[Frontend] Pass API server count to each process (#23717)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/config/parallel.py

@@ -193,6 +193,25 @@ class ParallelConfig:
     not change by dcp, it simply reuse the GPUs of TP group, and tp_size
     needs to be divisible by dcp_size."""
 
+    _api_process_count: int = 1
+    """
+    The number of API processes initialized.
+
+    Note:
+        This is an internal config that is only valid for and
+        should only be set by API server scale-out.
+    """
+
+    _api_process_rank: int = 0
+    """
+    The rank of this API process, or `-1` for engine core processes
+    under API server scale-out.
+
+    Note:
+        This is an internal config that is only valid for and
+        should only be set by API server scale-out.
+    """
+
     @property
     def world_size_across_dp(self) -> int:
         """world_size_across_dp is TPxPPxDP, it is the size of the world
@@ -428,6 +447,12 @@ class ParallelConfig:
         if self.distributed_executor_backend is None and self.world_size == 1:
             self.distributed_executor_backend = "uni"
 
+        if not -1 <= self._api_process_rank < self._api_process_count:
+            raise ValueError(
+                "Invalid value of `_api_process_rank`. "
+                f"Expected to be `-1` or `[0, {self._api_process_count})`, "
+                f"but found: {self._api_process_rank}")
+
     @property
     def use_ray(self) -> bool:
         return self.distributed_executor_backend == "ray" or (