修复(权限管理)：统一 PM 系统导航与权限校验

backend/app/api/v1/permission_monitoring.py

@@ -602,6 +602,7 @@ async def get_access_logs(
     end_time: Optional[datetime] = Query(None),
     client_ip: Optional[str] = Query(None),
     client_type: Optional[str] = Query(None),
+    event_type: str = Query("all", pattern="^(all|permission)$"),
     keyword: Optional[str] = Query(None),
     page: int = Query(1, ge=1),
     page_size: int = Query(50, ge=1, le=200),
@@ -660,7 +661,14 @@ async def get_access_logs(
             keyword_conditions.append(PermissionAccessLog.ip_address.in_(location_ips))
         conditions.append(or_(*keyword_conditions))
 
-    include_security_logs = scope.is_admin and study_id is None and role is None and endpoint_key is None and allowed is not True
+    include_security_logs = (
+        event_type != "permission"
+        and scope.is_admin
+        and study_id is None
+        and role is None
+        and endpoint_key is None
+        and allowed is not True
+    )
     event_conditions = list(conditions)
     if include_security_logs:
         matching_security_event = select(SecurityAccessLog.id).where(
@@ -1118,81 +1126,352 @@ async def get_trends(
     _=Depends(get_current_user),
     period: str = Query("24h", pattern="^(24h|7d|30d)$"),
 ) -> dict:
-    """获取趋势数据（从快照表或实时聚合）"""
+    """获取权限检查趋势、周期摘要和归因数据。"""
     scope = await resolve_monitoring_scope(db, _)
     if not scope.is_admin and not scope.study_ids:
         raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail="权限不足")
     now = datetime.now(timezone.utc)
-    period_map = {"24h": timedelta(hours=24), "7d": timedelta(days=7), "30d": timedelta(days=30)}
-    start_time = now - period_map[period]
+    bucket_delta, bucket_count = _trend_bucket_config(period)
+    current_bucket = _trend_bucket_time(now, period)
+    start_time = current_bucket - bucket_delta * (bucket_count - 1)
+    current_duration = now - start_time
+    previous_end = start_time
+    previous_start = previous_end - current_duration
 
-    # 先尝试从快照表获取
-    snapshot_query = (
-        select(PermissionMetricSnapshot)
-        .where(PermissionMetricSnapshot.bucket_time >= start_time)
-        .order_by(PermissionMetricSnapshot.bucket_time)
+    log_buckets = await _load_trend_log_buckets(db, scope, start_time, now, period)
+    cache_buckets, current_cache, previous_cache = await _load_trend_cache_buckets(
+        db,
+        scope,
+        previous_start,
+        previous_end,
+        start_time,
+        now,
+        period,
     )
-    result = await db.execute(snapshot_query)
-    snapshots = result.scalars().all()
+    summary = await _load_trend_summary(db, scope, start_time, now)
+    previous_summary = await _load_trend_summary(db, scope, previous_start, previous_end)
+    summary.update(current_cache)
+    previous_summary.update(previous_cache)
+    previous_summary.pop("observed_end_at", None)
 
-    if snapshots and scope.is_admin:
-        return {
-            "period": period,
-            "data_points": [
-                {
-                    "bucket_time": s.bucket_time.isoformat(),
-                    "total_checks": s.total_checks,
-                    "allowed_checks": s.allowed_checks,
-                    "denied_checks": s.denied_checks,
-                    "avg_elapsed_ms": round(s.avg_elapsed_ms, 2),
-                    "max_elapsed_ms": round(s.max_elapsed_ms, 2),
-                    "cache_hits": s.cache_hits,
-                    "cache_misses": s.cache_misses,
-                    "cache_hit_rate": round(
-                        s.cache_hits / (s.cache_hits + s.cache_misses) * 100, 1
-                    ) if (s.cache_hits + s.cache_misses) > 0 else 0,
-                    "error_count": s.error_count,
-                }
-                for s in snapshots
-            ],
-        }
-
-    # 如果没有快照数据，从原始日志实时聚合（适用于刚部署时）。
-    # 这里使用 Python 分桶，避免 SQLite 测试库不支持 PostgreSQL date_trunc。
