[Core] Change `execute_model_with_error_logging()` to be a ctx manager (#27060)

Signed-off-by: Nick Hill <nhill@redhat.com>

vllm/v1/engine/core.py

@@ -290,14 +290,11 @@ class EngineCore:
         # (i.e. client-aborted vs stop criteria met).
         self.scheduler.finish_requests(request_ids, RequestStatus.FINISHED_ABORTED)
 
-    def execute_model_with_error_logging(
-        self,
-        model_fn: Callable[[SchedulerOutput], ModelRunnerOutput],
-        scheduler_output: SchedulerOutput,
-    ) -> ModelRunnerOutput:
+    @contextmanager
+    def log_error_detail(self, scheduler_output: SchedulerOutput):
         """Execute the model and log detailed info on failure."""
         try:
-            return model_fn(scheduler_output)
+            yield
         except Exception as err:
             # We do not want to catch BaseException here since we're only
             # interested in dumping info when the exception is due to an
@@ -321,15 +318,16 @@ class EngineCore:
         if not self.scheduler.has_requests():
             return {}, False
         scheduler_output = self.scheduler.schedule()
-        model_output = self.execute_model_with_error_logging(
-            self.model_executor.execute_model,  # type: ignore
-            scheduler_output,
-        )
+
+        with self.log_error_detail(scheduler_output):
+            model_output = self.model_executor.execute_model(scheduler_output)
+
+        assert isinstance(model_output, ModelRunnerOutput)
         engine_core_outputs = self.scheduler.update_from_output(
             scheduler_output, model_output
         )
 
-        return (engine_core_outputs, scheduler_output.total_num_scheduled_tokens > 0)
+        return engine_core_outputs, scheduler_output.total_num_scheduled_tokens > 0
 
     def post_step(self, model_executed: bool) -> None:
         if self.use_spec_decode and model_executed:
@@ -386,14 +384,12 @@ class EngineCore:
 
         # Block until the next result is available.
         future, scheduler_output = batch_queue.pop()
-        model_output = self.execute_model_with_error_logging(
-            lambda _: future.result(), scheduler_output
-        )
+        with self.log_error_detail(scheduler_output):
+            model_output = future.result()
 
         engine_core_outputs = self.scheduler.update_from_output(
             scheduler_output, model_output
         )
-
         return engine_core_outputs, model_executed
 
     def shutdown(self):