assert expect_parent_task_id is not None
for task in tasks:
if f"{actor_method_name}" in task["name"]:
actual_parent_task_id = task["parent_task_id"]
assert expect_parent_task_id == actual_parent_task_id, task
return True
async_actor = AsyncActor.options(max_concurrency=actor_concurrency).remote()
ray.get([async_actor.main_task.remote(i) for i in range(20)])
wait_for_condition(
verify, actor_class_name="AsyncActor", actor_method_name="async_thd_task"
)
thd_actor = ThreadedActor.options(max_concurrency=actor_concurrency).remote()
ray.get([thd_actor.main_task.remote(i) for i in range(20)])
wait_for_condition(
verify, actor_class_name="ThreadedActor", actor_method_name="threaded_thd_task"
)
def test_is_debugger_paused(shutdown_only):
ray_context = ray.init(num_cpus=1, _system_config=_SYSTEM_CONFIG)
address = ray_context.address_info["address"]
node_id = ray_context.address_info["node_id"]
wait_for_aggregator_agent_if_enabled(address, node_id)
@ray.remote(max_retries=0)
def f():
import time
# Pause 5 seconds inside debugger
with ray._private.worker.global_worker.task_paused_by_debugger():
time.sleep(5)
def verify(num_paused):
tasks = list_tasks(filters=[("is_debugger_paused", "=", "True")])
return len(tasks) == num_paused
f_task = f.remote() # noqa
wait_for_condition(
verify,
timeout=20,
retry_interval_ms=100,
num_paused=1,
)
wait_for_condition(
verify,
timeout=20,
retry_interval_ms=100,
num_paused=0,
)
@pytest.mark.parametrize("actor_concurrency", [1, 3])
def test_is_debugger_paused_actor(shutdown_only, actor_concurrency):
ray.init(_system_config=_SYSTEM_CONFIG)
@ray.remote
class TestActor:
def main_task(self, i):
if i == 0:
import time
# Pause 5 seconds inside debugger
with ray._private.worker.global_worker.task_paused_by_debugger():
time.sleep(5)
def verify(expected_task_name):
tasks = list_tasks(filters=[("is_debugger_paused", "=", "True")])
return len(tasks) == 1 and f"{expected_task_name}_0" in tasks[0]["name"]
test_actor = TestActor.options(max_concurrency=actor_concurrency).remote()
refs = [ # noqa
test_actor.main_task.options(name=f"TestActor.main_task_{i}").remote(i)
for i in range(20)
]
wait_for_condition(verify, expected_task_name="TestActor.main_task")
@pytest.mark.parametrize("actor_concurrency", [1, 3])
def test_is_debugger_paused_threaded_actor(shutdown_only, actor_concurrency):
ray.init(_system_config=_SYSTEM_CONFIG)
@ray.remote
class ThreadedActor:
def main_task(self, i):
def thd_task():
@ray.remote
def thd_task():
if i == 0:
import time
# Pause 5 seconds inside debugger
with ray._private.worker.global_worker.task_paused_by_debugger(): # noqa: E501
time.sleep(5)
ray.get(thd_task.options(name=f"ThreadedActor.main_task_{i}").remote())
thd = threading.Thread(target=thd_task)
thd.start()
thd.join()
def verify(expected_task_name):
tasks = list_tasks(filters=[("is_debugger_paused", "=", "True")])
return len(tasks) == 1 and f"{expected_task_name}_0" in tasks[0]["name"]
threaded_actor = ThreadedActor.options(max_concurrency=actor_concurrency).remote()
refs = [ # noqa
threaded_actor.main_task.options(name=f"ThreadedActor.main_task_{i}").remote(i)
for i in range(20)
]
wait_for_condition(verify, expected_task_name="ThreadedActor.main_task")
@pytest.mark.parametrize("actor_concurrency", [1, 3])
def test_is_debugger_paused_async_actor(shutdown_only, actor_concurrency):
ray.init(_system_config=_SYSTEM_CONFIG)
@ray.remote
class AsyncActor:
async def main_task(self, i):
if i == 0:
import time
# Pause 5 seconds inside debugger
print()
with ray._private.worker.global_worker.task_paused_by_debugger():
time.sleep(5)
def verify(expected_task_name):
tasks = list_tasks(filters=[("is_debugger_paused", "=", "True")])
print(tasks)
return len(tasks) == 1 and f"{expected_task_name}_0" in tasks[0]["name"]
async_actor = AsyncActor.options(max_concurrency=actor_concurrency).remote()
refs = [ # noqa
async_actor.main_task.options(name=f"AsyncActor.main_task_{i}").remote(i)
for i in range(20)
]
wait_for_condition(verify, expected_task_name="AsyncActor.main_task")
if __name__ == "__main__":
sys.exit(pytest.main(["-sv", __file__]))