-    trend_query = select(
-        PermissionAccessLog.created_at,
-        PermissionAccessLog.allowed,
-        PermissionAccessLog.elapsed_ms,
-    ).where(PermissionAccessLog.created_at >= start_time)
-    result = await db.execute(
-        _apply_monitoring_scope_to_log_query(trend_query, scope)
-    )
-    buckets: dict[datetime, dict[str, float | int]] = defaultdict(
-        lambda: {"total": 0, "allowed": 0, "denied": 0, "elapsed_sum": 0.0, "max_ms": 0.0}
-    )
-    for created_at, allowed, elapsed_ms in result.all():
-        bucket = _trend_bucket_time(created_at, period)
-        buckets[bucket]["total"] += 1
-        buckets[bucket]["allowed" if allowed else "denied"] += 1
-        buckets[bucket]["elapsed_sum"] += float(elapsed_ms or 0)
-        buckets[bucket]["max_ms"] = max(float(buckets[bucket]["max_ms"]), float(elapsed_ms or 0))
+    data_points = []
+    for index in range(bucket_count):
+        bucket_time = start_time + bucket_delta * index
+        values = log_buckets.get(bucket_time, _empty_trend_bucket())
+        cache_values = cache_buckets.get(bucket_time, {"cache_hits": 0, "cache_misses": 0})
+        cache_total = int(cache_values["cache_hits"]) + int(cache_values["cache_misses"])
+        total = int(values["total"])
+        data_points.append(
+            {
+                "bucket_time": bucket_time.isoformat(),
+                "sample_state": "partial" if bucket_time == current_bucket else "complete",
+                "total_checks": total,
+                "allowed_checks": int(values["allowed"]),
+                "denied_checks": int(values["denied"]),
+                "avg_elapsed_ms": round(float(values["elapsed_sum"]) / total, 2) if total else 0,
+                "max_elapsed_ms": round(float(values["max_ms"]), 2),
+                "cache_hits": int(cache_values["cache_hits"]),
+                "cache_misses": int(cache_values["cache_misses"]),
+                "cache_hit_rate": round(int(cache_values["cache_hits"]) / cache_total * 100, 1) if cache_total else 0,
+                "cache_sample_available": cache_total > 0,
+                "error_count": 0,
+            }
+        )
 
     return {
         "period": period,
-        "data_points": [
+        "generated_at": now.isoformat(),
+        "range": {
+            "start_at": start_time.isoformat(),
+            "end_at": now.isoformat(),
+            "previous_start_at": previous_start.isoformat(),
+            "previous_end_at": previous_end.isoformat(),
+            "bucket_seconds": int(bucket_delta.total_seconds()),
+            "bucket_count": bucket_count,
+            "timezone": "UTC",
+            "observed_end_at": summary.pop("observed_end_at"),
+        },
+        "summary": summary,
+        "previous_summary": previous_summary,
+        "attribution": await _load_trend_attribution(db, scope, start_time, now),
+        "data_points": data_points,
+    }
+
+
+def _trend_bucket_config(period: str) -> tuple[timedelta, int]:
+    return {
+        "24h": (timedelta(hours=1), 24),
+        "7d": (timedelta(hours=6), 28),
+        "30d": (timedelta(days=1), 30),
+    }[period]
+
+
+def _empty_trend_bucket() -> dict[str, float | int]:
+    return {"total": 0, "allowed": 0, "denied": 0, "elapsed_sum": 0.0, "max_ms": 0.0}
+
+
+def _merge_trend_bucket(target: dict[str, float | int], *, total: int, allowed: int, denied: int, elapsed_sum: float, max_ms: float) -> None:
+    target["total"] = int(target["total"]) + total
+    target["allowed"] = int(target["allowed"]) + allowed
+    target["denied"] = int(target["denied"]) + denied
+    target["elapsed_sum"] = float(target["elapsed_sum"]) + elapsed_sum
+    target["max_ms"] = max(float(target["max_ms"]), max_ms)
+
+
+async def _load_trend_log_buckets(
+    db: AsyncSession,
+    scope: MonitoringScope,
+    start_time: datetime,
+    end_time: datetime,
+    period: str,
+) -> dict[datetime, dict[str, float | int]]:
