优化系统健康评分与缓存监测

backend/app/api/v1/permission_monitoring.py

@@ -15,7 +15,11 @@ from sqlalchemy import func, select, desc
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.core.deps import get_current_user, get_db_session, is_system_admin
-from app.core.permission_monitor import get_permission_monitor
+from app.core.permission_monitor import (
+    CACHE_HIT_RATE_HEALTH_MIN_ACCESSES,
+    CACHE_METRICS_WINDOW_SECONDS,
+    get_permission_monitor,
+)
 from app.models.permission_access_log import PermissionAccessLog
 from app.models.permission_metric_snapshot import PermissionMetricSnapshot
 from app.models.security_access_log import SecurityAccessLog
@@ -78,7 +82,7 @@ async def get_permission_metrics(
     row = result.one()
     total = row.total_checks or 0
     monitor = get_permission_monitor()
-    cache_metrics = monitor.metrics.cache_metrics.to_dict()
+    cache_metrics = monitor.metrics.cache_metrics.window_stats()
     return {
         "window_hours": hours,
         "check_metrics": {
@@ -178,33 +182,87 @@ async def permission_system_health(
     deny_rate = round(row.denied / total * 100, 2) if total else 0
 
     monitor = get_permission_monitor()
-    cache_metrics = monitor.metrics.cache_metrics
+    cache_metrics = monitor.metrics.cache_metrics.window_stats()
+    now_timestamp = datetime.now(timezone.utc).timestamp()
+    recent_alerts = [
+        alert
+        for alert in monitor.get_alerts(limit=1000)
+        if alert["timestamp"] >= now_timestamp - 3600
+    ]
 
-    health_score = 100
-    issues = []
+    deductions: dict[str, int] = {
+        "performance": 0,
+        "deny_rate": 0,
+        "cache": 0,
+        "alerts": 0,
+    }
+    issues: list[str] = []
+
+    if avg_ms > 200:
+        deductions["performance"] = 40
+    elif avg_ms > 50:
+        deductions["performance"] = 25
+    elif avg_ms > 10:
+        deductions["performance"] = 10
+    if deductions["performance"]:
+        issues.append(f"权限检查平均响应时间过长（{avg_ms:.2f}ms）")
 
-    if avg_ms > 10:
-        health_score -= 10
-        issues.append("权限检查响应时间过长")
     if deny_rate > 50:
-        health_score -= 5
-        issues.append("权限拒绝率过高")
-    if (
-        cache_metrics.total_accesses >= 10
-        and cache_metrics.hit_rate < 50
-    ):
-        health_score -= 10
-        issues.append("缓存命中率过低")
+        deductions["deny_rate"] = 20
+    elif deny_rate > 20:
+        deductions["deny_rate"] = 10
+    if deductions["deny_rate"]:
+        issues.append(f"权限拒绝率偏高（{deny_rate:.2f}%）")
+
+    cache_sample_sufficient = (
+        cache_metrics["total_accesses"] >= CACHE_HIT_RATE_HEALTH_MIN_ACCESSES
+    )
+    if cache_sample_sufficient:
+        if cache_metrics["hit_rate"] < 20:
+            deductions["cache"] = 25
+        elif cache_metrics["hit_rate"] < 50:
+            deductions["cache"] = 20
+        elif cache_metrics["hit_rate"] < 80:
+            deductions["cache"] = 10
+        if deductions["cache"]:
+            issues.append(f"近 1 小时缓存命中率偏低（{cache_metrics['hit_rate']:.2f}%）")
+
+    error_alert_count = sum(alert["level"] == "error" for alert in recent_alerts)
+    warning_alert_count = sum(alert["level"] == "warning" for alert in recent_alerts)
+    if error_alert_count:
+        deductions["alerts"] = 15
+        issues.append(f"近 1 小时发生 {error_alert_count} 条错误告警")
+    elif warning_alert_count:
+        deductions["alerts"] = 5
+        issues.append(f"近 1 小时发生 {warning_alert_count} 条警告")
+
+    health_score = max(0, 100 - sum(deductions.values()))
+    sample_sufficient = total >= 10
 
     return {
         "status": "healthy" if health_score >= 80 else "degraded" if health_score >= 50 else "unhealthy",
-        "health_score": max(0, health_score),
+        "health_score": health_score,
         "issues": issues,
+        "sample_sufficient": sample_sufficient,
+        "score_breakdown": {
+            "performance": {"weight": 40, "deduction": deductions["performance"]},
+            "deny_rate": {"weight": 20, "deduction": deductions["deny_rate"]},
+            "cache": {
+                "weight": 25,
+                "deduction": deductions["cache"],
+                "sample_sufficient": cache_sample_sufficient,
+                "scope": "sliding_window",
+                "window_seconds": CACHE_METRICS_WINDOW_SECONDS,
+            },
+            "alerts": {"weight": 15, "deduction": deductions["alerts"]},
+        },
         "last_hour": {
             "total_checks": total,
             "denied_checks": row.denied or 0,
             "deny_rate": deny_rate,
             "avg_elapsed_ms": round(avg_ms, 2),
+            "error_alerts": error_alert_count,
+            "warning_alerts": warning_alert_count,
         },
         "cache_stats": monitor.get_cache_stats(),
     }