[serve] Fix long running failure test (#11805)

This commit is contained in:
Edward Oakes
2020-11-09 11:21:03 -06:00
committed by GitHub
parent 407a212816
commit 2feba4409c
@@ -32,8 +32,8 @@ client = serve.start(detached=True)
@ray.remote
class RandomKiller:
def __init__(self, client, kill_period_s=1):
self.client = client
def __init__(self, kill_period_s=1):
self.client = serve.connect()
self.kill_period_s = kill_period_s
def _get_all_serve_actors(self):
@@ -55,8 +55,8 @@ class RandomKiller:
class RandomTest:
def __init__(self, client, max_endpoints=1):
self.client = client
def __init__(self, serve_client, max_endpoints=1):
self.client = serve_client
self.max_endpoints = max_endpoints
self.weighted_actions = [
(self.create_endpoint, 1),
@@ -114,8 +114,8 @@ class RandomTest:
iteration += 1
random_killer = RandomKiller.remote(client)
random_killer = RandomKiller.remote()
random_killer.run.remote()
# Subtract 4 from the CPUs available for master, router, HTTP proxy,
# and metric monitor actors.
RandomTest(client, max_endpoints=(num_nodes * cpus_per_node) - 4).run()
# Subtract 1 CPU for the controller and 1 CPU from each node for the HTTP
# server.
RandomTest(client, max_endpoints=(num_nodes * (cpus_per_node - 1)) - 1).run()