@agentunion/kite 1.4.0 → 1.5.0

package/extensions/services/proxy/agentcp/monitoring/sliding_window.py

@@ -0,0 +1,222 @@
+# -*- coding: utf-8 -*-
+"""
+滑动窗口指标统计
+
+支持多时间窗口的实时统计分析
+"""
+
+import time
+from typing import Dict, List, Tuple, Any
+
+
+class TimeWindow:
+    """单个时间窗口
+
+    维护一个固定时长的滑动窗口，自动清理过期数据点
+    """
+
+    def __init__(self, duration_seconds: int):
+        """初始化时间窗口
+
+        Args:
+            duration_seconds: 窗口时长（秒）
+        """
+        self.duration = duration_seconds
+        self.data_points: List[Tuple[float, dict]] = []  # [(timestamp, metrics), ...]
+
+    def add_snapshot(self, timestamp: float, metrics: dict):
+        """添加一个时间点的指标快照
+
+        Args:
+            timestamp: 时间戳
+            metrics: 指标字典，必须包含以下字段：
+                - received_delta: 增量接收消息数
+                - success_delta: 增量成功派发数
+                - failed_delta: 增量失败数
+                - avg_latency: 平均延迟（毫秒）
+                - queue_size: 当前队列大小
+        """
+        self.data_points.append((timestamp, metrics))
+
+        # 清理过期数据（保留窗口时长内的数据）
+        cutoff = timestamp - self.duration
+        self.data_points = [(t, m) for t, m in self.data_points if t >= cutoff]
+
+    def get_stats(self) -> Dict[str, Any]:
+        """计算窗口内的统计数据
+
+        Returns:
+            包含以下统计指标的字典：
+            - throughput_per_second: 吞吐量（消息/秒）
+            - avg_latency_ms: 平均延迟（毫秒）
+            - success_rate: 成功率（%）
+            - total_messages: 窗口内总消息数
+            - failed_messages: 窗口内失败消息数
+            - avg_queue_size: 平均队列大小
+            - window_duration: 窗口时长（秒）
+            - data_points_count: 数据点数量
+        """
+        if not self.data_points:
+            return {
+                'throughput_per_second': 0.0,
+                'avg_latency_ms': 0.0,
+                'success_rate': 0.0,
+                'total_messages': 0,
+                'failed_messages': 0,
+                'avg_queue_size': 0,
+                'window_duration': self.duration,
+                'data_points_count': 0,
+            }
+
+        # 计算总量
+        total_received = sum(m.get('received_delta', 0) for _, m in self.data_points)
+        total_success = sum(m.get('success_delta', 0) for _, m in self.data_points)
+        total_failed = sum(m.get('failed_delta', 0) for _, m in self.data_points)
+
+        # 计算实际时间跨度（可能小于窗口时长）
+        actual_duration = self.data_points[-1][0] - self.data_points[0][0]
+        if actual_duration < 1:
+            actual_duration = 1  # 避免除零
+
+        # 计算吞吐量 (msg/s)
+        throughput = total_received / actual_duration
+
+        # 计算平均延迟（只考虑有延迟数据的点）
+        latencies = [m.get('avg_latency', 0) for _, m in self.data_points if m.get('avg_latency', 0) > 0]
+        avg_latency = sum(latencies) / len(latencies) if latencies else 0.0
+
+        # 计算成功率
+        success_rate = (total_success / max(total_received, 1)) * 100
+
+        # 计算平均队列大小
+        queue_sizes = [m.get('queue_size', 0) for _, m in self.data_points]
+        avg_queue_size = sum(queue_sizes) / len(queue_sizes) if queue_sizes else 0
+
+        return {
+            'throughput_per_second': round(throughput, 2),
+            'avg_latency_ms': round(avg_latency, 2),
+            'success_rate': round(success_rate, 2),
+            'total_messages': total_received,
+            'failed_messages': total_failed,
+            'avg_queue_size': round(avg_queue_size, 1),
+            'window_duration': self.duration,
+            'data_points_count': len(self.data_points),
+        }
+
+
+class SlidingWindowMetrics:
+    """多时间窗口管理器
+
+    管理多个不同时长的滑动窗口，提供多粒度的统计视图
+    支持的窗口：1分钟、3分钟、5分钟、10分钟、15分钟
+    """
+
+    def __init__(self):
+        """初始化多时间窗口管理器"""
+        # 创建多个时间窗口
+        self.windows: Dict[str, TimeWindow] = {
+            '1m': TimeWindow(60),      # 1分钟
+            '3m': TimeWindow(180),     # 3分钟
+            '5m': TimeWindow(300),     # 5分钟
+            '10m': TimeWindow(600),    # 10分钟
+            '15m': TimeWindow(900),    # 15分钟
+        }
