chore: import upstream snapshot with attribution

This commit is contained in:
wehub-resource-sync
2026-07-13 13:17:40 +08:00
commit f1825c8ceb
10096 changed files with 2364182 additions and 0 deletions
@@ -0,0 +1,378 @@
import logging
import sys
import time
import traceback
import pytest
import requests
import ray
from ray._private.test_utils import format_web_url, wait_until_server_available
from ray.dashboard.modules.node import actor_consts
from ray.dashboard.tests.conftest import * # noqa
from ray.util.placement_group import placement_group
from ray.util.scheduling_strategies import PlacementGroupSchedulingStrategy
logger = logging.getLogger(__name__)
def test_actors(disable_aiohttp_cache, ray_start_with_dashboard):
"""
Tests the REST API dashboard calls on:
- alive actors
- infeasible actors
- dead actors
- pg actors (with pg_id set and required_resources formatted)
"""
@ray.remote
class Foo:
def __init__(self, num):
self.num = num
def do_task(self):
return self.num
def get_node_id(self):
return ray.get_runtime_context().get_node_id()
def get_pid(self):
import os
return os.getpid()
def __repr__(self) -> str:
return "Foo1"
@ray.remote(num_cpus=0, resources={"infeasible_actor": 1})
class InfeasibleActor:
pass
pg = placement_group([{"CPU": 1}])
@ray.remote(num_cpus=1)
class PgActor:
def __init__(self):
pass
def do_task(self):
return 1
def get_placement_group_id(self):
return ray.get_runtime_context().get_placement_group_id()
foo_actors = [Foo.options(name="first").remote(4), Foo.remote(5)]
infeasible_actor = InfeasibleActor.options(name="infeasible").remote() # noqa
dead_actor = Foo.options(name="dead").remote(1)
pg_actor = PgActor.options(
name="pg",
scheduling_strategy=PlacementGroupSchedulingStrategy(
placement_group=pg,
),
).remote()
ray.kill(dead_actor)
[actor.do_task.remote() for actor in foo_actors]
pg_actor.do_task.remote()
webui_url = ray_start_with_dashboard["webui_url"]
assert wait_until_server_available(webui_url)
webui_url = format_web_url(webui_url)
job_id = ray.get_runtime_context().get_job_id()
node_id = ray.get(foo_actors[0].get_node_id.remote())
pid = ray.get(foo_actors[0].get_pid.remote())
placement_group_id = ray.get(pg_actor.get_placement_group_id.remote())
timeout_seconds = 5
start_time = time.time()
last_ex = None
while True:
time.sleep(1)
try:
resp = requests.get(f"{webui_url}/logical/actors")
resp_json = resp.json()
resp_data = resp_json["data"]
actors = resp_data["actors"]
assert len(actors) == 5
for a in actors.values():
if a["name"] == "first":
actor_response = a
assert actor_response["jobId"] == job_id
assert "Foo" in actor_response["className"]
assert "address" in actor_response
assert type(actor_response["address"]) is dict
assert actor_response["address"]["nodeId"] == node_id
assert actor_response["state"] == "ALIVE"
assert actor_response["name"] == "first"
assert actor_response["numRestarts"] == "0"
assert actor_response["pid"] == pid
assert actor_response["startTime"] > 0
assert actor_response["requiredResources"] == {}
assert actor_response["endTime"] == 0
assert actor_response["exitDetail"] == "-"
assert actor_response["reprName"] == "Foo1"
for a in actors.values():
# "exitDetail always exits from the response"
assert "exitDetail" in a
# Check the dead actor metadata.
for a in actors.values():
if a["name"] == "dead":
dead_actor_response = a
assert dead_actor_response["endTime"] > 0
assert "ray.kill" in dead_actor_response["exitDetail"]
assert dead_actor_response["state"] == "DEAD"
assert dead_actor_response["name"] == "dead"
# Check the infeasible actor metadata.
for a in actors.values():
if a["name"] == "infeasible":
