perf(engine): avoid timed waits for Engine responses (#39486)

Signed-off-by: jthomson04 <jwillthomson19@gmail.com>
2026-09-21 20:42:03 -07:00
parent 56fee88e23
commit 15ba54bd5d
2 changed files with 92 additions and 3 deletions
@@ -1835,9 +1835,13 @@ class TokenizerManager(TokenizerControlMixin, TokenizerManagerScoreMixin):
        is_stream = getattr(obj, "stream", False)
        while True:
            try:
-                await asyncio.wait_for(
-                    state.event.wait(), timeout=_REQUEST_STATE_WAIT_TIMEOUT
-                )
+                if request is None:
+                    # Engine requests have no HTTP client to poll for disconnects.
+                    await state.event.wait()
+                else:
+                    await asyncio.wait_for(
+                        state.event.wait(), timeout=_REQUEST_STATE_WAIT_TIMEOUT
+                    )
            except asyncio.TimeoutError:
                if (
                    request is not None