[Misc] Add more scoping for improved trace (#28329)

Signed-off-by: Wei Wei <wwei6@meta.com>
This commit is contained in:
Wei Wei
2025-11-10 21:03:21 +00:00
committed by GitHub
parent 40d33264c6
commit bf6a3d0ff5
4 changed files with 192 additions and 148 deletions
+21 -16
View File
@@ -36,6 +36,7 @@ from vllm.v1.executor import Executor
from vllm.v1.metrics.loggers import StatLoggerFactory, StatLoggerManager
from vllm.v1.metrics.reader import Metric, get_metrics_snapshot
from vllm.v1.metrics.stats import IterationStats
from vllm.v1.utils import record_function_or_nullcontext
from vllm.v1.worker.worker_base import WorkerBase
logger = init_logger(__name__)
@@ -280,28 +281,32 @@ class LLMEngine:
return []
# 1) Get EngineCoreOutput from the EngineCore.
outputs = self.engine_core.get_output()
with record_function_or_nullcontext("llm_genine step: get_output"):
outputs = self.engine_core.get_output()
# 2) Process EngineCoreOutputs.
iteration_stats = IterationStats() if self.log_stats else None
processed_outputs = self.output_processor.process_outputs(
outputs.outputs,
engine_core_timestamp=outputs.timestamp,
iteration_stats=iteration_stats,
)
self.output_processor.update_scheduler_stats(outputs.scheduler_stats)
with record_function_or_nullcontext("llm_genine step: process_outputs"):
iteration_stats = IterationStats() if self.log_stats else None
processed_outputs = self.output_processor.process_outputs(
outputs.outputs,
engine_core_timestamp=outputs.timestamp,
iteration_stats=iteration_stats,
)
self.output_processor.update_scheduler_stats(outputs.scheduler_stats)
# 3) Abort any reqs that finished due to stop strings.
self.engine_core.abort_requests(processed_outputs.reqs_to_abort)
with record_function_or_nullcontext("llm_genine step: abort_requests"):
self.engine_core.abort_requests(processed_outputs.reqs_to_abort)
# 4) Record stats
if self.logger_manager is not None and outputs.scheduler_stats is not None:
self.logger_manager.record(
scheduler_stats=outputs.scheduler_stats,
iteration_stats=iteration_stats,
mm_cache_stats=self.processor.stat_mm_cache(),
)
self.do_log_stats_with_interval()
with record_function_or_nullcontext("llm_genine step: record_stats"):
if self.logger_manager is not None and outputs.scheduler_stats is not None:
self.logger_manager.record(
scheduler_stats=outputs.scheduler_stats,
iteration_stats=iteration_stats,
mm_cache_stats=self.processor.stat_mm_cache(),
)
self.do_log_stats_with_interval()
return processed_outputs.request_outputs