Files
ctms/backend/app/core/permission_monitor.py
T
2026-05-19 16:03:27 +08:00

335 lines
9.7 KiB
Python

"""权限系统监控
实现权限系统的监控功能,包括:
- 权限检查统计
- 缓存性能监控
- 异常检测
- 性能指标收集
"""
from __future__ import annotations
import time
import uuid
from dataclasses import dataclass, field
from typing import Any
from app.core.permission_cache import get_permission_cache
CACHE_HIT_RATE_HEALTH_MIN_ACCESSES = 10
@dataclass
class PermissionCheckMetrics:
"""权限检查指标"""
total_checks: int = 0 # 总检查次数
allowed_checks: int = 0 # 允许的检查次数
denied_checks: int = 0 # 拒绝的检查次数
total_time: float = 0.0 # 总耗时(秒)
min_time: float = float("inf") # 最小耗时(秒)
max_time: float = 0.0 # 最大耗时(秒)
errors: int = 0 # 错误次数
@property
def avg_time(self) -> float:
"""平均耗时(秒)"""
if self.total_checks == 0:
return 0.0
return self.total_time / self.total_checks
@property
def allow_rate(self) -> float:
"""允许率(百分比)"""
if self.total_checks == 0:
return 0.0
return (self.allowed_checks / self.total_checks) * 100
@property
def deny_rate(self) -> float:
"""拒绝率(百分比)"""
if self.total_checks == 0:
return 0.0
return (self.denied_checks / self.total_checks) * 100
@property
def error_rate(self) -> float:
"""错误率(百分比)"""
if self.total_checks == 0:
return 0.0
return (self.errors / self.total_checks) * 100
def to_dict(self) -> dict[str, Any]:
"""转换为字典"""
return {
"total_checks": self.total_checks,
"allowed_checks": self.allowed_checks,
"denied_checks": self.denied_checks,
"total_time": round(self.total_time, 3),
"min_time": round(self.min_time, 3) if self.min_time != float("inf") else 0,
"max_time": round(self.max_time, 3),
"avg_time": round(self.avg_time, 3),
"allow_rate": round(self.allow_rate, 2),
"deny_rate": round(self.deny_rate, 2),
"error_rate": round(self.error_rate, 2),
"errors": self.errors,
}
@dataclass
class CacheMetrics:
"""缓存指标"""
total_accesses: int = 0 # 总访问次数
cache_hits: int = 0 # 缓存命中次数
cache_misses: int = 0 # 缓存未命中次数
cache_invalidations: int = 0 # 缓存失效次数
@property
def hit_rate(self) -> float:
"""缓存命中率(百分比)"""
if self.total_accesses == 0:
return 0.0
return (self.cache_hits / self.total_accesses) * 100
@property
def miss_rate(self) -> float:
"""缓存未命中率(百分比)"""
if self.total_accesses == 0:
return 0.0
return (self.cache_misses / self.total_accesses) * 100
def to_dict(self) -> dict[str, Any]:
"""转换为字典"""
return {
"total_accesses": self.total_accesses,
"cache_hits": self.cache_hits,
"cache_misses": self.cache_misses,
"cache_invalidations": self.cache_invalidations,
"hit_rate": round(self.hit_rate, 2),
"miss_rate": round(self.miss_rate, 2),
}
@dataclass
class PermissionSystemMetrics:
"""权限系统指标"""
check_metrics: PermissionCheckMetrics = field(default_factory=PermissionCheckMetrics)
cache_metrics: CacheMetrics = field(default_factory=CacheMetrics)
last_reset_time: float = field(default_factory=time.time)
def reset(self) -> None:
"""重置所有指标"""
self.check_metrics = PermissionCheckMetrics()
self.cache_metrics = CacheMetrics()
self.last_reset_time = time.time()
def to_dict(self) -> dict[str, Any]:
"""转换为字典"""
return {
"check_metrics": self.check_metrics.to_dict(),
"cache_metrics": self.cache_metrics.to_dict(),
"uptime_seconds": time.time() - self.last_reset_time,
}
class PermissionMonitor:
"""权限系统监控器
收集权限系统的运行指标,用于监控和告警。
"""
def __init__(self):
"""初始化监控器"""
self.metrics = PermissionSystemMetrics()
self._alerts: list[dict[str, Any]] = []
def record_permission_check(
self,
allowed: bool,
elapsed_time: float,
error: Exception | None = None,
) -> None:
"""记录权限检查
Args:
allowed: 是否允许
elapsed_time: 耗时(秒)
error: 错误对象(如果有)
"""
metrics = self.metrics.check_metrics
metrics.total_checks += 1
