wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-08-12 12:20:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

Use gcs server to replace raylet monitor when RAY_GCS_SERVICE_ENABLED=true (#7166)

Browse Source
This commit is contained in:
ZhuSenlin
2020-03-12 22:13:56 +08:00
committed by GitHub
parent 428fb79b27
commit b663bc6d67
17 changed files with 269 additions and 142 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/tests/test_multinode_failures_2.py
+17 -6
View File
@@ -132,7 +132,10 @@ def test_driver_lives_sequential(ray_start_regular):
ray.worker._global_node.kill_plasma_store()
ray.worker._global_node.kill_log_monitor()
ray.worker._global_node.kill_monitor()
ray.worker._global_node.kill_raylet_monitor()
if os.environ.get(ray_constants.RAY_GCS_SERVICE_ENABLED, None):
ray.worker._global_node.kill_gcs_server()
else:
ray.worker._global_node.kill_raylet_monitor()
# If the driver can reach the tearDown method, then it is still alive.
@@ -142,11 +145,19 @@ def test_driver_lives_sequential(ray_start_regular):
reason="Hanging with new GCS API.")
def test_driver_lives_parallel(ray_start_regular):
all_processes = ray.worker._global_node.all_processes
process_infos = (all_processes[ray_constants.PROCESS_TYPE_PLASMA_STORE] +
all_processes[ray_constants.PROCESS_TYPE_RAYLET] +
all_processes[ray_constants.PROCESS_TYPE_LOG_MONITOR] +
all_processes[ray_constants.PROCESS_TYPE_MONITOR] +
all_processes[ray_constants.PROCESS_TYPE_RAYLET_MONITOR])
if os.environ.get(ray_constants.RAY_GCS_SERVICE_ENABLED, None):
process_infos = (all_processes[ray_constants.PROCESS_TYPE_PLASMA_STORE]
+ all_processes[ray_constants.PROCESS_TYPE_GCS_SERVER]
+ all_processes[ray_constants.PROCESS_TYPE_RAYLET] +
all_processes[ray_constants.PROCESS_TYPE_LOG_MONITOR]
+ all_processes[ray_constants.PROCESS_TYPE_MONITOR])
else:
process_infos = (
all_processes[ray_constants.PROCESS_TYPE_PLASMA_STORE] +
all_processes[ray_constants.PROCESS_TYPE_RAYLET] +
all_processes[ray_constants.PROCESS_TYPE_LOG_MONITOR] +
all_processes[ray_constants.PROCESS_TYPE_MONITOR] +
all_processes[ray_constants.PROCESS_TYPE_RAYLET_MONITOR])
assert len(process_infos) == 5
# Kill all the components in parallel.