dev_1.0.0_sundp 完成Metric模块及模块e2e测试 (#18)
Co-authored-by: sundapeng.sdp <sundapeng@hashdata.cn> Reviewed-on: #18 Reviewed-by: xuxt <xuxt@zgclab.edu.cn> Reviewed-by: yuyr <yuyr@zgclab.edu.cn> Reviewed-by: huhy <husteryezi@163.com>
This commit was merged in pull request #18.
This commit is contained in:
903
src/metric/client-plugins/all-in-one-demo/scripts/install_artifact.sh
Executable file
903
src/metric/client-plugins/all-in-one-demo/scripts/install_artifact.sh
Executable file
@@ -0,0 +1,903 @@
|
||||
#!/bin/bash
|
||||
|
||||
set -e
|
||||
|
||||
RED='\033[0;31m'
|
||||
GREEN='\033[0;32m'
|
||||
YELLOW='\033[1;33m'
|
||||
BLUE='\033[0;34m'
|
||||
NC='\033[0m'
|
||||
|
||||
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} $1"
|
||||
}
|
||||
|
||||
# 配置变量
|
||||
INSTALL_DIR="${1:-$(pwd)}" # 使用第一个参数作为安装目录，如果没有参数则使用当前目录
|
||||
TEMP_DIR="/tmp/metrics-install-$$"
|
||||
VERSION_FILE="version.json"
|
||||
|
||||
|
||||
# 加载配置文件
|
||||
load_config() {
|
||||
local script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||||
local config_file="$script_dir/config.env"
|
||||
|
||||
if [[ -f "$config_file" ]]; then
|
||||
log_info "加载配置文件: $config_file"
|
||||
# 导出配置文件中的环境变量
|
||||
set -a # 自动导出所有变量
|
||||
source "$config_file"
|
||||
set +a # 关闭自动导出
|
||||
log_success "配置文件加载完成"
|
||||
else
|
||||
log_warning "配置文件不存在: $config_file，使用默认配置"
|
||||
fi
|
||||
}
|
||||
|
||||
# 复制配置文件到安装目录
|
||||
copy_config_files() {
|
||||
log_info "复制配置文件到安装目录..."
|
||||
|
||||
local script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||||
local source_config="$script_dir/../config/config.env"
|
||||
local target_config="$INSTALL_DIR/config.env"
|
||||
|
||||
if [[ -f "$source_config" ]]; then
|
||||
# 检查源文件和目标文件是否是同一个文件
|
||||
if [[ "$source_config" == "$target_config" ]]; then
|
||||
log_info "配置文件已在目标位置，跳过复制"
|
||||
log_success "配置文件已存在: $target_config"
|
||||
else
|
||||
if cp "$source_config" "$target_config"; then
|
||||
log_success "配置文件复制完成: $target_config"
|
||||
else
|
||||
log_error "配置文件复制失败"
|
||||
return 1
|
||||
fi
|
||||
fi
|
||||
else
|
||||
log_warning "源配置文件不存在: $source_config"
|
||||
fi
|
||||
|
||||
# 复制版本校验脚本
|
||||
log_info "复制版本校验脚本到安装目录..."
|
||||
local target_check_version="$INSTALL_DIR/check_version.sh"
|
||||
|
||||
# 检查目标文件是否已存在（从 artifact 包中解压出来的）
|
||||
if [[ -f "$target_check_version" ]]; then
|
||||
log_info "版本校验脚本已存在，设置执行权限..."
|
||||
chmod +x "$target_check_version"
|
||||
log_success "版本校验脚本权限设置完成: $target_check_version"
|
||||
else
|
||||
log_warning "版本校验脚本不存在: $target_check_version"
|
||||
log_info "请确保 check_version.sh 已包含在 artifact 包中"
|
||||
fi
|
||||
}
|
||||
|
||||
check_root() {
|
||||
if [[ $EUID -ne 0 ]]; then
|
||||
log_error "此脚本需要 root 权限运行"
|
||||
log_info "请使用: sudo $0 [安装目录]"
|
||||
log_info "如果不指定安装目录，将使用当前目录: $(pwd)"
|
||||
exit 1
|
||||
fi
|
||||
}
|
||||
|
||||
# 检查系统要求
|
||||
check_system() {
|
||||
log_info "检查系统要求..."