+    """PostgreSQL 先按小时聚合，SQLite 测试库使用 Python 分桶。"""
+    buckets: dict[datetime, dict[str, float | int]] = defaultdict(_empty_trend_bucket)
+    dialect_name = db.get_bind().dialect.name
+    if dialect_name == "postgresql":
+        hour_bucket = func.date_trunc("hour", PermissionAccessLog.created_at).label("hour_bucket")
+        query = select(
+            hour_bucket,
+            func.count().label("total"),
+            func.count().filter(PermissionAccessLog.allowed.is_(True)).label("allowed"),
+            func.count().filter(PermissionAccessLog.allowed.is_(False)).label("denied"),
+            func.coalesce(func.sum(PermissionAccessLog.elapsed_ms), 0).label("elapsed_sum"),
+            func.coalesce(func.max(PermissionAccessLog.elapsed_ms), 0).label("max_ms"),
+        ).where(
+            PermissionAccessLog.created_at >= start_time,
+            PermissionAccessLog.created_at <= end_time,
+        ).group_by(hour_bucket)
+        result = await db.execute(_apply_monitoring_scope_to_log_query(query, scope))
+        for row in result.all():
+            bucket_time = _trend_bucket_time(row.hour_bucket, period)
+            _merge_trend_bucket(
+                buckets[bucket_time],
+                total=int(row.total or 0),
+                allowed=int(row.allowed or 0),
+                denied=int(row.denied or 0),
+                elapsed_sum=float(row.elapsed_sum or 0),
+                max_ms=float(row.max_ms or 0),
+            )
+        return buckets
+
+    query = select(
+        PermissionAccessLog.created_at,
+        PermissionAccessLog.allowed,
+        PermissionAccessLog.elapsed_ms,
+    ).where(
+        PermissionAccessLog.created_at >= start_time,
+        PermissionAccessLog.created_at <= end_time,
+    )
+    result = await db.execute(_apply_monitoring_scope_to_log_query(query, scope))
+    for created_at, allowed, elapsed_ms in result.all():
+        bucket_time = _trend_bucket_time(created_at, period)
+        _merge_trend_bucket(
+            buckets[bucket_time],
+            total=1,
+            allowed=1 if allowed else 0,
+            denied=0 if allowed else 1,
+            elapsed_sum=float(elapsed_ms or 0),
+            max_ms=float(elapsed_ms or 0),
+        )
+    return buckets
+
+
+def _percentile(values: list[float], percentile: float) -> float:
+    if not values:
+        return 0.0
+    ordered = sorted(values)
+    position = (len(ordered) - 1) * percentile
+    lower = int(position)
+    upper = min(lower + 1, len(ordered) - 1)
+    fraction = position - lower
+    return ordered[lower] + (ordered[upper] - ordered[lower]) * fraction
+
+
+async def _load_trend_summary(
+    db: AsyncSession,
+    scope: MonitoringScope,
+    start_time: datetime,
+    end_time: datetime,
+) -> dict:
+    query = select(
+        func.count().label("total"),
+        func.count().filter(PermissionAccessLog.allowed.is_(True)).label("allowed"),
+        func.count().filter(PermissionAccessLog.allowed.is_(False)).label("denied"),
+        func.count().filter(PermissionAccessLog.elapsed_ms > 50).label("slow"),
+        func.count(func.distinct(PermissionAccessLog.study_id)).label("studies"),
+        func.count(func.distinct(PermissionAccessLog.user_id)).label("users"),
+        func.count(func.distinct(PermissionAccessLog.endpoint_key)).label("endpoints"),
+        func.coalesce(func.avg(PermissionAccessLog.elapsed_ms), 0).label("avg_ms"),
+        func.coalesce(func.max(PermissionAccessLog.elapsed_ms), 0).label("max_ms"),
+        func.max(PermissionAccessLog.created_at).label("observed_end_at"),
+    ).where(
+        PermissionAccessLog.created_at >= start_time,
+        PermissionAccessLog.created_at <= end_time,
+    )
+    row = (await db.execute(_apply_monitoring_scope_to_log_query(query, scope))).one()
+    if db.get_bind().dialect.name == "postgresql":
+        percentile_query = select(