infeasible_actor_response = a
# Make sure the infeasible actor's resource name is correct.
assert infeasible_actor_response["requiredResources"] == {
"infeasible_actor": 1
}
assert infeasible_actor_response["pid"] == 0
all_pids = {entry["pid"] for entry in actors.values()}
assert 0 in all_pids # The infeasible actor
assert len(all_pids) > 1
# Check the pg actor metadata.
for a in actors.values():
if a["name"] == "pg":
pg_actor_response = a
assert pg_actor_response["placementGroupId"] == placement_group_id
assert pg_actor_response["requiredResources"] == {"CPU": 1.0}
break
except Exception as ex:
last_ex = ex
finally:
if time.time() > start_time + timeout_seconds:
ex_stack = (
traceback.format_exception(
type(last_ex), last_ex, last_ex.__traceback__
)
if last_ex
else []
)
ex_stack = "".join(ex_stack)
raise Exception(f"Timed out while testing, {ex_stack}")
def test_actor_with_ids(disable_aiohttp_cache, ray_start_with_dashboard):
"""
Tests the REST API dashboard calls with actor ids in the URL
"""
@ray.remote
class Actor:
def __init__(self, num):
self.num = num
def do_task(self):
return self.num
def get_actor_id(self):
return ray.get_runtime_context().get_actor_id()
webui_url = ray_start_with_dashboard["webui_url"]
assert wait_until_server_available(webui_url)
webui_url = format_web_url(webui_url)
actors = [Actor.options(name=f"Actor{i}").remote(i) for i in range(5)]
for actor in actors:
actor.do_task.remote()
actor_ids = [ray.get(actor.get_actor_id.remote()) for actor in actors]
timeout_seconds = 5
start_time = time.time()
last_ex = None
while True:
time.sleep(1)
try:
actor_idx = 2
resp = requests.get(f"{webui_url}/logical/actors/{actor_ids[actor_idx]}")
resp_json = resp.json()
resp_data = resp_json["data"]
actor_detail = resp_data["detail"]
assert actor_detail["state"] == "ALIVE"
assert actor_detail["name"] == f"Actor{actor_idx}"
assert actor_detail["numRestarts"] == "0"
assert actor_detail["startTime"] > 0
assert actor_detail["requiredResources"] == {}
assert actor_detail["endTime"] == 0
assert actor_detail["exitDetail"] == "-"
actor_idxs = [0, 1, 4]
actor_idxs_to_id_str = ",".join([str(actor_ids[i]) for i in actor_idxs])
resp = requests.get(
f"{webui_url}/logical/actors?ids={actor_idxs_to_id_str}"
)
resp_json = resp.json()
resp_actors = resp_json["data"]["actors"]
assert len(resp_actors) == len(actor_idxs)
for actor_idx in actor_idxs:
actor_detail = resp_actors[str(actor_ids[actor_idx])]
assert actor_detail["state"] == "ALIVE"
assert actor_detail["name"] == f"Actor{actor_idx}"
assert actor_detail["numRestarts"] == "0"
assert actor_detail["startTime"] > 0
assert actor_detail["requiredResources"] == {}
assert actor_detail["endTime"] == 0
assert actor_detail["exitDetail"] == "-"
break
except Exception as ex:
last_ex = ex
finally:
if time.time() > start_time + timeout_seconds:
ex_stack = (
traceback.format_exception(
type(last_ex), last_ex, last_ex.__traceback__
)
if last_ex
else []
)
ex_stack = "".join(ex_stack)
raise Exception(f"Timed out while testing, {ex_stack}")
def test_nil_node(enable_test_module, disable_aiohttp_cache, ray_start_with_dashboard):
assert wait_until_server_available(ray_start_with_dashboard["webui_url"]) is True
webui_url = ray_start_with_dashboard["webui_url"]
assert wait_until_server_available(webui_url)
webui_url = format_web_url(webui_url)
@ray.remote(num_gpus=1)
class InfeasibleActor:
pass
infeasible_actor = InfeasibleActor.remote() # noqa
timeout_seconds = 5
start_time = time.time()
last_ex = None
while True:
time.sleep(1)
try:
resp = requests.get(f"{webui_url}/logical/actors")
resp_json = resp.json()
resp_data = resp_json["data"]
actors = resp_data["actors"]
assert len(actors) == 1
response = requests.get(webui_url + "/test/dump?key=node_actors")
response.raise_for_status()