|
||||
|
||||
# 检查操作系统
|
||||
if [[ ! -f /etc/os-release ]]; then
|
||||
log_error "无法检测操作系统版本"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
source /etc/os-release
|
||||
log_info "检测到操作系统: $NAME $VERSION"
|
||||
|
||||
# 检查系统架构
|
||||
arch=$(uname -m)
|
||||
log_info "系统架构: $arch"
|
||||
|
||||
# 检查磁盘空间
|
||||
available_space=$(df / | awk 'NR==2 {print $4}')
|
||||
if [[ $available_space -lt 10485760 ]]; then # 10GB in KB
|
||||
log_warning "可用磁盘空间不足 10GB，当前可用: $(($available_space / 1024 / 1024))GB"
|
||||
fi
|
||||
|
||||
# 检查内存
|
||||
total_mem=$(free -m | awk 'NR==2{print $2}')
|
||||
if [[ $total_mem -lt 4096 ]]; then # 4GB
|
||||
log_warning "系统内存不足 4GB，当前: ${total_mem}MB"
|
||||
fi
|
||||
}
|
||||
|
||||
# 查找版本文件
|
||||
find_version_file() {
|
||||
log_info "查找版本信息文件..."
|
||||
|
||||
# 在当前目录查找
|
||||
if [[ -f "$VERSION_FILE" ]]; then
|
||||
VERSION_FILE_PATH="$(pwd)/$VERSION_FILE"
|
||||
log_success "找到版本文件: $VERSION_FILE"
|
||||
return 0
|
||||
fi
|
||||
|
||||
# 在 artifact 目录查找
|
||||
for version_dir in artifact/*/; do
|
||||
if [[ -f "${version_dir}${VERSION_FILE}" ]]; then
|
||||
VERSION_FILE_PATH="$(cd "$(dirname "${version_dir}${VERSION_FILE}")" && pwd)/$(basename "${version_dir}${VERSION_FILE}")"
|
||||
log_success "找到版本文件: $VERSION_FILE_PATH"
|
||||
return 0
|
||||
fi
|
||||
done
|
||||
|
||||
log_error "未找到版本信息文件 $VERSION_FILE"
|
||||
exit 1
|
||||
}
|
||||
|
||||
# 解析版本信息
|
||||
parse_version_info() {
|
||||
log_info "解析版本信息..."
|
||||
|
||||
if [[ ! -f "$VERSION_FILE_PATH" ]]; then
|
||||
log_error "版本文件不存在: $VERSION_FILE_PATH"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 使用 jq 解析 JSON（如果可用）
|
||||
if command -v jq &> /dev/null; then
|
||||
# 验证JSON文件格式
|
||||
if ! jq empty "$VERSION_FILE_PATH" 2>/dev/null; then
|
||||
log_error "JSON文件格式错误，请检查 $VERSION_FILE_PATH"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
VERSION=$(jq -r '.version' "$VERSION_FILE_PATH")
|
||||
BUILD_TIME=$(jq -r '.build_time' "$VERSION_FILE_PATH")
|
||||
|
||||
# 解析 artifact_list
|
||||
if jq -e '.artifact_list' "$VERSION_FILE_PATH" > /dev/null 2>&1; then
|
||||
jq -r '.artifact_list | to_entries[] | "\(.key):\(.value)"' "$VERSION_FILE_PATH" > "$TEMP_DIR/components.txt"
|
||||
else
|
||||
log_error "version.json 中缺少 artifact_list 字段"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 解析 checksums
|
||||
if jq -e '.checksums' "$VERSION_FILE_PATH" > /dev/null 2>&1; then
|
||||
jq -r '.checksums | to_entries[] | "\(.key):\(.value)"' "$VERSION_FILE_PATH" > "$TEMP_DIR/checksums.txt"
|
||||
else
|
||||
log_error "version.json 中缺少 checksums 字段"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 解析 install_order（现在包含完整的文件名）
|
||||
if jq -e '.install_order' "$VERSION_FILE_PATH" > /dev/null 2>&1; then
|
||||
jq -r '.install_order[]' "$VERSION_FILE_PATH" > "$TEMP_DIR/install_order.txt"
|
||||
else
|
||||
log_error "version.json 中缺少 install_order 字段"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
else
|
||||
log_warning "jq 未安装，使用简单的 JSON 解析"
|
||||
# 简单的 JSON 解析
|
||||
VERSION=$(grep '"version"' "$VERSION_FILE_PATH" | sed 's/.*"version": *"\([^"]*\)".*/\1/')
|
||||
BUILD_TIME=$(grep '"build_time"' "$VERSION_FILE_PATH" | sed 's/.*"build_time": *"\([^"]*\)".*/\1/')
|
||||
|
