chore: initial commit

smart_health_check.sh

@@ -0,0 +1,95 @@
+#\!/bin/bash
+# 智能健康检查 - 只在有问题时干预
+# 每12小时运行一次
+
+LOG_FILE="logs/smart_health_check.log"
+ERROR_LOG="logs/integrated_bot_errors.log"
+DETAIL_LOG="logs/integrated_bot_detailed.log"
+
+log() {
+    echo "[$(date "+%Y-%m-%d %H:%M:%S")] $1" | tee -a "$LOG_FILE"
+}
+
+log "=========================================="
+log "开始健康检查..."
+
+# 1. 检查机器人进程是否运行
+if \! pgrep -f "python.*integrated_bot_ai.py" >/dev/null; then
+    log "❌ 机器人进程未运行，需要启动"
+    ./manage_bot.sh restart >> "$LOG_FILE" 2>&1
+    log "✅ 已重启机器人"
+    exit 0
+fi
+
+# 2. 检查 Pyrogram 客户端状态
+PYROGRAM_RUNNING=$(tail -100 "$DETAIL_LOG" 2>/dev/null | grep -c "✅ Pyrogram客户端已启动")
+if [ "$PYROGRAM_RUNNING" -gt 0 ]; then
+    log "✅ Pyrogram 客户端状态: 正常"
+    PYROGRAM_OK=true
+else
+    log "⚠️  Pyrogram 客户端状态: 未知"
+    PYROGRAM_OK=false
+fi
+
+# 3. 检查最近1小时是否有 Connection lost 错误
+HOUR_AGO=$(date -d "1 hour ago" "+%Y-%m-%d %H" 2>/dev/null || date -v-1H "+%Y-%m-%d %H")
+RECENT_CONNECTION_ERRORS=$(tail -200 "$ERROR_LOG" 2>/dev/null | grep "Connection lost" | grep "$HOUR_AGO" | wc -l)
+
+log "最近1小时 Connection lost 错误: $RECENT_CONNECTION_ERRORS 个"
+
+# 4. 检查最近1小时是否有 AUTH_KEY 错误
+RECENT_AUTH_ERRORS=$(tail -200 "$ERROR_LOG" 2>/dev/null | grep "AUTH_KEY_UNREGISTERED" | grep "$HOUR_AGO" | wc -l)
+
+log "最近1小时 AUTH_KEY 错误: $RECENT_AUTH_ERRORS 个"
+
+# 5. 决策逻辑
+NEED_RESTART=false
+
+if [ "$RECENT_CONNECTION_ERRORS" -gt 5 ]; then
+    log "❌ 检测到过多连接错误 ($RECENT_CONNECTION_ERRORS 个)"
+    NEED_RESTART=true
+fi
+
+if [ "$RECENT_AUTH_ERRORS" -gt 0 ]; then
+    log "❌ 检测到 AUTH_KEY 错误"
+    NEED_RESTART=true
+fi
+
+if [ "$PYROGRAM_OK" = false ] && [ "$RECENT_CONNECTION_ERRORS" -gt 2 ]; then
+    log "❌ Pyrogram 状态异常且有连接错误"
+    NEED_RESTART=true
+fi
+
+# 6. 执行操作
+if [ "$NEED_RESTART" = true ]; then
+    log "==========================================",
+    log "🔄 检测到问题，准备重启机器人..."
+    log "==========================================",
+    
+    # 重启
+    ./manage_bot.sh restart >> "$LOG_FILE" 2>&1
+    
+    # 等待启动
+    sleep 10
+    
+    # 验证
+    if grep -q "✅ Pyrogram客户端已启动" "$DETAIL_LOG" 2>/dev/null; then
+        log "✅ 重启成功 - Pyrogram 已重新连接"
+    else
+        log "⚠️  重启完成 - 状态待确认"
+    fi
+    
+    log "=========================================="
+else
+    log "✅ 一切正常，无需干预"
+    log "=========================================="
+fi
+
+# 7. 状态摘要
+log "状态摘要:"
+log "  - 进程: 运行中"
+log "  - Pyrogram: $([ "$PYROGRAM_OK" = true ] && echo "正常" || echo "未知")"
+log "  - 连接错误: $RECENT_CONNECTION_ERRORS 个"
+log "  - AUTH错误: $RECENT_AUTH_ERRORS 个"
+log "  - 操作: $([ "$NEED_RESTART" = true ] && echo "已重启" || echo "无操作")"
+log ""