wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-06-28 12:10:40 +08:00
Code Issues Packages Projects Releases Wiki Activity

Revert heartbeat interval to keep ci stable (#12836)

Browse Source 
* Revert heartbeat interval to keep ci stable

* fix missing one
This commit is contained in:
Tao Wang
2020-12-14 16:58:40 +08:00
committed by GitHub
parent 22c1968d62
commit 35f7d84dbe
7 changed files with 14 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/tests/test_actor_resources.py
+1 -1
View File
@@ -237,7 +237,7 @@ def test_actor_multiple_gpus_from_multiple_tasks(ray_start_cluster):
cluster.add_node(
num_cpus=10 * num_gpus_per_raylet,
num_gpus=num_gpus_per_raylet,
_system_config={"num_heartbeats_timeout": 100} if i == 0 else {})
_system_config={"num_heartbeats_timeout": 1000} if i == 0 else {})
ray.init(address=cluster.address)
@ray.remote
python/ray/tests/test_advanced_3.py
+3 -4
View File
@@ -610,10 +610,9 @@ def test_lease_request_leak(shutdown_only):
del obj_ref
ray.get(tasks)
def _no_objects():
return len(ray.objects()) == 0
wait_for_condition(_no_objects, timeout=10)
time.sleep(
1) # Sleep for an amount longer than the reconstruction timeout.
assert len(ray.objects()) == 0, ray.objects()
@pytest.mark.parametrize(
python/ray/tests/test_component_failures_2.py
+1 -1
View File
@@ -143,7 +143,7 @@ def check_components_alive(cluster, component_type, check_component_alive):
"num_cpus": 8,
"num_nodes": 4,
"_system_config": {
"num_heartbeats_timeout": 10
"num_heartbeats_timeout": 100
},
}],
indirect=True)
python/ray/tests/test_gcs_fault_tolerance.py
+1 -1
View File
@@ -76,7 +76,7 @@ def test_gcs_server_restart_during_actor_creation(ray_start_regular):
@pytest.mark.parametrize(
"ray_start_cluster_head", [
generate_system_config_map(
num_heartbeats_timeout=2, ping_gcs_rpc_server_max_retries=60)
num_heartbeats_timeout=20, ping_gcs_rpc_server_max_retries=60)
],
indirect=True)
def test_node_failure_detector_when_gcs_server_restart(ray_start_cluster_head):
python/ray/tests/test_multi_node_2.py
+3 -3
View File
@@ -34,7 +34,7 @@ def test_shutdown():
@pytest.mark.parametrize(
"ray_start_cluster_head", [
generate_system_config_map(
num_heartbeats_timeout=2, object_timeout_milliseconds=12345)
num_heartbeats_timeout=20, object_timeout_milliseconds=12345)
],
indirect=True)
def test_system_config(ray_start_cluster_head):
@@ -52,12 +52,12 @@ def test_system_config(ray_start_cluster_head):
@ray.remote
def f():
assert ray._config.object_timeout_milliseconds() == 12345
assert ray._config.num_heartbeats_timeout() == 2
assert ray._config.num_heartbeats_timeout() == 20
ray.get([f.remote() for _ in range(5)])
cluster.remove_node(worker, allow_graceful=False)
time.sleep(0.9)
time.sleep(1)
assert ray.cluster_resources()["CPU"] == 2
time.sleep(2)
python/ray/tests/test_placement_group.py
+3 -3
View File
@@ -1172,7 +1172,7 @@ ray.shutdown()
@pytest.mark.parametrize(
"ray_start_cluster_head", [
generate_system_config_map(
num_heartbeats_timeout=3, ping_gcs_rpc_server_max_retries=60)
num_heartbeats_timeout=20, ping_gcs_rpc_server_max_retries=60)
],
indirect=True)
def test_create_placement_group_after_gcs_server_restart(
@@ -1210,7 +1210,7 @@ def test_create_placement_group_after_gcs_server_restart(
@pytest.mark.parametrize(
"ray_start_cluster_head", [
generate_system_config_map(
num_heartbeats_timeout=3, ping_gcs_rpc_server_max_retries=60)
num_heartbeats_timeout=20, ping_gcs_rpc_server_max_retries=60)
],
indirect=True)
def test_create_actor_with_placement_group_after_gcs_server_restart(
@@ -1234,7 +1234,7 @@ def test_create_actor_with_placement_group_after_gcs_server_restart(
@pytest.mark.parametrize(
"ray_start_cluster_head", [
generate_system_config_map(
num_heartbeats_timeout=3, ping_gcs_rpc_server_max_retries=60)
num_heartbeats_timeout=20, ping_gcs_rpc_server_max_retries=60)
],
indirect=True)
def test_create_placement_group_during_gcs_server_restart(