result = response.json()
assert actor_consts.NIL_NODE_ID not in result["data"]["nodeActors"]
break
except Exception as ex:
last_ex = ex
finally:
if time.time() > start_time + timeout_seconds:
ex_stack = (
traceback.format_exception(
type(last_ex), last_ex, last_ex.__traceback__
)
if last_ex
else []
)
ex_stack = "".join(ex_stack)
raise Exception(f"Timed out while testing, {ex_stack}")
def test_actor_cleanup(
disable_aiohttp_cache, reduce_actor_cache, ray_start_with_dashboard
):
@ray.remote
class Foo:
def __init__(self, num):
self.num = num
def do_task(self):
return self.num
@ray.remote(num_gpus=1)
class InfeasibleActor:
pass
infeasible_actor = InfeasibleActor.remote() # noqa
foo_actors = [
Foo.remote(1),
Foo.remote(2),
Foo.remote(3),
Foo.remote(4),
Foo.remote(5),
Foo.remote(6),
]
results = [actor.do_task.remote() for actor in foo_actors] # noqa
webui_url = ray_start_with_dashboard["webui_url"]
assert wait_until_server_available(webui_url)
webui_url = format_web_url(webui_url)
timeout_seconds = 8
start_time = time.time()
last_ex = None
while True:
time.sleep(1)
try:
resp = requests.get(f"{webui_url}/logical/actors")
resp_json = resp.json()
resp_data = resp_json["data"]
actors = resp_data["actors"]
# Although max cache is 3, there should be 7 actors
# because they are all still alive.
assert len(actors) == 7
break
except Exception as ex:
last_ex = ex
finally:
if time.time() > start_time + timeout_seconds:
ex_stack = (
traceback.format_exception(
type(last_ex), last_ex, last_ex.__traceback__
)
if last_ex
else []
)
ex_stack = "".join(ex_stack)
raise Exception(f"Timed out while testing, {ex_stack}")
# kill
ray.kill(infeasible_actor)
[ray.kill(foo_actor) for foo_actor in foo_actors]
# Wait 5 seconds for cleanup to finish
time.sleep(5)
# Check only three remaining in cache
start_time = time.time()
while True:
time.sleep(1)
try:
resp = requests.get(f"{webui_url}/logical/actors")
resp_json = resp.json()
resp_data = resp_json["data"]
actors = resp_data["actors"]
# Max cache is 3 so only 3 actors should be left.
assert len(actors) == 3
break
except Exception as ex:
last_ex = ex
finally:
if time.time() > start_time + timeout_seconds:
ex_stack = (
traceback.format_exception(
type(last_ex), last_ex, last_ex.__traceback__
)
if last_ex
else []
)
ex_stack = "".join(ex_stack)
raise Exception(f"Timed out while testing, {ex_stack}")
if __name__ == "__main__":
sys.exit(pytest.main(["-v", __file__]))
@@ -0,0 +1,416 @@
import logging
import os
import random
import sys
import threading
import time
import traceback
from datetime import datetime, timedelta
import pytest
import requests
import ray
from ray._common.test_utils import wait_for_condition
from ray._private.test_utils import (
format_web_url,
wait_until_server_available,
)
from ray.cluster_utils import Cluster
from ray.dashboard.consts import RAY_DASHBOARD_STATS_UPDATING_INTERVAL
from ray.dashboard.tests.conftest import * # noqa
logger = logging.getLogger(__name__)
def test_nodes_update(enable_test_module, ray_start_with_dashboard):
assert wait_until_server_available(ray_start_with_dashboard["webui_url"]) is True
webui_url = ray_start_with_dashboard["webui_url"]
webui_url = format_web_url(webui_url)
timeout_seconds = 10
start_time = time.time()
while True:
time.sleep(1)
try:
response = requests.get(webui_url + "/test/dump")
response.raise_for_status()
try:
dump_info = response.json()
except Exception as ex:
logger.info("failed response: %s", response.text)
raise ex
assert dump_info["result"] is True
dump_data = dump_info["data"]
assert len(dump_data["nodes"]) == 1
break
except (AssertionError, requests.exceptions.ConnectionError):
logger.exception("Retry")
finally:
if time.time() > start_time + timeout_seconds:
raise Exception("Timed out while testing.")