+
+        # 记录上一次快照的状态
+        self.last_snapshot_time = 0
+        self.last_metrics_snapshot: Dict[str, Any] = {}
+
+    def update(self, current_metrics: dict):
+        """更新所有窗口（应该每10秒调用一次）
+
+        Args:
+            current_metrics: 当前累计指标，必须包含以下字段：
+                - received_total: 累计接收消息总数
+                - dispatched_success: 累计派发成功数
+                - dispatched_failed: 累计派发失败数
+                - dispatch_queue_size: 当前派发队列大小
+                - avg_dispatch_latency_ms: 平均派发延迟（可选）
+                - avg_handler_latency_ms: 平均处理延迟（可选）
+        """
+        now = time.time()
+
+        # 计算增量指标 (delta)
+        metrics_delta = self._calculate_delta(current_metrics)
+
+        # 更新所有窗口
+        for window in self.windows.values():
+            window.add_snapshot(now, metrics_delta)
+
+        # 保存当前快照状态
+        self.last_snapshot_time = now
+        self.last_metrics_snapshot = current_metrics.copy()
+
+    def _calculate_delta(self, current: dict) -> dict:
+        """计算两次快照之间的增量
+
+        Args:
+            current: 当前累计指标
+
+        Returns:
+            增量指标字典
+        """
+        if not self.last_metrics_snapshot:
+            # 第一次快照，返回零增量
+            return {
+                'received_delta': 0,
+                'success_delta': 0,
+                'failed_delta': 0,
+                'avg_latency': 0.0,
+                'queue_size': current.get('dispatch_queue_size', 0),
+            }
+
+        # 计算增量
+        received_delta = current.get('received_total', 0) - self.last_metrics_snapshot.get('received_total', 0)
+        success_delta = current.get('dispatched_success', 0) - self.last_metrics_snapshot.get('dispatched_success', 0)
+        failed_delta = current.get('dispatched_failed', 0) - self.last_metrics_snapshot.get('dispatched_failed', 0)
+
+        # 获取平均延迟（优先使用 dispatch_latency，其次使用 handler_latency）
+        avg_latency = 0.0
+        if 'avg_dispatch_latency_ms' in current:
+            avg_latency = current['avg_dispatch_latency_ms']
+        elif 'avg_handler_latency_ms' in current:
+            avg_latency = current['avg_handler_latency_ms']
+
+        return {
+            'received_delta': max(0, received_delta),  # 防止负数
+            'success_delta': max(0, success_delta),
+            'failed_delta': max(0, failed_delta),
+            'avg_latency': avg_latency,
+            'queue_size': current.get('dispatch_queue_size', 0),
+        }
+
+    def get_window_stats(self, window_name: str) -> Dict[str, Any]:
+        """获取指定窗口的统计数据
+
+        Args:
+            window_name: 窗口名称（'1m', '3m', '5m', '10m', '15m'）
+
+        Returns:
+            窗口统计数据字典，如果窗口不存在则返回空字典
+        """
+        if window_name not in self.windows:
+            return {}
+        return self.windows[window_name].get_stats()
+
+    def get_all_windows(self) -> Dict[str, Dict[str, Any]]:
+        """获取所有窗口的统计数据
+
+        Returns:
+            字典，key为窗口名称，value为统计数据
+        """
+        return {
+            name: window.get_stats()
+            for name, window in self.windows.items()
+        }
+
+    def reset(self):
+        """重置所有窗口数据"""
+        for window in self.windows.values():
+            window.data_points.clear()
+        self.last_snapshot_time = 0
+        self.last_metrics_snapshot.clear()

package/extensions/services/proxy/agentcp/monitoring/standalone_reader.py

@@ -0,0 +1,224 @@
+# -*- coding: utf-8 -*-
+"""
+独立监控数据读取器
+
+支持跨进程读取监控数据，无需依赖全局变量或进程内通信
+"""
+
+import os
+import time
+from typing import Dict, Any, Optional, List
+
+from .metrics_store import MetricsStore
+from .sliding_window import SlidingWindowMetrics
+
+
+class StandaloneMonitoringReader:
+    """独立监控数据读取器
+
+    可以在任何进程中使用，通过直接读取 SQLite 数据库获取监控数据
+    """
+
+    def __init__(self, db_path: str = None):
+        """初始化读取器
+
+        Args:
+            db_path: 数据库路径，如果为空则使用默认路径
+        """
+        if db_path is None:
+            # 使用默认路径
+            project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), '..', '..'))
