Features/scheduler refactor cursor

.gitignore

@@ -230,3 +230,6 @@ src/ui/next-env.d.ts
 !src/ui/public/manifest.json
 !src/ui/serve.json
 .eslintcache
+
+# vllm-sim
+bin/

src/guidellm/__main__.py

@@ -180,6 +180,17 @@ def benchmark():
         "If None, will run until max_seconds or the data is exhausted."
     ),
 )
+# TODO: Review Cursor generated code (start)
+@click.option(
+    "--max-error-rate",
+    type=float,
+    default=None,
+    help=(
+        "The maximum error rate allowed (0.0 to 1.0) before stopping the benchmark. "
+        "If None, no error rate constraint will be applied."
+    ),
+)
+# TODO: Review Cursor generated code (end)
 @click.option(
     "--warmup-percent",
     type=float,
@@ -261,6 +272,9 @@ def run(
     rate,
     max_seconds,
     max_requests,
+    # TODO: Review Cursor generated code (start)
+    max_error_rate,
+    # TODO: Review Cursor generated code (end)
     warmup_percent,
     cooldown_percent,
     disable_progress,
@@ -288,6 +302,9 @@ def run(
         rate=rate,
         max_seconds=max_seconds,
         max_requests=max_requests,
+        # TODO: Review Cursor generated code (start)
+        max_error_rate=max_error_rate,
+        # TODO: Review Cursor generated code (end)
         warmup_percent=warmup_percent,
         cooldown_percent=cooldown_percent,
         output_sampling=output_sampling,

src/guidellm/backend/openai.py

@@ -354,12 +354,53 @@ async def resolve(
                 request_info.request_timings.request_end = time.time()
                 response.request_output_tokens = usage_stats.output_tokens
                 response.request_prompt_tokens = usage_stats.prompt_tokens
+                # TODO: Review Cursor generated code (start)
+                logger.debug(
+                    f"OpenAI Backend: Got usage_stats - prompt_tokens={usage_stats.prompt_tokens}, output_tokens={usage_stats.output_tokens}"
+                )
+                # TODO: Review Cursor generated code (end)
+
+            # TODO: Review Cursor generated code (start)
+            # Debug what we're actually yielding
+            from loguru import logger
+            # TODO: Review Cursor generated code (end)
+
+            # TODO: Review Cursor generated code (start)
+            logger.debug("OpenAI Backend: About to yield response, request_info")
+            logger.debug(
+                f"OpenAI Backend: request_info.request_timings id: {id(request_info.request_timings)}"
+            )
+            if request_info.request_timings:
+                logger.debug(
+                    f"OpenAI Backend: Yielding with first_iteration={request_info.request_timings.first_iteration}, last_iteration={request_info.request_timings.last_iteration}"
+                )
+            else:
+                logger.debug("OpenAI Backend: Yielding with request_timings=None")
+            # TODO: Review Cursor generated code (end)
 
             yield response, request_info
 
         if request_info.request_timings.request_end is None:
             request_info.request_timings.request_end = time.time()
         response.delta = None
+
+        # TODO: Review Cursor generated code (start)
+        # Debug final yield
+        from loguru import logger
+        # TODO: Review Cursor generated code (end)
+
+        # TODO: Review Cursor generated code (start)
+        logger.debug(
+            f"OpenAI Backend: Final yield - request_info.request_timings id: {id(request_info.request_timings)}"
+        )
+        if request_info.request_timings:
+            logger.debug(
+                f"OpenAI Backend: Final yield with first_iteration={request_info.request_timings.first_iteration}, last_iteration={request_info.request_timings.last_iteration}"
+            )
+        else:
+            logger.debug("OpenAI Backend: Final yield with request_timings=None")
+        # TODO: Review Cursor generated code (end)
+
         yield response, request_info
 
     async def text_completions(