def test_node_info(disable_aiohttp_cache, ray_start_with_dashboard):
@ray.remote
class Actor:
def getpid(self):
print(f"actor pid={os.getpid()}")
return os.getpid()
actors = [Actor.remote(), Actor.remote()]
actor_pids = [actor.getpid.remote() for actor in actors]
actor_pids = set(ray.get(actor_pids))
assert wait_until_server_available(ray_start_with_dashboard["webui_url"]) is True
webui_url = ray_start_with_dashboard["webui_url"]
webui_url = format_web_url(webui_url)
node_id = ray_start_with_dashboard["node_id"]
# NOTE: Leaving sum buffer time for data to get refreshed
timeout_seconds = RAY_DASHBOARD_STATS_UPDATING_INTERVAL * 1.5
start_time = time.time()
last_ex = None
while True:
time.sleep(1)
try:
response = requests.get(webui_url + "/nodes?view=hostnamelist")
response.raise_for_status()
hostname_list = response.json()
assert hostname_list["result"] is True, hostname_list["msg"]
hostname_list = hostname_list["data"]["hostNameList"]
assert len(hostname_list) == 1
hostname = hostname_list[0]
response = requests.get(webui_url + f"/nodes/{node_id}")
response.raise_for_status()
detail = response.json()
assert detail["result"] is True, detail["msg"]
detail = detail["data"]["detail"]
assert detail["hostname"] == hostname
assert detail["raylet"]["state"] == "ALIVE"
assert detail["raylet"]["isHeadNode"] is True
assert "raylet" in detail["cmdline"][0]
assert len(detail["workers"]) >= 2
assert len(detail["actors"]) == 2, detail["actors"]
actor_worker_pids = set()
for worker in detail["workers"]:
if "ray::Actor" in worker["cmdline"][0]:
actor_worker_pids.add(worker["pid"])
assert actor_worker_pids == actor_pids
response = requests.get(webui_url + "/nodes?view=summary")
response.raise_for_status()
summary = response.json()
assert summary["result"] is True, summary["msg"]
assert len(summary["data"]["summary"]) == 1
summary = summary["data"]["summary"][0]
assert summary["hostname"] == hostname
assert summary["raylet"]["state"] == "ALIVE"
assert "raylet" in summary["cmdline"][0]
assert "workers" not in summary
assert "actors" not in summary
assert "objectStoreAvailableMemory" in summary["raylet"]
assert "objectStoreUsedMemory" in summary["raylet"]
break
except Exception as ex:
last_ex = ex
finally:
if time.time() > start_time + timeout_seconds:
ex_stack = (
traceback.format_exception(
type(last_ex), last_ex, last_ex.__traceback__
)
if last_ex
else []
)
ex_stack = "".join(ex_stack)
raise Exception(f"Timed out while testing, {ex_stack}")
@pytest.mark.parametrize(
"ray_start_cluster_head_with_env_vars",
[
{
"include_dashboard": True,
"env_vars": {
"RAY_maximum_gcs_dead_node_cached_count": "1",
},
"_system_config": {
"health_check_initial_delay_ms": 0,
"health_check_timeout_ms": 100,
"health_check_failure_threshold": 3,
"health_check_period_ms": 100,
},
}
],
indirect=True,
)
def test_dead_node_cache_contains_latest_dead_node_if_cache_overflows(
enable_test_module, disable_aiohttp_cache, ray_start_cluster_head_with_env_vars
):
cluster: Cluster = ray_start_cluster_head_with_env_vars
assert wait_until_server_available(
cluster.webui_url
), "Failed to connect to the Dashboard Server"
webui_url = format_web_url(cluster.webui_url)
def _compare_dead_node_set(expected_alive_nodes, expected_dead_nodes):
try:
response = requests.get(webui_url + "/nodes?view=summary")
response.raise_for_status()
summary = response.json()