+            db_path = os.path.join(project_root, 'backup', 'metrics_timeseries.db')
+
+        self.db_path = db_path
+        self.metrics_store = MetricsStore(db_path)
+
+    def get_realtime_metrics(self) -> Dict[str, Any]:
+        """获取实时监控指标
+
+        Returns:
+            包含累计指标和时间窗口统计的字典
+        """
+        # 检查数据库文件是否存在
+        if not os.path.exists(self.db_path):
+            raise FileNotFoundError(f"监控数据库不存在: {self.db_path}")
+
+        # 获取最新的一条记录作为累计指标
+        latest_records = self.metrics_store.query_latest(limit=1)
+        if not latest_records:
+            raise ValueError("数据库中没有监控数据")
+
+        cumulative = latest_records[0]
+
+        # 计算时间窗口统计
+        windows = self._calculate_windows()
+
+        return {
+            'agent_id': cumulative.get('agent_id', 'unknown'),
+            'timestamp': cumulative.get('timestamp', time.time()),
+            'cumulative': self._format_cumulative(cumulative),
+            'windows': windows
+        }
+
+    def _format_cumulative(self, record: dict) -> dict:
+        """格式化累计指标"""
+        # 计算成功率
+        received_total = record.get('received_total', 0)
+        dispatched_success = record.get('dispatched_success', 0)
+
+        dispatch_success_rate = "0.00%"
+        if received_total > 0:
+            rate = (dispatched_success / received_total) * 100
+            dispatch_success_rate = f"{rate:.2f}%"
+
+        # 计算运行时长（使用最早的记录时间）
+        store_stats = self.metrics_store.get_stats()
+        min_ts = store_stats.get('min_timestamp', record['timestamp'])
+        uptime_seconds = record['timestamp'] - min_ts if min_ts else 0
+
+        return {
+            'received_total': record.get('received_total', 0),
+            'dispatched_success': record.get('dispatched_success', 0),
+            'dispatched_failed': record.get('dispatched_failed', 0),
+            'handler_success': record.get('handler_success', 0),
+            'handler_failed': record.get('handler_failed', 0),
+            'dispatch_queue_size': record.get('dispatch_queue_size', 0),
+            'dispatch_success_rate': dispatch_success_rate,
+            'uptime_seconds': str(int(uptime_seconds)),
+            'dispatch_latency': {
+                'avg_ms': f"{record.get('avg_dispatch_latency_ms', 0):.2f}",
+                'p50_ms': f"{record.get('p50_dispatch_latency_ms', 0):.2f}",
+                'p95_ms': f"{record.get('p95_dispatch_latency_ms', 0):.2f}",
+                'p99_ms': f"{record.get('p99_dispatch_latency_ms', 0):.2f}",
+            }
+        }
+
+    def _calculate_windows(self) -> Dict[str, Dict[str, Any]]:
+        """计算时间窗口统计"""
+        now = int(time.time())
+
+        windows = {
+            '1m': self._calculate_window(now, 60),
+            '3m': self._calculate_window(now, 180),
+            '5m': self._calculate_window(now, 300),
+            '10m': self._calculate_window(now, 600),
+            '15m': self._calculate_window(now, 900),
+        }
+
+        return windows
+
+    def _calculate_window(self, now: int, duration: int) -> Dict[str, Any]:
+        """计算单个时间窗口的统计数据"""
+        from_ts = now - duration
+        to_ts = now
+
+        # 查询时间范围内的数据
+        records = self.metrics_store.query_range(from_ts, to_ts)
+
+        if len(records) < 2:
+            # 数据不足，返回空统计
+            return {
+                'throughput_per_second': 0.0,
+                'avg_latency_ms': 0.0,
+                'success_rate': 0.0,
+                'total_messages': 0,
+                'failed_messages': 0,
+                'avg_queue_size': 0.0,
+                'window_duration': duration,
+                'data_points_count': len(records)
