fix(proxy): resolve high CPU when router_settings in DB by avoiding REGISTRY.collect() in PrometheusServicesLogger

litellm/integrations/prometheus_services.py

@@ -105,6 +105,11 @@ def _get_service_metrics_initialize(
         return metrics
 
     def is_metric_registered(self, metric_name) -> bool:
+        # Use _names_to_collectors (O(1)) instead of REGISTRY.collect() (O(n)) to avoid
+        # perf regression when a new Router is created per request (e.g. router_settings in DB).
+        names_to_collectors = getattr(self.REGISTRY, "_names_to_collectors", None)
+        if names_to_collectors is not None:
+            return metric_name in names_to_collectors
         for metric in self.REGISTRY.collect():
             if metric_name == metric.name:
                 return True

tests/test_litellm/integrations/test_prometheus_services.py

@@ -1,6 +1,7 @@
 import json
 import os
 import sys
+import time
 from unittest.mock import AsyncMock, patch
 
 import pytest
@@ -17,6 +18,63 @@
 )  # Adds the parent directory to the system path
 
 
+def test_is_metric_registered_does_not_use_registry_collect():
+    """is_metric_registered() must use _names_to_collectors, not REGISTRY.collect() (perf; #19921)."""
+    from prometheus_client import CollectorRegistry, Counter, Histogram
+
+    registry = CollectorRegistry()
+    for i in range(80):
+        Counter(
+            f"litellm_service_{i}_total_requests",
+            "Total requests",
+            labelnames=["service"],
+            registry=registry,
+        )
+        Histogram(
+            f"litellm_service_{i}_latency",
+            "Latency",
+            labelnames=["service"],
+            registry=registry,
+        )
+
+    pl = PrometheusServicesLogger()
+    pl.REGISTRY = registry
+
+    original_collect = registry.collect
+    collect_called = []
+
+    def track_collect(*args, **kwargs):
+        collect_called.append(1)
+        return original_collect(*args, **kwargs)
+
+    registry.collect = track_collect
+
+    n_calls = 30 * 2
+    start = time.perf_counter()
+    for _ in range(30):
+        pl.is_metric_registered("litellm_service_0_latency")
+        pl.is_metric_registered("litellm_service_79_total_requests")
+    elapsed_s = time.perf_counter() - start
+    elapsed_ms = elapsed_s * 1000
+    per_call_us = (elapsed_s / n_calls) * 1_000_000 if n_calls else 0
+    n_collect = len(collect_called)
+
+    path = "slow (REGISTRY.collect)" if n_collect else "fast (_names_to_collectors)"
+    print(
+        f"\n  is_metric_registered: {elapsed_ms:.2f} ms total | "
+        f"{per_call_us:.1f} µs/call | {n_calls} calls | {n_collect} collect() | {path}\n"
+    )
+
+    assert n_collect == 0, (
+        f"is_metric_registered() must not use REGISTRY.collect() when _names_to_collectors "
+        f"is available. Latency: {elapsed_ms:.2f} ms, {per_call_us:.1f} µs/call, {n_calls} calls, "
+        f"collect() called {n_collect} times."
+    )
+    assert elapsed_s < 0.05, (
+        f"is_metric_registered() took {elapsed_ms:.2f} ms for {n_calls} calls; expected <50 ms."
+    )
+
+
 def test_create_gauge_new():
     """Test creating a new gauge"""
     pl = PrometheusServicesLogger()