[Misc] Have AsyncLLM custom_stat_loggers extend default logger list

tests/v1/metrics/test_engine_logger_apis.py

@@ -0,0 +1,36 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+import pytest
+
+from vllm.v1.engine.async_llm import AsyncEngineArgs, AsyncLLM
+from vllm.v1.metrics.ray_wrappers import RayPrometheusStatLogger
+
+DEFAULT_ENGINE_ARGS = AsyncEngineArgs(
+    model="distilbert/distilgpt2",
+    dtype="half",
+    disable_log_stats=False,
+    enforce_eager=True,
+)
+
+
+@pytest.mark.asyncio
+async def test_async_llm_replace_default_loggers():
+    # Empty stat_loggers removes default loggers
+    engine = AsyncLLM.from_engine_args(DEFAULT_ENGINE_ARGS, stat_loggers=[])
+    await engine.add_logger(RayPrometheusStatLogger)
+
+    # Verify that only this logger is present in shared loggers
+    assert len(engine.logger_manager.shared_loggers) == 1
+    assert isinstance(engine.logger_manager.shared_loggers[0],
+                      RayPrometheusStatLogger)
+
+
+@pytest.mark.asyncio
+async def test_async_llm_add_to_default_loggers():
+    # Start with default loggers, including PrometheusStatLogger
+    engine = AsyncLLM.from_engine_args(DEFAULT_ENGINE_ARGS)
+
+    # Add another PrometheusStatLogger subclass
+    await engine.add_logger(RayPrometheusStatLogger)
+
+    assert len(engine.logger_manager.shared_loggers) == 2

vllm/v1/engine/async_llm.py

@@ -36,7 +36,8 @@
 from vllm.v1.engine.parallel_sampling import ParentRequest
 from vllm.v1.engine.processor import Processor
 from vllm.v1.executor.abstract import Executor
-from vllm.v1.metrics.loggers import StatLoggerFactory, StatLoggerManager
+from vllm.v1.metrics.loggers import (DpSharedStatLoggerFactory,
+                                     StatLoggerFactory, StatLoggerManager)
 from vllm.v1.metrics.prometheus import shutdown_prometheus
 from vllm.v1.metrics.stats import IterationStats
 
@@ -55,7 +56,8 @@ def __init__(
         use_cached_outputs: bool = False,
         log_requests: bool = True,
         start_engine_loop: bool = True,
-        stat_loggers: Optional[list[StatLoggerFactory]] = None,
+        stat_loggers: Optional[list[Union[StatLoggerFactory,
+                                          DpSharedStatLoggerFactory]]] = None,
         client_addresses: Optional[dict[str, str]] = None,
         client_index: int = 0,
     ) -> None:
@@ -144,7 +146,8 @@ def from_vllm_config(
         vllm_config: VllmConfig,
         start_engine_loop: bool = True,
         usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
-        stat_loggers: Optional[list[StatLoggerFactory]] = None,
+        stat_loggers: Optional[list[Union[StatLoggerFactory,
+                                          DpSharedStatLoggerFactory]]] = None,
         disable_log_requests: bool = False,
         disable_log_stats: bool = False,
         client_addresses: Optional[dict[str, str]] = None,
@@ -176,7 +179,8 @@ def from_engine_args(
         engine_args: AsyncEngineArgs,
         start_engine_loop: bool = True,
         usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
-        stat_loggers: Optional[list[StatLoggerFactory]] = None,
+        stat_loggers: Optional[list[Union[StatLoggerFactory,
+                                          DpSharedStatLoggerFactory]]] = None,
     ) -> "AsyncLLM":
         """Create an AsyncLLM from the EngineArgs."""
 
@@ -596,6 +600,17 @@ async def collective_rpc(self,
         return await self.engine_core.collective_rpc_async(
             method, timeout, args, kwargs)
 
+    async def add_logger(
+        self, logger_factory: Union[StatLoggerFactory,
+                                    DpSharedStatLoggerFactory]
+    ) -> None:
+        if self.logger_manager is None:
+            raise RuntimeError(
+                "Stat logging is disabled. Set `disable_log_stats=False` "
+                "engine argument to enable.")
+
+        self.logger_manager.add_logger(logger_factory)
+
     async def wait_for_requests_to_drain(self, drain_timeout: int = 300):
         """Wait for all requests to be drained."""
         start_time = time.time()

vllm/v1/metrics/loggers.py

@@ -20,6 +20,8 @@
 logger = init_logger(__name__)
 