+            }
+
+        # 计算增量
+        first_record = records[0]
+        last_record = records[-1]
+
+        received_delta = last_record['received_total'] - first_record['received_total']
+        success_delta = last_record['dispatched_success'] - first_record['dispatched_success']
+        failed_delta = last_record['dispatched_failed'] - first_record['dispatched_failed']
+
+        # 计算实际时间跨度
+        actual_duration = last_record['timestamp'] - first_record['timestamp']
+        if actual_duration < 1:
+            actual_duration = 1
+
+        # 计算吞吐量
+        throughput = received_delta / actual_duration
+
+        # 计算成功率
+        success_rate = 0.0
+        if received_delta > 0:
+            success_rate = (success_delta / received_delta) * 100
+
+        # 计算平均延迟
+        latencies = [r['avg_dispatch_latency_ms'] for r in records if r['avg_dispatch_latency_ms'] > 0]
+        avg_latency = sum(latencies) / len(latencies) if latencies else 0.0
+
+        # 计算平均队列大小
+        queue_sizes = [r['dispatch_queue_size'] for r in records]
+        avg_queue_size = sum(queue_sizes) / len(queue_sizes) if queue_sizes else 0.0
+
+        return {
+            'throughput_per_second': round(throughput, 2),
+            'avg_latency_ms': round(avg_latency, 2),
+            'success_rate': round(success_rate, 2),
+            'total_messages': received_delta,
+            'failed_messages': failed_delta,
+            'avg_queue_size': round(avg_queue_size, 1),
+            'window_duration': duration,
+            'data_points_count': len(records)
+        }
+
+    def get_window_metrics(self, window_names: List[str]) -> Dict[str, Dict[str, Any]]:
+        """获取指定时间窗口的指标"""
+        now = int(time.time())
+
+        window_durations = {
+            '1m': 60,
+            '3m': 180,
+            '5m': 300,
+            '10m': 600,
+            '15m': 900,
+        }
+
+        result = {}
+        for name in window_names:
+            if name in window_durations:
+                result[name] = self._calculate_window(now, window_durations[name])
+
+        return result
+
+    def get_history(self, from_ts: int, to_ts: int, limit: int = 1000) -> List[Dict[str, Any]]:
+        """获取历史数据"""
+        return self.metrics_store.query_range(from_ts, to_ts, limit=limit)
+
+    def get_latest_history(self, limit: int = 100) -> List[Dict[str, Any]]:
+        """获取最新的历史数据"""
+        return self.metrics_store.query_latest(limit=limit)
+
+    def get_service_info(self) -> Dict[str, Any]:
+        """获取监控服务信息"""
+        store_stats = self.metrics_store.get_stats()
+
+        # 检查是否有最近的数据更新（最近30秒内）
+        max_ts = store_stats.get('max_timestamp', 0)
+        is_running = (time.time() - max_ts) < 30 if max_ts else False
+
+        return {
+            'agent_id': 'unknown',
+            'running': is_running,
+            'snapshot_interval': 10,
+            'snapshot_count': store_stats.get('total_records', 0),
+            'store_stats': store_stats
+        }
+
+
+def get_standalone_reader(db_path: str = None) -> StandaloneMonitoringReader:
+    """获取独立监控读取器实例
+
+    Args:
+        db_path: 数据库路径，如果为空则使用默认路径
+
+    Returns:
+        StandaloneMonitoringReader 实例
+    """
+    return StandaloneMonitoringReader(db_path)

package/extensions/services/proxy/agentcp/msg/__init__.py

@@ -0,0 +1,21 @@
+# Copyright 2025 AgentUnion Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from . import message_client
+from . import open_ai_message
+from . import message_serialize
+from . import session_manager
+from . import stream_client
+from . import wss_binary_message
+from . import websocket_file_receiver