[Bugfix] Backport request id validation to v0 (vllm-project#11036)

Signed-off-by: Joe Runde <[email protected]>
sleepwalker2017 · Dec 13, 2024 · 0f22716 · 0f22716
1 parent 73be551
commit 0f22716
Show file tree

Hide file tree

Showing 2 changed files with 5 additions and 1 deletion.
diff --git a/vllm/engine/multiprocessing/client.py b/vllm/engine/multiprocessing/client.py
@@ -576,6 +576,10 @@ async def _process_request(
         if self._errored_with is not None:
             raise ENGINE_DEAD_ERROR(self._errored_with)
 
+        # Ensure the request id is unique among running requests
+        if request_id in self.output_queues:
+            raise ValueError(f"Request {request_id} already exists")
+
         # Constructing guided decoding logits processors is expensive, so we do
         # it here to avoid contending with cpu resources and the GIL on the
         # backend process.

diff --git a/vllm/v1/engine/async_llm.py b/vllm/v1/engine/async_llm.py
@@ -152,7 +152,7 @@ async def add_request(
         """Add new request to the AsyncLLM."""
 
         if self.detokenizer.is_request_active(request_id):
-            raise KeyError(f"Request {request_id} already exists.")
+            raise ValueError(f"Request {request_id} already exists.")
 
         # 1) Create a new AsyncStream for the request.
         stream = self._add_request_to_streams(request_id)