novalon-website/scripts/production-diagnosis.sh

#!/bin/bash

# 生产环境连接超时诊断脚本
# 用途：系统化排查生产环境连接超时问题

set -e

# 颜色定义
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m'

# 日志函数
log_info() {
    echo -e "${BLUE}[INFO]${NC} $1"
}

log_success() {
    echo -e "${GREEN}[SUCCESS]${NC} $1"
}

log_warning() {
    echo -e "${YELLOW}[WARNING]${NC} $1"
}

log_error() {
    echo -e "${RED}[ERROR]${NC} $1"
}

separator() {
    echo "======================================================================"
}

# 1. 收集错误信息
collect_error_info() {
    separator
    log_info "Phase 1.1: 收集错误信息"
    separator
    
    echo ""
    log_info "测试基本网络连接..."
    
    # 测试本地网络
    if ping -c 3 8.8.8.8 > /dev/null 2>&1; then
        log_success "本地网络正常"
    else
        log_error "本地网络异常"
        return 1
    fi
    
    echo ""
    log_info "测试DNS解析..."
    
    # 测试生产服务器域名
    if nslookup novalon.cn > /dev/null 2>&1; then
        log_success "novalon.cn DNS解析成功"
        PROD_IP=$(nslookup novalon.cn | grep "Address" | tail -1 | awk '{print $2}')
        log_info "生产服务器IP: $PROD_IP"
    else
        log_error "novalon.cn DNS解析失败"
    fi
    
    # 测试Git服务器
    if nslookup git.f.novalon.cn > /dev/null 2>&1; then
        log_success "git.f.novalon.cn DNS解析成功"
        GIT_IP=$(nslookup git.f.novalon.cn | grep "Address" | tail -1 | awk '{print $2}')
        log_info "Git服务器IP: $GIT_IP"
    else
        log_error "git.f.novalon.cn DNS解析失败"
    fi
    
    # 测试CI服务器
    if nslookup ci.f.novalon.cn > /dev/null 2>&1; then
        log_success "ci.f.novalon.cn DNS解析成功"
        CI_IP=$(nslookup ci.f.novalon.cn | grep "Address" | tail -1 | awk '{print $2}')
        log_info "CI服务器IP: $CI_IP"
    else
        log_error "ci.f.novalon.cn DNS解析失败"
    fi
    
    echo ""
    log_info "测试TCP连接..."
    
    # 测试HTTP端口
    if nc -zv -w 5 novalon.cn 80 2>&1 | grep -q "succeeded"; then
        log_success "novalon.cn:80 连接成功"
    else
        log_error "novalon.cn:80 连接超时"
    fi
    
    # 测试HTTPS端口
    if nc -zv -w 5 novalon.cn 443 2>&1 | grep -q "succeeded"; then
        log_success "novalon.cn:443 连接成功"
    else
        log_error "novalon.cn:443 连接超时"
    fi
    
    # 测试Git服务器
    if nc -zv -w 5 git.f.novalon.cn 443 2>&1 | grep -q "succeeded"; then
        log_success "git.f.novalon.cn:443 连接成功"
    else
        log_error "git.f.novalon.cn:443 连接超时"
    fi
    
    # 测试CI服务器
    if nc -zv -w 5 ci.f.novalon.cn 443 2>&1 | grep -q "succeeded"; then
        log_success "ci.f.novalon.cn:443 连接成功"
    else
        log_error "ci.f.novalon.cn:443 连接超时"
    fi
    
    echo ""
    log_info "测试HTTP响应..."
    
    # 测试HTTP请求
    HTTP_CODE=$(curl -s -o /dev/null -w "%{http_code}" --connect-timeout 10 --max-time 15 https://novalon.cn 2>&1)
    if [ "$HTTP_CODE" = "200" ] || [ "$HTTP_CODE" = "301" ] || [ "$HTTP_CODE" = "302" ]; then
        log_success "novalon.cn HTTP响应正常 (HTTP $HTTP_CODE)"
    else
        log_error "novalon.cn HTTP响应异常 (HTTP $HTTP_CODE)"
    fi
    
