[serve] Fix long running failure test (#8863)

This commit is contained in:
Edward Oakes 2020-06-09 12:13:20 -05:00 committed by GitHub
parent 3d473600a8
commit f8d20d25bd
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23

View file

@ -6,7 +6,6 @@ import requests
import ray import ray
from ray import serve from ray import serve
from ray.serve.utils import retry_actor_failures
from ray.cluster_utils import Cluster from ray.cluster_utils import Cluster
num_redis_shards = 1 num_redis_shards = 1
@ -42,11 +41,10 @@ class RandomKiller:
def _get_all_serve_actors(self): def _get_all_serve_actors(self):
master = serve.api._get_master_actor() master = serve.api._get_master_actor()
[router] = retry_actor_failures(master.get_router) [router] = ray.get(master.get_router.remote())
[http_proxy] = retry_actor_failures(master.get_http_proxy) [http_proxy] = ray.get(master.get_http_proxy.remote())
all_handles = [master, router, http_proxy] all_handles = [master, router, http_proxy]
worker_handle_dict = retry_actor_failures( worker_handle_dict = ray.get(master.get_all_worker_handles.remote())
master.get_all_worker_handles)
for _, replica_dict in worker_handle_dict.items(): for _, replica_dict in worker_handle_dict.items():
all_handles.extend(list(replica_dict.values())) all_handles.extend(list(replica_dict.values()))