metrics: support customer labels specified in request header (#10143)

python/sglang/srt/entrypoints/openai/serving_completions.py

@@ -59,6 +59,7 @@ class OpenAIServingCompletion(OpenAIServingBase):
     def _convert_to_internal_request(
         self,
         request: CompletionRequest,
+        raw_request: Request = None,
     ) -> tuple[GenerateReqInput, CompletionRequest]:
         """Convert OpenAI completion request to internal format"""
         # NOTE: with openai API, the prompt's logprobs are always not computed
@@ -89,6 +90,9 @@ class OpenAIServingCompletion(OpenAIServingBase):
         else:
             prompt_kwargs = {"input_ids": prompt}
 
+        # Extract customer labels from raw request headers
+        customer_labels = self.extract_customer_labels(raw_request)
+
         adapted_request = GenerateReqInput(
             **prompt_kwargs,
             sampling_params=sampling_params,
@@ -103,6 +107,7 @@ class OpenAIServingCompletion(OpenAIServingBase):
             bootstrap_room=request.bootstrap_room,
             return_hidden_states=request.return_hidden_states,
             rid=request.rid,
+            customer_labels=customer_labels,
         )
 
         return adapted_request, request