fix: safely parse metric labels (#948)

tests/unit/test_query/test_ontology_monitoring.py

@@ -0,0 +1,73 @@
+"""
+Tests for ontology monitoring metrics.
+"""
+
+import importlib.util
+import sys
+from pathlib import Path
+
+
+MODULE_PATH = (
+    Path(__file__).resolve().parents[3]
+    / "trustgraph-flow"
+    / "trustgraph"
+    / "query"
+    / "ontology"
+    / "monitoring.py"
+)
+spec = importlib.util.spec_from_file_location("ontology_monitoring", MODULE_PATH)
+assert spec is not None and spec.loader is not None
+monitoring = importlib.util.module_from_spec(spec)
+sys.modules[spec.name] = monitoring
+spec.loader.exec_module(monitoring)
+
+PerformanceMonitor = monitoring.PerformanceMonitor
+_extract_metric_label = monitoring._extract_metric_label
+
+
+def test_extract_metric_label_reads_unquoted_label_value():
+    metric_name = "cache_requests_total{cache_type=entity,component=ontology}"
+
+    assert _extract_metric_label(metric_name, "cache_type") == "entity"
+
+
+def test_extract_metric_label_reads_quoted_label_value():
+    metric_name = 'cache_requests_total{cache_type="entity",component="ontology"}'
+
+    assert _extract_metric_label(metric_name, "cache_type") == "entity"
+
+
+def test_extract_metric_label_returns_none_when_label_missing():
+    metric_name = "cache_requests_total{component=ontology}"
+
+    assert _extract_metric_label(metric_name, "cache_type") is None
+
+
+def test_performance_report_ignores_counters_without_cache_type_label():
+    monitor = PerformanceMonitor({"enabled": False})
+    monitor.metrics_collector.increment(
+        "cache_requests_total",
+        labels={"component": "ontology"},
+    )
+    monitor.metrics_collector.increment(
+        "cache_type=not_a_label",
+        labels={"component": "ontology"},
+    )
+    monitor.metrics_collector.increment(
+        "cache_requests_total",
+        labels={"cache_type": "entity"},
+    )
+    monitor.metrics_collector.increment(
+        "cache_hits_total",
+        labels={"cache_type": "entity"},
+    )
+
+    report = monitor.get_performance_report()
+
+    assert report["cache_performance"] == {
+        "entity": {
+            "hit_rate": 1.0,
+            "total_requests": 1.0,
+            "total_hits": 1.0,
+        }
+    }

trustgraph-flow/trustgraph/query/ontology/monitoring.py

@@ -4,6 +4,7 @@ Provides comprehensive monitoring of system performance, query patterns, and res
 """
 
 import logging
+import re
 import time
 import asyncio
 import inspect
@@ -276,6 +277,26 @@ class MetricsCollector:
         return f"{name}{{{label_str}}}"
 
 
+def _extract_metric_label(metric_name: str, label: str) -> Optional[str]:
+    """Extract a label value from an internal metric key."""
+    labels_start = metric_name.find('{')
+    labels_end = metric_name.find('}', labels_start + 1)
+
+    if labels_start == -1 or labels_end == -1:
+        return None
+
+    labels = metric_name[labels_start + 1:labels_end]
+    label_match = re.search(
+        rf'(?:^|,){re.escape(label)}=(?:"([^"]*)"|([^,]*))',
+        labels,
+    )
+    if not label_match:
+        return None
+
+    quoted_value, unquoted_value = label_match.groups()
+    return quoted_value if quoted_value is not None else unquoted_value
+
+
 class PerformanceMonitor:
     """Monitors system performance and component health."""
 
@@ -474,8 +495,8 @@ class PerformanceMonitor:
         # Cache performance
         cache_types = set()
         for metric_name in self.metrics_collector.counters.keys():
-            if 'cache_type=' in metric_name:
-                cache_type = metric_name.split('cache_type=')[1].split(',')[0].split('}')[0]
+            cache_type = _extract_metric_label(metric_name, 'cache_type')
+            if cache_type is not None:
                 cache_types.add(cache_type)
 
         for cache_type in cache_types: