[core] Fix get actor timeout multiplier (#54525)

Signed-off-by: Varun Bhandary <varun.bhandary@databricks.com>
Signed-off-by: DataErrata <varun.bhandary@gmail.com>
Co-authored-by: Varun Bhandary <varun.bhandary@databricks.com>
Co-authored-by: Dhyey Shah <dhyey2019@gmail.com>

Author: 3 people

python/ray/tests/test_state_api.py

@@ -111,6 +111,7 @@
     WorkerState,
     StateSchema,
     state_column,
+    GetApiOptions,
 )
 from ray.dashboard.utils import ray_address_to_api_server_url
 from ray.util.state.exception import DataSourceUnavailable, RayStateApiException
@@ -3800,5 +3801,40 @@ def test_hang_driver_has_no_is_running_task(monkeypatch, ray_start_cluster):
     assert not all_job_info[my_job_id].HasField("is_running_tasks")
 
 
+def test_get_actor_timeout_multiplier(shutdown_only):
+    """Test that GetApiOptions applies the same timeout multiplier as ListApiOptions.
+
+    This test reproduces the issue where get_actor with timeout=1 fails even though
+    the actual operation takes less than 1 second, because GetApiOptions doesn't
+    apply the 0.8 server timeout multiplier that ListApiOptions uses.
+
+    Related issue: https://github.com/ray-project/ray/issues/54153
+    """
+
+    @ray.remote
+    class TestActor:
+        def ready(self):
+            pass
+
+    actor = TestActor.remote()
+    ray.get(actor.ready.remote())
+
+    # Test that both options classes apply the same timeout multiplier
+    test_timeout = 1
+    get_options = GetApiOptions(timeout=test_timeout)
+    list_options = ListApiOptions(timeout=test_timeout)
+
+    # After __post_init__, both should have the same effective timeout
+    assert get_options.timeout == list_options.timeout
+
+    # Test that get_actor works with a 1-second timeout
+    actors = list_actors()
+    actor_id = actors[0]["actor_id"]
+
+    # This should work without timeout issues
+    result = get_actor(actor_id, timeout=1)
+    assert result["actor_id"] == actor_id
+
+
 if __name__ == "__main__":
     sys.exit(pytest.main(["-sv", __file__]))

python/ray/util/state/common.py

@@ -152,7 +152,7 @@ def __post_init__(self):
         # To return the data to users, when there's a partial failure
         # we need to have a timeout that's smaller than the users' timeout.
         # 80% is configured arbitrarily.
-        self.timeout = int(self.timeout * self.server_timeout_multiplier)
+        self.timeout = max(1, int(self.timeout * self.server_timeout_multiplier))
         assert self.timeout != 0, "0 second timeout is not supported."
         if self.filters is None:
             self.filters = []
@@ -197,6 +197,18 @@ def has_conflicting_filters(self) -> bool:
 class GetApiOptions:
     # Timeout for the HTTP request
     timeout: int = DEFAULT_RPC_TIMEOUT
+    # When the request is processed on the server side,
+    # we should apply multiplier so that server side can finish
+    # processing a request within timeout. Otherwise,
+    # timeout will always lead Http timeout.
+    server_timeout_multiplier: float = 0.8
+
+    def __post_init__(self):
+        # To return the data to users, when there's a partial failure
+        # we need to have a timeout that's smaller than the users' timeout.
+        # 80% is configured arbitrarily.
+        self.timeout = max(1, int(self.timeout * self.server_timeout_multiplier))
+        assert self.timeout != 0, "0 second timeout is not supported."
 
 
 @dataclass(init=not IS_PYDANTIC_2)