assert summary["result"] is True, summary["msg"]
summary = summary["data"]["summary"]
dead_nodes = set()
alive_nodes = set()
for node_info in summary:
node_id = node_info["raylet"]["nodeId"]
response = requests.get(webui_url + f"/nodes/{node_id}")
response.raise_for_status()
if node_info["raylet"]["state"] == "DEAD":
dead_nodes.add(node_id)
if node_info["raylet"]["state"] == "ALIVE":
alive_nodes.add(node_id)
assert alive_nodes == expected_alive_nodes
assert dead_nodes == expected_dead_nodes
return True
except Exception as ex:
logger.info(ex)
return False
node_1 = cluster.add_node()
head_node_id = ray.get_runtime_context().get_node_id()
node_1_id = node_1.node_id
curr_alive_nodes = {head_node_id, node_1_id}
curr_dead_nodes = set()
wait_for_condition(
_compare_dead_node_set,
10,
expected_alive_nodes=curr_alive_nodes,
expected_dead_nodes=curr_dead_nodes,
)
node_2 = cluster.add_node()
node_2_id = node_2.node_id
curr_alive_nodes.add(node_2_id)
wait_for_condition(
_compare_dead_node_set,
10,
expected_alive_nodes=curr_alive_nodes,
expected_dead_nodes=curr_dead_nodes,
)
cluster.remove_node(node_1, allow_graceful=False)
curr_alive_nodes.remove(node_1_id)
curr_dead_nodes.add(node_1_id)
wait_for_condition(
_compare_dead_node_set,
10,
expected_alive_nodes=curr_alive_nodes,
expected_dead_nodes=curr_dead_nodes,
)
cluster.remove_node(node_2, allow_graceful=False)
curr_alive_nodes.remove(node_2_id)
curr_dead_nodes.remove(node_1_id)
curr_dead_nodes.add(node_2_id)
wait_for_condition(
_compare_dead_node_set,
10,
expected_alive_nodes=curr_alive_nodes,
expected_dead_nodes=curr_dead_nodes,
)
@pytest.mark.parametrize(
"ray_start_cluster_head",
[
{
"include_dashboard": True,
"_system_config": {
"health_check_initial_delay_ms": 0,
"health_check_timeout_ms": 100,
"health_check_failure_threshold": 3,
"health_check_period_ms": 100,
},
}
],
indirect=True,
)
def test_multi_nodes_info(
enable_test_module, disable_aiohttp_cache, ray_start_cluster_head
):
cluster: Cluster = ray_start_cluster_head
assert wait_until_server_available(cluster.webui_url) is True
webui_url = cluster.webui_url
webui_url = format_web_url(webui_url)
cluster.add_node()
cluster.add_node()
dead_node = cluster.add_node()
cluster.remove_node(dead_node, allow_graceful=False)
def _check_nodes():
try:
response = requests.get(webui_url + "/nodes?view=summary")
response.raise_for_status()
summary = response.json()
assert summary["result"] is True, summary["msg"]
summary = summary["data"]["summary"]
assert len(summary) == 4
for node_info in summary:
node_id = node_info["raylet"]["nodeId"]
response = requests.get(webui_url + f"/nodes/{node_id}")
response.raise_for_status()
detail = response.json()
assert detail["result"] is True, detail["msg"]
detail = detail["data"]["detail"]
if node_id != dead_node.node_id:
assert detail["raylet"]["state"] == "ALIVE"
else:
assert detail["raylet"]["state"] == "DEAD"
assert detail["raylet"].get("objectStoreAvailableMemory", 0) == 0
return True
except Exception as ex:
logger.info(ex)
return False
wait_for_condition(_check_nodes, timeout=15)
@pytest.mark.parametrize(
"ray_start_cluster_head", [{"include_dashboard": True}], indirect=True
)
def test_multi_node_churn(
enable_test_module, disable_aiohttp_cache, ray_start_cluster_head
):
cluster: Cluster = ray_start_cluster_head
assert wait_until_server_available(cluster.webui_url) is True