if allowed:
metrics.allowed_checks += 1
else:
metrics.denied_checks += 1
metrics.total_time += elapsed_time
metrics.min_time = min(metrics.min_time, elapsed_time)
metrics.max_time = max(metrics.max_time, elapsed_time)
if error:
metrics.errors += 1
self._check_error_alert(error)
# 检查性能告警
self._check_performance_alert(elapsed_time)
def record_cache_hit(self) -> None:
"""记录缓存命中"""
metrics = self.metrics.cache_metrics
metrics.total_accesses += 1
metrics.cache_hits += 1
def record_cache_miss(self) -> None:
"""记录缓存未命中"""
metrics = self.metrics.cache_metrics
metrics.total_accesses += 1
metrics.cache_misses += 1
def record_cache_invalidation(self) -> None:
"""记录缓存失效"""
self.metrics.cache_metrics.cache_invalidations += 1
def _check_performance_alert(self, elapsed_time: float) -> None:
"""检查性能告警
如果权限检查耗时过长,发出告警。
"""
if elapsed_time > 0.05: # 50ms
self._add_alert(
level="warning",
type="slow_permission_check",
message=f"权限检查耗时过长: {elapsed_time*1000:.2f}ms",
data={"elapsed_time": elapsed_time},
)
def _check_error_alert(self, error: Exception) -> None:
"""检查错误告警"""
self._add_alert(
level="error",
type="permission_check_error",
message=f"权限检查出错: {str(error)}",
data={"error": str(error)},
)
def _add_alert(
self,
level: str,
type: str,
message: str,
data: dict[str, Any] | None = None,
) -> None:
"""添加告警
Args:
level: 告警级别 (info, warning, error)
type: 告警类型
message: 告警消息
data: 额外数据
"""
alert = {
"timestamp": time.time(),
"level": level,
"type": type,
"message": message,
"data": data or {},
}
self._alerts.append(alert)
# 只保留最近1000条告警
if len(self._alerts) > 1000:
self._alerts = self._alerts[-1000:]
def get_alerts(self, level: str | None = None, limit: int = 100) -> list[dict[str, Any]]:
"""获取告警列表
Args:
level: 告警级别过滤(可选)
limit: 返回的最大告警数
Returns:
告警列表
"""
alerts = self._alerts
if level:
alerts = [a for a in alerts if a["level"] == level]
return alerts[-limit:]
def get_metrics(self) -> dict[str, Any]:
"""获取指标"""
return self.metrics.to_dict()
def get_cache_stats(self) -> dict[str, Any]:
"""获取缓存统计"""
cache = get_permission_cache()
return {
"cache_items": cache.get_cache_stats(),
"cache_metrics": self.metrics.cache_metrics.to_dict(),
}
def reset_metrics(self) -> None:
"""重置指标"""
self.metrics.reset()
def clear_alerts(self) -> None:
"""清除所有告警"""
self._alerts.clear()
# 全局监控器实例
_permission_monitor: PermissionMonitor | None = None
def get_permission_monitor() -> PermissionMonitor:
"""获取全局权限监控器实例"""
global _permission_monitor
if _permission_monitor is None:
_permission_monitor = PermissionMonitor()
return _permission_monitor
def set_permission_monitor(monitor: PermissionMonitor) -> None:
"""设置全局权限监控器实例(用于测试)"""
global _permission_monitor
_permission_monitor = monitor
def evaluate_permission_system_health(metrics: dict[str, Any], cache_stats: dict[str, Any]) -> dict[str, Any]:
"""根据权限系统指标评估健康状态"""
check_metrics = metrics["check_metrics"]
cache_metrics = metrics["cache_metrics"]
health_score = 100
issues = []
if check_metrics["error_rate"] > 1:
health_score -= 20
issues.append("权限检查错误率过高")
if (
cache_metrics["total_accesses"] >= CACHE_HIT_RATE_HEALTH_MIN_ACCESSES
and cache_metrics["hit_rate"] < 50
):
health_score -= 10
issues.append("缓存命中率过低")
if check_metrics["avg_time"] > 0.01:
health_score -= 10
issues.append("权限检查响应时间过长")
if check_metrics["deny_rate"] > 50:
health_score -= 5
issues.append("权限拒绝率过高")
return {
"status": "healthy" if health_score >= 80 else "degraded" if health_score >= 50 else "unhealthy",
"health_score": max(0, health_score),
"issues": issues,
"metrics": metrics,
"cache_stats": cache_stats,
}