    # 测试Git服务器
    if curl -I --connect-timeout 10 --max-time 15 https://git.f.novalon.cn 2>&1 | grep -q "HTTP"; then
        log_success "git.f.novalon.cn HTTP响应正常"
    else
        log_error "git.f.novalon.cn HTTP响应超时"
    fi
    
    # 测试CI服务器
    if curl -I --connect-timeout 10 --max-time 15 https://ci.f.novalon.cn 2>&1 | grep -q "HTTP"; then
        log_success "ci.f.novalon.cn HTTP响应正常"
    else
        log_error "ci.f.novalon.cn HTTP响应超时"
    fi
}

# 2. 检查最近的变更
check_recent_changes() {
    separator
    log_info "Phase 1.2: 检查最近的变更"
    separator
    
    echo ""
    log_info "检查本地Git状态..."
    
    # 检查当前分支
    CURRENT_BRANCH=$(git branch --show-current)
    log_info "当前分支: $CURRENT_BRANCH"
    
    # 检查未提交的更改
    if git status --porcelain | grep -q .; then
        log_warning "存在未提交的更改:"
        git status --short
    else
        log_success "工作区干净"
    fi
    
    # 检查最近的提交
    echo ""
    log_info "最近5次提交:"
    git log --oneline -5
    
    # 检查最近的推送
    echo ""
    log_info "检查远程仓库连接..."
    if git remote -v | grep -q "git.f.novalon.cn"; then
        log_info "远程仓库: git.f.novalon.cn"
        
        # 尝试连接远程仓库
        if git ls-remote --heads origin > /dev/null 2>&1; then
            log_success "远程仓库连接成功"
        else
            log_error "远程仓库连接失败"
        fi
    fi
}

# 3. 追踪数据流
trace_data_flow() {
    separator
    log_info "Phase 1.3: 追踪数据流"
    separator
    
    echo ""
    log_info "网络路由追踪..."
    
    # 追踪到生产服务器的路由
    log_info "追踪到 novalon.cn 的路由:"
    traceroute -m 15 novalon.cn 2>&1 | head -20 || log_warning "traceroute命令不可用"
    
    echo ""
    log_info "追踪到 git.f.novalon.cn 的路由:"
    traceroute -m 15 git.f.novalon.cn 2>&1 | head -20 || log_warning "traceroute命令不可用"
    
    echo ""
    log_info "检查防火墙规则..."
    
    # 检查防火墙状态
    if command -v ufw > /dev/null; then
        sudo ufw status verbose
    elif command -v firewall-cmd > /dev/null; then
        sudo firewall-cmd --list-all
    else
        log_warning "未检测到防火墙"
    fi
    
    echo ""
    log_info "检查网络代理设置..."
    env | grep -i proxy || log_info "未设置网络代理"
}

# 4. 分析系统资源
analyze_system_resources() {
    separator
    log_info "Phase 1.4: 分析系统资源"
    separator
    
    echo ""
    log_info "检查DNS配置..."
    cat /etc/resolv.conf
    
    echo ""
    log_info "检查网络接口..."
    ifconfig | grep -E "^[a-z]|inet " || ip addr show
    
    echo ""
    log_info "检查系统负载..."
    uptime
    
    echo ""
    log_info "检查磁盘空间..."
    df -h | grep -E "Filesystem|/$|/home"
}

# 生成诊断报告
generate_report() {
    separator
    log_info "生成诊断报告"
    separator
    
    REPORT_FILE="/tmp/production-diagnosis-$(date +%Y%m%d_%H%M%S).log"
    
    {
        echo "生产环境连接超时诊断报告"
        echo "生成时间: $(date)"
        echo ""
        collect_error_info
        echo ""
        check_recent_changes
        echo ""
        trace_data_flow
        echo ""
        analyze_system_resources
    } > "$REPORT_FILE" 2>&1
    
    log_success "诊断报告已生成: $REPORT_FILE"
    
    echo ""
    log_info "报告摘要:"
    head -50 "$REPORT_FILE"
}

# 主函数
main() {
    log_info "开始生产环境连接超时诊断"
    
    if [ "$1" = "--report" ]; then
        generate_report
    else
        collect_error_info
        echo ""
        check_recent_changes
        echo ""
        trace_data_flow
        echo ""
        analyze_system_resources
    fi
}

main "$@"