+            func.percentile_cont(0.95).within_group(PermissionAccessLog.elapsed_ms)
+        ).where(
+            PermissionAccessLog.created_at >= start_time,
+            PermissionAccessLog.created_at <= end_time,
+        )
+        percentile_result = await db.execute(
+            _apply_monitoring_scope_to_log_query(percentile_query, scope)
+        )
+        p95_ms = float(percentile_result.scalar() or 0)
+    else:
+        elapsed_query = select(PermissionAccessLog.elapsed_ms).where(
+            PermissionAccessLog.created_at >= start_time,
+            PermissionAccessLog.created_at <= end_time,
+        )
+        elapsed_result = await db.execute(
+            _apply_monitoring_scope_to_log_query(elapsed_query, scope)
+        )
+        p95_ms = _percentile(
+            [float(value) for value in elapsed_result.scalars().all()],
+            0.95,
+        )
+    total = int(row.total or 0)
+    denied = int(row.denied or 0)
+    return {
+        "total_checks": total,
+        "allowed_checks": int(row.allowed or 0),
+        "denied_checks": denied,
+        "deny_rate": round(denied / total * 100, 2) if total else 0,
+        "avg_elapsed_ms": round(float(row.avg_ms or 0), 2),
+        "p95_elapsed_ms": round(p95_ms, 2),
+        "max_elapsed_ms": round(float(row.max_ms or 0), 2),
+        "slow_check_count": int(row.slow or 0),
+        "active_study_count": int(row.studies or 0),
+        "active_user_count": int(row.users or 0),
+        "active_endpoint_count": int(row.endpoints or 0),
+        "cache_hits": 0,
+        "cache_misses": 0,
+        "cache_hit_rate": 0,
+        "cache_sample_available": False,
+        "observed_end_at": row.observed_end_at.isoformat() if row.observed_end_at else None,
+    }
+
+
+async def _load_trend_cache_buckets(
+    db: AsyncSession,
+    scope: MonitoringScope,
+    previous_start: datetime,
+    previous_end: datetime,
+    current_start: datetime,
+    current_end: datetime,
+    period: str,
+) -> tuple[dict[datetime, dict[str, int]], dict, dict]:
+    empty_summary = {"cache_hits": 0, "cache_misses": 0, "cache_hit_rate": 0, "cache_sample_available": False}
+    if not scope.is_admin:
+        return {}, dict(empty_summary), dict(empty_summary)
+    result = await db.execute(
+        select(PermissionMetricSnapshot).where(
+            PermissionMetricSnapshot.bucket_time >= previous_start,
+            PermissionMetricSnapshot.bucket_time <= current_end,
+        )
+    )
+    buckets: dict[datetime, dict[str, int]] = defaultdict(lambda: {"cache_hits": 0, "cache_misses": 0})
+    current_counts = {"cache_hits": 0, "cache_misses": 0}
+    previous_counts = {"cache_hits": 0, "cache_misses": 0}
+    for snapshot in result.scalars().all():
+        snapshot_time = snapshot.bucket_time
+        if snapshot_time.tzinfo is None:
+            snapshot_time = snapshot_time.replace(tzinfo=timezone.utc)
+        if snapshot_time >= current_start:
+            current_counts["cache_hits"] += int(snapshot.cache_hits or 0)
+            current_counts["cache_misses"] += int(snapshot.cache_misses or 0)
+            bucket = buckets[_trend_bucket_time(snapshot_time, period)]
+            bucket["cache_hits"] += int(snapshot.cache_hits or 0)
+            bucket["cache_misses"] += int(snapshot.cache_misses or 0)
+        elif snapshot_time < previous_end:
+            previous_counts["cache_hits"] += int(snapshot.cache_hits or 0)
+            previous_counts["cache_misses"] += int(snapshot.cache_misses or 0)
+
+    def serialize(counts: dict[str, int]) -> dict:
+        total = counts["cache_hits"] + counts["cache_misses"]
+        return {
+            **counts,
+            "cache_hit_rate": round(counts["cache_hits"] / total * 100, 2) if total else 0,
+            "cache_sample_available": total > 0,
+        }
+
+    return dict(buckets), serialize(current_counts), serialize(previous_counts)
+
+
+async def _load_trend_attribution(
