fix: fix logical error when destroying ranks group

Author: specture724

checkpoint_engine/ps.py

@@ -969,6 +969,7 @@ def update(
                 which is useful in disaggregated architecture.
         """
         assert req_func is not None, "req_func is required"
+        ranks_group = None
         try:
             master_addr = os.getenv("MASTER_ADDR")
             assert master_addr, "master_addr is required"
@@ -993,22 +994,23 @@ def update(
                 self._update_per_bucket(checkpoint_name, req_func, ranks_group, ranks)
 
             self.store_based_barrier(manager_store)
-            logger.info(
-                f"[rank{self._rank}] update checkpoint {checkpoint_name} with ranks {ranks} done. "
-                f"Current CUDA allocated {self.device_manager.device_module.memory_allocated() / 1024 / 1024} MB, "
-                f"reserved {self.device_manager.device_module.memory_reserved() / 1024 / 1024} MB."
-            )
+
         except Exception as e:
             logger.exception(
                 f"[rank{self._rank}] update checkpoint {checkpoint_name} with ranks {ranks} error {e}"
             )
             raise
         finally:
-            if not ranks_group:
+            if ranks_group:
                 dist.destroy_process_group(ranks_group)
             if self._auto_pg and dist.is_initialized():
                 dist.destroy_process_group()
             self.device_manager.device_module.empty_cache()
+            logger.info(
+                f"[rank{self._rank}] update checkpoint {checkpoint_name} with ranks {ranks} done. "
+                f"Current CUDA allocated {self.device_manager.device_module.memory_allocated() / 1024 / 1024} MB, "
+                f"reserved {self.device_manager.device_module.memory_reserved() / 1024 / 1024} MB."
+            )
 
     def _bind_zmq_socket(self) -> tuple[zmq.Socket, list[tuple[str, str]]]:
         def zmq_handle(device_uuid: str) -> str: