Files
ctms/backend/app/core/permission_monitor.py
2026-06-30 10:43:24 +08:00

205 lines
6.6 KiB
Python

"""权限系统监控
收集缓存性能指标和告警信息。
权限检查的统计数据(总次数、允许/拒绝、耗时)已持久化在
permission_access_logs 表中,通过 /metrics 端点实时聚合查询。
"""
from __future__ import annotations
import time
from collections import deque
from dataclasses import dataclass, field
from typing import Any
from app.core.permission_cache import get_permission_cache
CACHE_HIT_RATE_HEALTH_MIN_ACCESSES = 50
CACHE_METRICS_WINDOW_SECONDS = 3600
@dataclass
class CacheMetrics:
"""缓存指标"""
total_accesses: int = 0
cache_hits: int = 0
cache_misses: int = 0
cache_invalidations: int = 0
_access_events: deque[tuple[float, str, bool]] = field(
default_factory=deque,
repr=False,
)
_invalidation_events: deque[float] = field(default_factory=deque, repr=False)
@property
def hit_rate(self) -> float:
if self.total_accesses == 0:
return 0.0
return (self.cache_hits / self.total_accesses) * 100
@property
def miss_rate(self) -> float:
if self.total_accesses == 0:
return 0.0
return (self.cache_misses / self.total_accesses) * 100
def to_dict(self) -> dict[str, Any]:
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),
}
def record_access(self, cache_name: str, *, hit: bool) -> None:
self.total_accesses += 1
if hit:
self.cache_hits += 1
else:
self.cache_misses += 1
self._access_events.append((time.time(), cache_name, hit))
def record_invalidation(self) -> None:
self.cache_invalidations += 1
self._invalidation_events.append(time.time())
def window_stats(self, seconds: int = CACHE_METRICS_WINDOW_SECONDS) -> dict[str, Any]:
cutoff = time.time() - seconds
while self._access_events and self._access_events[0][0] < cutoff:
self._access_events.popleft()
while self._invalidation_events and self._invalidation_events[0] < cutoff:
self._invalidation_events.popleft()
def summarize(cache_name: str | None = None) -> dict[str, Any]:
events = [
event for event in self._access_events
if cache_name is None or event[1] == cache_name
]
hits = sum(event[2] for event in events)
total = len(events)
misses = total - hits
return {
"total_accesses": total,
"cache_hits": hits,
"cache_misses": misses,
"cache_invalidations": len(self._invalidation_events),
"hit_rate": round(hits / total * 100, 2) if total else 0.0,
"miss_rate": round(misses / total * 100, 2) if total else 0.0,
}
stats = summarize()
stats.update({
"window_seconds": seconds,
"scope": "sliding_window",
"by_cache": {
"project_permissions": summarize("project_permissions"),
"member_roles": summarize("member_roles"),
},
})
return stats
@dataclass
class PermissionSystemMetrics:
cache_metrics: CacheMetrics = field(default_factory=CacheMetrics)
start_time: float = field(default_factory=time.time)
@property
def uptime_seconds(self) -> float:
return time.time() - self.start_time
def reset(self) -> None:
self.cache_metrics = CacheMetrics()
self.start_time = time.time()
class PermissionMonitor:
"""权限系统监控器(仅缓存指标 + 告警)"""
def __init__(self):
self.metrics = PermissionSystemMetrics()
self._alerts: list[dict[str, Any]] = []
def record_cache_hit(self, cache_name: str = "project_permissions") -> None:
self.metrics.cache_metrics.record_access(cache_name, hit=True)
def record_cache_miss(self, cache_name: str = "project_permissions") -> None:
self.metrics.cache_metrics.record_access(cache_name, hit=False)
def record_cache_invalidation(self) -> None:
self.metrics.cache_metrics.record_invalidation()
def record_error_alert(self, error: Exception) -> None:
self._add_alert(
level="error",
type="permission_check_error",
message=f"权限检查出错: {str(error)}",
data={"error": str(error)},
)
def record_slow_check_alert(self, elapsed_ms: float) -> None:
if elapsed_ms > 50:
self._add_alert(
level="warning",
type="slow_permission_check",
message=f"权限检查耗时过长: {elapsed_ms:.2f}ms",
data={"elapsed_ms": elapsed_ms},
)
def _add_alert(self, level: str, type: str, message: str, data: dict[str, Any] | None = None) -> None:
self._alerts.append({
"timestamp": time.time(),
"level": level,
"type": type,
"message": message,
"data": data or {},
})
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]]:
alerts = self._alerts
if level:
alerts = [a for a in alerts if a["level"] == level]
return alerts[-limit:]
def get_cache_stats(self) -> dict[str, Any]:
cache = get_permission_cache()
return {
"cache_items": cache.get_cache_stats(),
"cache_metrics": self.metrics.cache_metrics.window_stats(),
"cache_metrics_lifetime": self.metrics.cache_metrics.to_dict(),
}
def get_metrics(self) -> dict[str, Any]:
return {
"cache_metrics": self.metrics.cache_metrics.window_stats(),
"cache_metrics_lifetime": self.metrics.cache_metrics.to_dict(),
"uptime_seconds": self.metrics.uptime_seconds,
}
def reset_metrics(self) -> None:
self.metrics.reset()
def clear_alerts(self) -> None:
self._alerts.clear()
_permission_monitor: PermissionMonitor | None = None
def get_permission_monitor() -> PermissionMonitor:
global _permission_monitor
if _permission_monitor is None:
_permission_monitor = PermissionMonitor()
return _permission_monitor
def set_permission_monitor(monitor: PermissionMonitor) -> None:
global _permission_monitor
_permission_monitor = monitor