[Perf] API-server scaleout with many-to-many server-engine comms (#17546)

vllm/v1/request.py

@@ -26,12 +26,13 @@ class Request:
         multi_modal_placeholders: Optional[list[PlaceholderRange]],
         sampling_params: SamplingParams,
         eos_token_id: Optional[int],
-        arrival_time: float,
+        client_index: int = 0,
         lora_request: Optional["LoRARequest"] = None,
         structured_output_request: Optional["StructuredOutputRequest"] = None,
         cache_salt: Optional[str] = None,
     ) -> None:
         self.request_id = request_id
+        self.client_index = client_index
         self.sampling_params = sampling_params
         # Because of LoRA, the eos token id can be different for each request.
         self.eos_token_id = eos_token_id
@@ -90,13 +91,13 @@ class Request:
 
         return cls(
             request_id=request.request_id,
+            client_index=request.client_index,
             prompt_token_ids=request.prompt_token_ids,
             multi_modal_inputs=request.mm_inputs,
             multi_modal_hashes=request.mm_hashes,
             multi_modal_placeholders=request.mm_placeholders,
             sampling_params=request.sampling_params,
             eos_token_id=request.eos_token_id,
-            arrival_time=request.arrival_time,
             lora_request=request.lora_request,
             structured_output_request=StructuredOutputRequest(
                 sampling_params=request.sampling_params),