[Core] Add span metrics for model_forward, scheduler and sampler time (#7089)

vllm/executor/executor_base.py

@@ -2,8 +2,8 @@ from abc import ABC, abstractmethod
 from typing import List, Optional, Set, Tuple
 
 from vllm.config import (CacheConfig, DeviceConfig, LoadConfig, LoRAConfig,
-                         ModelConfig, MultiModalConfig, ParallelConfig,
-                         PromptAdapterConfig, SchedulerConfig,
+                         ModelConfig, MultiModalConfig, ObservabilityConfig,
+                         ParallelConfig, PromptAdapterConfig, SchedulerConfig,
                          SpeculativeConfig)
 from vllm.lora.request import LoRARequest
 from vllm.prompt_adapter.request import PromptAdapterRequest
@@ -32,6 +32,7 @@ class ExecutorBase(ABC):
         multimodal_config: Optional[MultiModalConfig],
         speculative_config: Optional[SpeculativeConfig],
         prompt_adapter_config: Optional[PromptAdapterConfig],
+        observability_config: Optional[ObservabilityConfig],
     ) -> None:
         self.model_config = model_config
         self.cache_config = cache_config
@@ -43,7 +44,7 @@ class ExecutorBase(ABC):
         self.multimodal_config = multimodal_config
         self.speculative_config = speculative_config
         self.prompt_adapter_config = prompt_adapter_config
-
+        self.observability_config = observability_config
         self._init_executor()
 
     @abstractmethod

vllm/executor/gpu_executor.py

@@ -60,6 +60,7 @@ class GPUExecutor(ExecutorBase):
             prompt_adapter_config=self.prompt_adapter_config,
             is_driver_worker=(not self.parallel_config)
             or (rank % self.parallel_config.tensor_parallel_size == 0),
+            observability_config=self.observability_config,
         )
 
     def _get_create_worker_kwargs(