 StatLoggerFactory = Callable[[VllmConfig, int], "StatLoggerBase"]
+DpSharedStatLoggerFactory = Callable[[VllmConfig, Optional[list[int]]],
+                                     "PrometheusStatLogger"]
 
 
 class StatLoggerBase(ABC):
@@ -633,37 +635,67 @@ def __init__(
         self,
         vllm_config: VllmConfig,
         engine_idxs: Optional[list[int]] = None,
-        custom_stat_loggers: Optional[list[StatLoggerFactory]] = None,
+        custom_stat_loggers: Optional[list[Union[
+            StatLoggerFactory, DpSharedStatLoggerFactory]]] = None,
     ):
+        """
+        Initializes the StatLoggerManager.
+
+        Args:
+            vllm_config (VllmConfig): The configuration object for vLLM.
+            engine_idxs (Optional[list[int]]): List of engine indices. If None,
+                defaults to [0].
+            custom_stat_loggers (Optional[list[Union[
+                StatLoggerFactory, DpSharedStatLoggerFactory
+            ]]]): 
+                Optional list of custom stat logger factories to use. If None, 
+                default loggers are used.
+        """
         self.engine_idxs = engine_idxs if engine_idxs else [0]
+        self.vllm_config = vllm_config
 
-        factories: list[StatLoggerFactory]
+        factories: list[StatLoggerFactory] = []
+        shared_logger_factories: list[DpSharedStatLoggerFactory] = []
         if custom_stat_loggers is not None:
-            factories = custom_stat_loggers
+            for factory in custom_stat_loggers:
+                if isinstance(factory, type) and issubclass(
+                        factory, PrometheusStatLogger):
+                    shared_logger_factories.append(factory)  # type: ignore
+                else:
+                    factories.append(factory)  # type: ignore
         else:
-            factories = []
             if logger.isEnabledFor(logging.INFO):
                 factories.append(LoggingStatLogger)
 
+            shared_logger_factories.append(PrometheusStatLogger)
+
+        self.shared_loggers = []
+        if len(shared_logger_factories) > 0:
+            for factory in shared_logger_factories:
+                self.shared_loggers.append(factory(vllm_config, engine_idxs))
+
         # engine_idx: StatLogger
         self.per_engine_logger_dict: dict[int, list[StatLoggerBase]] = {}
-        prometheus_factory = PrometheusStatLogger
         for engine_idx in self.engine_idxs:
             loggers: list[StatLoggerBase] = []
             for logger_factory in factories:
-                # If we get a custom prometheus logger, use that
-                # instead. This is typically used for the ray case.
-                if (isinstance(logger_factory, type)
-                        and issubclass(logger_factory, PrometheusStatLogger)):
-                    prometheus_factory = logger_factory
-                    continue
                 loggers.append(logger_factory(vllm_config,
                                               engine_idx))  # type: ignore
             self.per_engine_logger_dict[engine_idx] = loggers
 
-        # For Prometheus, need to share the metrics between EngineCores.
-        # Each EngineCore's metrics are expressed as a unique label.
-        self.prometheus_logger = prometheus_factory(vllm_config, engine_idxs)
+    def add_logger(
+        self, logger_factory: Union[StatLoggerFactory,
+                                    DpSharedStatLoggerFactory]
+    ) -> None:
+        if (isinstance(logger_factory, type)
+                and issubclass(logger_factory, PrometheusStatLogger)):
+            self.shared_loggers.append(
+                logger_factory(self.vllm_config,
+                               self.engine_idxs))  # type: ignore
+        else:
+            for engine_idx, logger_list in self.per_engine_logger_dict.items():
+                logger_list.append(logger_factory(self.vllm_config,
+                                                  engine_idx))  # type: ignore
 
     def record(
         self,
@@ -678,17 +710,18 @@ def record(
         for logger in per_engine_loggers:
             logger.record(scheduler_stats, iteration_stats, engine_idx)
 
-        self.prometheus_logger.record(scheduler_stats, iteration_stats,
-                                      engine_idx)
+        for logger in self.shared_loggers:
+            logger.record(scheduler_stats, iteration_stats, engine_idx)
 
     def log(self):
         for per_engine_loggers in self.per_engine_logger_dict.values():
             for logger in per_engine_loggers:
                 logger.log()
 
     def log_engine_initialized(self):
-        self.prometheus_logger.log_engine_initialized()
+        for shared_logger in self.shared_loggers:
+            shared_logger.log_engine_initialized()
 
         for per_engine_loggers in self.per_engine_logger_dict.values():
-            for logger in per_engine_loggers:
-                logger.log_engine_initialized()
+            for per_engine_logger in per_engine_loggers:
+                per_engine_logger.log_engine_initialized()