fix

python/ray/rllib/evaluation/worker_set.py

@@ -123,10 +123,10 @@ def foreach_worker_with_index(self, func):
         return local_result + remote_results
 
     @staticmethod
-    def _from_existing(local_worker, remote_workers):
+    def _from_existing(local_worker, remote_workers=None):
         workers = WorkerSet(None, None, {}, _setup=False)
         workers._local_worker = local_worker
-        workers._remote_workers = remote_workers
+        workers._remote_workers = remote_workers or []
         return workers
 
     def _make_worker(self, cls, env_creator, policy, worker_index, config):

python/ray/rllib/tests/test_external_multi_agent_env.py

@@ -11,6 +11,7 @@
 from ray.rllib.agents.pg.pg_policy import PGTFPolicy
 from ray.rllib.optimizers import SyncSamplesOptimizer
 from ray.rllib.evaluation.rollout_worker import RolloutWorker
+from ray.rllib.evaluation.worker_set import WorkerSet
 from ray.rllib.env.external_multi_agent_env import ExternalMultiAgentEnv
 from ray.rllib.tests.test_rollout_worker import MockPolicy
 from ray.rllib.tests.test_external_env import make_simple_serving
@@ -75,7 +76,7 @@ def testTrainExternalMultiCartpoleManyPolicies(self):
             policy=policies,
             policy_mapping_fn=lambda agent_id: random.choice(policy_ids),
             batch_steps=100)
-        optimizer = SyncSamplesOptimizer(ev, [])
+        optimizer = SyncSamplesOptimizer(WorkerSet._from_existing(ev), [])
         for i in range(100):
             optimizer.step()
             result = collect_metrics(ev)