[Metrics] Add --show-hidden-metrics-for-version CLI arg (#13295)

vllm/engine/arg_utils.py

@@ -10,6 +10,7 @@ from typing import (TYPE_CHECKING, Any, Dict, List, Literal, Mapping, Optional,
 import torch
 
 import vllm.envs as envs
+from vllm import version
 from vllm.config import (CacheConfig, CompilationConfig, ConfigFormat,
                          DecodingConfig, DeviceConfig, HfOverrides,
                          KVTransferConfig, LoadConfig, LoadFormat, LoRAConfig,
@@ -188,6 +189,7 @@ class EngineArgs:
     qlora_adapter_name_or_path: Optional[str] = None
     disable_logprobs_during_spec_decoding: Optional[bool] = None
 
+    show_hidden_metrics_for_version: Optional[str] = None
     otlp_traces_endpoint: Optional[str] = None
     collect_detailed_traces: Optional[str] = None
     disable_async_output_proc: bool = False
@@ -909,6 +911,18 @@ class EngineArgs:
                             default=None,
                             help='Name or path of the QLoRA adapter.')
 
+        parser.add_argument('--show-hidden-metrics-for-version',
+                            type=str,
+                            default=None,
+                            help='Enable deprecated Prometheus metrics that '
+                            'have been hidden since the specified version. '
+                            'For example, if a previously deprecated metric '
+                            'has been hidden since the v0.7.0 release, you '
+                            'use --show-hidden-metrics-for-version=0.7 as a '
+                            'temporary escape hatch while you migrate to new '
+                            'metrics. The metric is likely to be removed '
+                            'completely in an upcoming release.')
+
         parser.add_argument(
             '--otlp-traces-endpoint',
             type=str,
@@ -1317,6 +1331,11 @@ class EngineArgs:
         decoding_config = DecodingConfig(
             guided_decoding_backend=self.guided_decoding_backend)
 
+        show_hidden_metrics = False
+        if self.show_hidden_metrics_for_version is not None:
+            show_hidden_metrics = version._prev_minor_version_was(
+                self.show_hidden_metrics_for_version)
+
         detailed_trace_modules = []
         if self.collect_detailed_traces is not None:
             detailed_trace_modules = self.collect_detailed_traces.split(",")
@@ -1326,6 +1345,7 @@ class EngineArgs:
                     f"Invalid module {m} in collect_detailed_traces. "
                     f"Valid modules are {ALLOWED_DETAILED_TRACE_MODULES}")
         observability_config = ObservabilityConfig(
+            show_hidden_metrics=show_hidden_metrics,
             otlp_traces_endpoint=self.otlp_traces_endpoint,
             collect_model_forward_time="model" in detailed_trace_modules
             or "all" in detailed_trace_modules,