[Bugfix][Frontend] Honor request_id and X-Request-Id in chat and completions render (#61236)

Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>

tests/entrypoints/scale_out/render/test_render.py

@@ -367,6 +367,74 @@ async def test_completion_render_multiple_prompts(client):
         assert prompt["request_id"].startswith("cmpl-")
 
 
+CHAT_BODY = {
+    "model": MODEL_NAME,
+    "messages": [{"role": "user", "content": "Hello"}],
+}
+
+
+@pytest.mark.asyncio
+async def test_chat_render_honors_body_request_id(client):
+    response = await client.post(
+        "/v1/chat/completions/render", json={**CHAT_BODY, "request_id": "abc"}
+    )
+
+    assert response.status_code == 200
+    assert response.json()["request_id"] == "chatcmpl-abc"
+
+
+@pytest.mark.asyncio
+async def test_chat_render_header_request_id_wins(client):
+    response = await client.post(
+        "/v1/chat/completions/render",
+        json={**CHAT_BODY, "request_id": "abc"},
+        headers={"X-Request-Id": "hdr"},
+    )
+
+    assert response.status_code == 200
+    assert response.json()["request_id"] == "chatcmpl-hdr"
+
+
+@pytest.mark.asyncio
+async def test_chat_render_generates_unique_request_ids(client):
+    ids = set()
+    for _ in range(2):
+        response = await client.post("/v1/chat/completions/render", json=CHAT_BODY)
+        assert response.status_code == 200
+        request_id = response.json()["request_id"]
+        assert request_id.startswith("chatcmpl-")
+        ids.add(request_id)
+
+    assert len(ids) == 2
+
+
+@pytest.mark.asyncio
+async def test_completion_render_request_id_single_prompt(client):
+    response = await client.post(
+        "/v1/completions/render",
+        json={"model": MODEL_NAME, "prompt": "Hello world", "request_id": "abc"},
+    )
+
+    assert response.status_code == 200
+    assert [r["request_id"] for r in response.json()] == ["cmpl-abc"]
+
+
+@pytest.mark.asyncio
+async def test_completion_render_request_id_multiple_prompts(client):
+    response = await client.post(
+        "/v1/completions/render",
+        json={
+            "model": MODEL_NAME,
+            "prompt": ["Hello world", "Goodbye world"],
+            "request_id": "abc",
+        },
+        headers={"X-Request-Id": "hdr"},
+    )
+
+    assert response.status_code == 200
+    assert [r["request_id"] for r in response.json()] == ["cmpl-hdr-0", "cmpl-hdr-1"]
+
+
 @pytest.mark.asyncio
 async def test_chat_completion_render_multi_turn(client):
     """Test chat completion render with multi-turn conversation."""

vllm/entrypoints/scale_out/render/api_router.py

@@ -43,7 +43,7 @@ async def render_chat_completion(request: ChatCompletionRequest, raw_request: Re
             "The model does not support Chat Completions Render API"
         )
 
-    result = await handler.render_chat_request(request)
+    result = await handler.render_chat_request(request, raw_request)
 
     if isinstance(result, ErrorResponse):
         return JSONResponse(content=result.model_dump(), status_code=result.error.code)
@@ -67,7 +67,7 @@ async def render_messages(request: AnthropicMessagesRequest, raw_request: Reques
     if handler is None:
         raise NotImplementedError("The model does not support Messages Render API")
 
-    result = await handler.render_messages_request(request)
+    result = await handler.render_messages_request(request, raw_request)
 
     if isinstance(result, ErrorResponse):
         return JSONResponse(content=result.model_dump(), status_code=result.error.code)
@@ -90,7 +90,7 @@ async def render_completion(request: CompletionRequest, raw_request: Request):
     if handler is None:
         raise NotImplementedError("The model does not support Completions Render API")
 
-    result = await handler.render_completion_request(request)
+    result = await handler.render_completion_request(request, raw_request)
 
     if isinstance(result, ErrorResponse):
         return JSONResponse(content=result.model_dump(), status_code=result.error.code)

vllm/entrypoints/scale_out/render/serving.py

@@ -2,6 +2,8 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 from typing import TYPE_CHECKING
 
+from fastapi import Request
+
 from vllm.entrypoints.anthropic.protocol import AnthropicMessagesRequest
 from vllm.entrypoints.anthropic.serving import AnthropicServingMessages
 from vllm.entrypoints.openai.chat_completion.protocol import ChatCompletionRequest
@@ -31,7 +33,6 @@ from vllm.renderers.inputs.preprocess import (
     extract_prompt_len,
 )
 from vllm.renderers.online_renderer import OnlineRenderer
-from vllm.utils import random_uuid
 
 logger = init_logger(__name__)
 
@@ -77,6 +78,7 @@ class ServingRender(BaseServing):
     async def render_chat_request(
         self,
         request: ChatCompletionRequest,
+        raw_request: Request | None = None,
     ) -> GenerateRequest | ErrorResponse:
         """Validate the model and preprocess a chat completion request.
 
@@ -139,7 +141,9 @@ class ServingRender(BaseServing):
             )
         reasoning_ended = request.resolve_reasoning_ended(parser, token_ids)
 
-        request_id = f"chatcmpl-{random_uuid()}"
+        request_id = (
+            f"chatcmpl-{self._base_request_id(raw_request, request.request_id)}"
+        )
 
         return GenerateRequest(
             request_id=request_id,
@@ -161,6 +165,7 @@ class ServingRender(BaseServing):
     async def render_messages_request(
         self,
         request: AnthropicMessagesRequest,
+        raw_request: Request | None = None,
     ) -> GenerateRequest | ErrorResponse:
         """Validate the model and preprocess an Anthropic Messages request.
 
@@ -172,11 +177,12 @@ class ServingRender(BaseServing):
             request, merge_inline_system=self._merge_inline_system
         )
         chat_req.return_mm_kwargs = request.return_mm_kwargs
-        return await self.render_chat_request(chat_req)
+        return await self.render_chat_request(chat_req, raw_request)
 
     async def render_completion_request(
         self,
         request: CompletionRequest,
+        raw_request: Request | None = None,
     ) -> list[GenerateRequest] | ErrorResponse:
         """Validate the model and preprocess a completion request.
 
@@ -197,8 +203,9 @@ class ServingRender(BaseServing):
         )
         if isinstance(result, ErrorResponse):
             return result
+        base_id = f"cmpl-{self._base_request_id(raw_request, request.request_id)}"
         generate_requests: list[GenerateRequest] = []
-        for engine_input in result:
+        for i, engine_input in enumerate(result):
             prompt_components = extract_prompt_components(
                 self.model_config, engine_input
             )
@@ -220,7 +227,7 @@ class ServingRender(BaseServing):
                 max_tokens, self.default_sampling_params
             )
 
-            request_id = f"cmpl-{random_uuid()}"
+            request_id = base_id if len(result) == 1 else f"{base_id}-{i}"
 
             generate_requests.append(
                 GenerateRequest(