[GCS] global state query node info table from GCS. (#8498)

This commit is contained in:
Lingxuan Zuo
2020-05-28 16:39:13 +08:00
committed by GitHub
parent 675ccbc799
commit e594524ed3
27 changed files with 162 additions and 101 deletions
+13 -7
View File
@@ -11,9 +11,15 @@ import ray
import ray.ray_constants as ray_constants
import ray.test_utils
import ray.cluster_utils
from ray.test_utils import (relevant_errors, wait_for_condition,
wait_for_errors, wait_for_pid_to_exit,
generate_internal_config_map)
from ray.test_utils import (
relevant_errors,
wait_for_condition,
wait_for_errors,
wait_for_pid_to_exit,
generate_internal_config_map,
get_non_head_nodes,
get_other_nodes,
)
SIGKILL = signal.SIGKILL if sys.platform != "win32" else signal.SIGTERM
@@ -305,7 +311,7 @@ def test_actor_restart_on_node_failure(ray_start_cluster):
ray.get(actor.ready.remote())
results = [actor.increase.remote() for _ in range(100)]
# Kill actor node, while the above task is still being executed.
cluster.remove_node(cluster.list_all_nodes()[-1])
cluster.remove_node(get_non_head_nodes(cluster)[-1])
cluster.add_node(num_cpus=1, _internal_config=config)
cluster.wait_for_nodes()
# Check that none of the tasks failed and the actor is restarted.
@@ -821,7 +827,7 @@ def test_decorated_method(ray_start_regular):
@pytest.mark.parametrize(
"ray_start_cluster", [{
"num_cpus": 1,
"num_nodes": 2,
"num_nodes": 3,
}], indirect=True)
def test_ray_wait_dead_actor(ray_start_cluster):
"""Tests that methods completed by dead actors are returned as ready"""
@@ -857,8 +863,8 @@ def test_ray_wait_dead_actor(ray_start_cluster):
except ray.exceptions.RayActorError:
return True
# Kill a node.
cluster.remove_node(cluster.list_all_nodes()[-1])
# Kill a node that must not be driver node or head node.
cluster.remove_node(get_other_nodes(cluster, exclude_head=True)[-1])
# Repeatedly submit tasks and call ray.wait until the exception for the
# dead actor is received.
assert wait_for_condition(actor_dead)