[PlacementGroup]Fix placement group wait api disorder bug (#12827)

* [PlacementGroup]Fix placment group wait api disorder bug

* fix review comment

* fix review comment

* fix review comment

* fix review comments

* increase num_heartbeats_timeout

Co-authored-by: 灵洵 <fengbin.ffb@antgroup.com>

python/ray/_raylet.pyx

@@ -1207,14 +1207,17 @@ cdef class CoreWorker:
 
     def wait_placement_group_ready(self,
                                    PlacementGroupID placement_group_id,
-                                   int32_t timeout_ms):
+                                   int32_t timeout_seconds):
         cdef CRayStatus status
         cdef CPlacementGroupID cplacement_group_id = (
             CPlacementGroupID.FromBinary(placement_group_id.binary()))
-        cdef int ctimeout_ms = timeout_ms
+        cdef int ctimeout_seconds = timeout_seconds
         with nogil:
             status = CCoreWorkerProcess.GetCoreWorker() \
-                .WaitPlacementGroupReady(cplacement_group_id, ctimeout_ms)
+                .WaitPlacementGroupReady(cplacement_group_id, ctimeout_seconds)
+            if status.IsNotFound():
+                raise Exception("Placement group {} does not exist.".format(
+                    placement_group_id))
         return status.ok()
 
     def submit_actor_task(self,

python/ray/tests/test_placement_group.py

@@ -1256,5 +1256,37 @@ def test_create_placement_group_during_gcs_server_restart(
         ray.get(placement_groups[i].ready())
 
 
+@pytest.mark.parametrize(
+    "ray_start_cluster_head", [
+        generate_system_config_map(
+            num_heartbeats_timeout=20, ping_gcs_rpc_server_max_retries=60)
+    ],
+    indirect=True)
+def test_placement_group_wait_api(ray_start_cluster_head):
+    cluster = ray_start_cluster_head
+    cluster.add_node(num_cpus=2)
+    cluster.add_node(num_cpus=2)
+    cluster.wait_for_nodes()
+
+    # Create placement group 1 successfully.
+    placement_group1 = ray.util.placement_group([{"CPU": 1}, {"CPU": 1}])
+    assert placement_group1.wait(10)
+
+    # Restart gcs server.
+    cluster.head_node.kill_gcs_server()
+    cluster.head_node.start_gcs_server()
+
+    # Create placement group 2 successfully.
+    placement_group2 = ray.util.placement_group([{"CPU": 1}, {"CPU": 1}])
+    assert placement_group2.wait(10)
+
+    # Remove placement group 1.
+    ray.util.remove_placement_group(placement_group1)
+
+    # Wait for placement group 1 after it is removed.
+    with pytest.raises(Exception):
+        placement_group1.wait(10)
+
+
 if __name__ == "__main__":
     sys.exit(pytest.main(["-v", __file__]))

python/ray/util/placement_group.py

@@ -83,6 +83,19 @@ class PlacementGroup:
             placement_group_bundle_index=bundle_index,
             resources=resources).remote(self)
 
+    def wait(self, timeout_seconds: int) -> bool:
+        """Wait for the placement group to be ready within the specified time.
+        Args:
+             timeout_seconds(str): Timeout in seconds.
+        Return:
+             True if the placement group is created. False otherwise.
+        """
+        worker = ray.worker.global_worker
+        worker.check_connected()
+
+        return worker.core_worker.wait_placement_group_ready(
+            self.id, timeout_seconds)
+
     @property
     def bundle_specs(self) -> List[Dict]:
         """List[Dict]: Return bundles belonging to this placement group."""