How do let vllm v1 to use a customized stat logger extended from StatLoggerBase?

[liuzijing2014]

For V0, we are allowed to pass in customized stat loggers when we init the llm_engine:

vllm/vllm/engine/async_llm_engine.py

Lines 633 to 658 in 53ea6ad

	def from_engine_args(
	cls,
	engine_args: AsyncEngineArgs,
	engine_config: Optional[VllmConfig] = None,
	start_engine_loop: bool = True,
	usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
	stat_loggers: Optional[Dict[str, StatLoggerBase]] = None,
	) -> "AsyncLLMEngine":
	"""Creates an async LLM engine from the engine arguments."""
	# Create the engine configs.
	if engine_config is None:
	engine_config = engine_args.create_engine_config(usage_context)
	executor_class = cls._get_executor_cls(engine_config)
	# Create the async LLM engine.
	engine = cls(
	vllm_config=engine_config,
	executor_class=executor_class,
	log_requests=not engine_args.disable_log_requests,
	log_stats=not engine_args.disable_log_stats,
	start_engine_loop=start_engine_loop,
	usage_context=usage_context,
	stat_loggers=stat_loggers,
	)
	return engine

However, looking at v1, it seems that we are no longer taking in stat_logger:

vllm/vllm/v1/engine/async_llm.py

Lines 39 to 63 in 53ea6ad

	class AsyncLLM(EngineClient):
	def __init__(
	self,
	vllm_config: VllmConfig,
	executor_class: type[Executor],
	log_stats: bool,
	usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
	input_registry: InputRegistry = INPUT_REGISTRY,
	use_cached_outputs: bool = False,
	log_requests: bool = True,
	start_engine_loop: bool = True,
	) -> None:
	assert start_engine_loop
	self.model_config = vllm_config.model_config
	self.log_requests = log_requests
	self.log_stats = log_stats
	self.stat_loggers: list[StatLoggerBase] = []
	if self.log_stats:
	if logger.isEnabledFor(logging.INFO):
	self.stat_loggers.append(LoggingStatLogger())
	self.stat_loggers.append(PrometheusStatLogger(vllm_config))

The purpose is to log stats to a customized destination for persistent storage and analysis.