[bugfix][distributed] fix multi-node bug for shared memory (vllm-proj…

…ect#6597)
KuntaiDu · Jul 19, 2024 · 98fb2f4 · 98fb2f4
1 parent 9dd56cf
commit 98fb2f4
Showing 1 changed file with 8 additions and 2 deletions.
diff --git a/vllm/distributed/device_communicators/shm_broadcast.py b/vllm/distributed/device_communicators/shm_broadcast.py
@@ -108,8 +108,14 @@ def __init__(self,
             # created by the process. The following patch is a workaround.
             with patch("multiprocessing.resource_tracker.register",
                        lambda *args, **kwargs: None):
-                self.shared_memory = shared_memory.SharedMemory(name=name)
-            assert self.shared_memory.size == self.total_bytes_of_buffer
+                try:
+                    self.shared_memory = shared_memory.SharedMemory(name=name)
+                    assert self.shared_memory.size == self.total_bytes_of_buffer  # noqa
+                except FileNotFoundError:
+                    # we might deserialize the object in a different node
+                    # in this case, this object is not used,
+                    # and we should suppress the error
+                    pass
 
     def __reduce__(self):
         return (