webui_url = format_web_url(cluster.webui_url)
success = True
def verify():
nonlocal success
while True:
try:
resp = requests.get(webui_url)
resp.raise_for_status()
resp = requests.get(webui_url + "/nodes?view=summary")
resp.raise_for_status()
summary = resp.json()
assert summary["result"] is True, summary["msg"]
assert summary["data"]["summary"]
time.sleep(1)
except Exception:
success = False
break
t = threading.Thread(target=verify, daemon=True)
t.start()
t_st = datetime.now()
duration = timedelta(seconds=60)
worker_nodes = []
while datetime.now() < t_st + duration:
time.sleep(5)
if len(worker_nodes) < 2:
worker_nodes.append(cluster.add_node())
continue
should_add_node = random.randint(0, 1)
if should_add_node:
worker_nodes.append(cluster.add_node())
else:
node_index = random.randrange(0, len(worker_nodes))
node_to_remove = worker_nodes.pop(node_index)
cluster.remove_node(node_to_remove)
assert success
@pytest.mark.skipif(
sys.platform == "win32", reason="setproctitle does not change psutil.cmdline"
)
def test_node_physical_stats(enable_test_module, shutdown_only):
"""
Tests NodeHead._update_node_physical_stats.
"""
addresses = ray.init(include_dashboard=True, num_cpus=6)
@ray.remote(num_cpus=1)
class Actor:
def getpid(self):
return os.getpid()
actors = [Actor.remote() for _ in range(6)]
actor_pids = ray.get([actor.getpid.remote() for actor in actors])
actor_pids = set(actor_pids)
webui_url = addresses["webui_url"]
assert wait_until_server_available(webui_url) is True
webui_url = format_web_url(webui_url)
def _check_workers():
try:
resp = requests.get(webui_url + "/test/dump?key=node_physical_stats")
resp.raise_for_status()
result = resp.json()
assert result["result"] is True
node_physical_stats = result["data"]["nodePhysicalStats"]
assert len(node_physical_stats) == 1
current_stats = node_physical_stats[addresses["node_id"]]
# Check Actor workers
current_actor_pids = set()
for worker in current_stats["workers"]:
if "ray::Actor" in worker["cmdline"][0]:
current_actor_pids.add(worker["pid"])
assert current_actor_pids == actor_pids
# Check raylet cmdline
assert "raylet" in current_stats["cmdline"][0]
return True
except Exception as ex:
logger.info(ex)
return False
wait_for_condition(_check_workers, timeout=10)
def test_worker_pids_reported(enable_test_module, ray_start_with_dashboard):
assert wait_until_server_available(ray_start_with_dashboard["webui_url"]) is True
webui_url = ray_start_with_dashboard["webui_url"]
webui_url = format_web_url(webui_url)
node_id = ray_start_with_dashboard["node_id"]
@ray.remote(runtime_env={"uv": {"packages": ["requests==2.32.5"]}})
class UvActor:
def get_pid(self):
return os.getpid()
uv_actor = UvActor.remote()
uv_actor_pid = ray.get(uv_actor.get_pid.remote())
driver_pid = os.getpid()
def _check_worker_pids():
try:
response = requests.get(webui_url + f"/nodes/{node_id}")
response.raise_for_status()
dump_info = response.json()
assert dump_info["result"] is True
detail = dump_info["data"]["detail"]
pids = [worker["pid"] for worker in detail["workers"]]
assert len(pids) >= 2 # might include idle worker
assert uv_actor_pid in pids
assert driver_pid in pids
return True
except Exception as ex:
logger.info(ex)
return False
wait_for_condition(_check_worker_pids, timeout=20)
if __name__ == "__main__":
sys.exit(pytest.main(["-v", __file__]))