+    db: AsyncSession,
+    scope: MonitoringScope,
+    start_time: datetime,
+    end_time: datetime,
+) -> dict:
+    base_conditions = (
+        PermissionAccessLog.created_at >= start_time,
+        PermissionAccessLog.created_at <= end_time,
+    )
+    denied_query = select(
+        PermissionAccessLog.endpoint_key,
+        PermissionAccessLog.role,
+        func.count().label("denied_count"),
+    ).where(*base_conditions, PermissionAccessLog.allowed.is_(False)).group_by(
+        PermissionAccessLog.endpoint_key,
+        PermissionAccessLog.role,
+    ).order_by(desc("denied_count")).limit(5)
+    denied_rows = (await db.execute(_apply_monitoring_scope_to_log_query(denied_query, scope))).all()
+
+    slow_query = select(
+        PermissionAccessLog.endpoint_key,
+        func.count().label("sample_count"),
+        func.count().filter(PermissionAccessLog.elapsed_ms > 50).label("slow_count"),
+        func.coalesce(func.avg(PermissionAccessLog.elapsed_ms), 0).label("avg_ms"),
+        func.coalesce(func.max(PermissionAccessLog.elapsed_ms), 0).label("max_ms"),
+    ).where(*base_conditions).group_by(PermissionAccessLog.endpoint_key).order_by(
+        desc("slow_count"), desc("avg_ms"), desc("sample_count")
+    ).limit(5)
+    slow_rows = (await db.execute(_apply_monitoring_scope_to_log_query(slow_query, scope))).all()
+
+    client_type = func.coalesce(PermissionAccessLog.client_type, "unknown")
+    client_version = func.coalesce(PermissionAccessLog.client_version, "-")
+    client_platform = func.coalesce(PermissionAccessLog.client_platform, "-")
+    client_query = select(
+        client_type.label("client_type"),
+        client_version.label("client_version"),
+        client_platform.label("client_platform"),
+        func.count().label("total_count"),
+        func.count().filter(PermissionAccessLog.allowed.is_(False)).label("denied_count"),
+        func.coalesce(func.avg(PermissionAccessLog.elapsed_ms), 0).label("avg_ms"),
+    ).where(*base_conditions).group_by(client_type, client_version, client_platform).order_by(
+        desc("total_count")
+    ).limit(5)
+    client_rows = (await db.execute(_apply_monitoring_scope_to_log_query(client_query, scope))).all()
+
+    return {
+        "top_denied": [
+            {"endpoint_key": row.endpoint_key, "role": row.role, "denied_count": int(row.denied_count)}
+            for row in denied_rows
+        ],
+        "top_slow": [
             {
-                "bucket_time": bucket.isoformat(),
-                "total_checks": values["total"],
-                "allowed_checks": values["allowed"],
-                "denied_checks": values["denied"],
-                "avg_elapsed_ms": round(float(values["elapsed_sum"]) / int(values["total"]), 2),
-                "max_elapsed_ms": round(float(values["max_ms"]), 2),
-                "cache_hits": 0,
-                "cache_misses": 0,
-                "cache_hit_rate": 0,
-                "error_count": 0,
+                "endpoint_key": row.endpoint_key,
+                "sample_count": int(row.sample_count),
+                "slow_count": int(row.slow_count),
+                "avg_elapsed_ms": round(float(row.avg_ms), 2),
+                "max_elapsed_ms": round(float(row.max_ms), 2),
             }
-            for bucket, values in sorted(buckets.items())
+            for row in slow_rows
+        ],
+        "client_breakdown": [
+            {
+                "client_type": row.client_type,
+                "client_version": row.client_version,
+                "client_platform": row.client_platform,
+                "total_count": int(row.total_count),
+                "denied_count": int(row.denied_count),
+                "deny_rate": round(int(row.denied_count) / int(row.total_count) * 100, 2) if row.total_count else 0,
+                "avg_elapsed_ms": round(float(row.avg_ms), 2),
+            }
+            for row in client_rows
         ],
     }