||||
# 解析 artifact_list
|
||||
grep -A 100 '"artifact_list"' "$VERSION_FILE_PATH" | grep -E '^\s*"[^"]+":\s*"[^"]+"' | while read line; do
|
||||
component=$(echo "$line" | sed 's/.*"\([^"]*\)":\s*"[^"]*".*/\1/')
|
||||
version=$(echo "$line" | sed 's/.*"[^"]*":\s*"\([^"]*\)".*/\1/')
|
||||
echo "$component:$version" >> "$TEMP_DIR/components.txt"
|
||||
done
|
||||
|
||||
# 解析 checksums
|
||||
grep -A 100 '"checksums"' "$VERSION_FILE_PATH" | grep -E '^\s*"[^"]+":\s*"[^"]+"' | while read line; do
|
||||
component=$(echo "$line" | sed 's/.*"\([^"]*\)":\s*"[^"]*".*/\1/')
|
||||
checksum=$(echo "$line" | sed 's/.*"[^"]*":\s*"\([^"]*\)".*/\1/')
|
||||
echo "$component:$checksum" >> "$TEMP_DIR/checksums.txt"
|
||||
done
|
||||
|
||||
# 解析 install_order
|
||||
grep -A 100 '"install_order"' "$VERSION_FILE_PATH" | grep -E '^\s*"[^"]+"' | while read line; do
|
||||
component=$(echo "$line" | sed 's/.*"\([^"]*\)".*/\1/')
|
||||
echo "$component" >> "$TEMP_DIR/install_order.txt"
|
||||
done
|
||||
|
||||
# 验证解析结果
|
||||
if [[ ! -f "$TEMP_DIR/components.txt" || ! -s "$TEMP_DIR/components.txt" ]]; then
|
||||
log_error "无法解析 artifact_list，请检查 version.json 格式"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
if [[ ! -f "$TEMP_DIR/checksums.txt" || ! -s "$TEMP_DIR/checksums.txt" ]]; then
|
||||
log_error "无法解析 checksums，请检查 version.json 格式"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
if [[ ! -f "$TEMP_DIR/install_order.txt" || ! -s "$TEMP_DIR/install_order.txt" ]]; then
|
||||
log_error "无法解析 install_order，请检查 version.json 格式"
|
||||
exit 1
|
||||
fi
|
||||
fi
|
||||
|
||||
log_success "版本信息解析完成"
|
||||
log_info " 版本: $VERSION"
|
||||
log_info " 构建时间: $BUILD_TIME"
|
||||
|
||||
component_count=0
|
||||
if [[ -f "$TEMP_DIR/components.txt" ]]; then
|
||||
component_count=$(wc -l < "$TEMP_DIR/components.txt")
|
||||
log_info " 组件数量: $component_count"
|
||||
log_info " 组件列表:"
|
||||
while IFS= read -r line; do
|
||||
component=$(echo "$line" | cut -d':' -f1)
|
||||
version=$(echo "$line" | cut -d':' -f2)
|
||||
log_info " - $component v$version"
|
||||
done < "$TEMP_DIR/components.txt"
|
||||
else
|
||||
log_error "components.txt 文件不存在"
|
||||
exit 1
|
||||
fi
|
||||
}
|
||||
|
||||
# 验证文件完整性
|
||||
verify_checksums() {
|
||||
log_info "验证文件完整性..."
|
||||
|
||||
artifact_dir=$(dirname "$VERSION_FILE_PATH")
|
||||
log_info "Artifact 目录: $artifact_dir"
|
||||
failed_verification=0
|
||||
|
||||
if [[ -f "$TEMP_DIR/checksums.txt" ]]; then
|
||||
while IFS= read -r line; do
|
||||
component=$(echo "$line" | cut -d':' -f1)
|
||||
expected_checksum=$(echo "$line" | cut -d':' -f2-)
|
||||
|
||||
# 查找匹配的 tar 文件
|
||||
actual_file=""
|
||||
for file in "$artifact_dir/${component}-"*.tar.gz; do
|
||||
if [[ -f "$file" ]]; then
|
||||
actual_file="$file"
|
||||
break
|
||||
fi
|
||||
done
|
||||
|
||||
if [[ -z "$actual_file" ]]; then
|
||||
log_error "找不到组件文件: $component"
|
||||
failed_verification=1
|
||||
continue
|
||||
fi
|
||||
|
||||
# 计算实际校验和
|
||||
actual_checksum="sha256:$(sha256sum "$actual_file" | cut -d' ' -f1)"
|
||||
|
||||
if [[ "$actual_checksum" == "$expected_checksum" ]]; then
|
||||
log_success " $component: 校验通过"
|
||||
else
|
||||
log_error " $component: 校验失败"
|
||||
log_error " 期望: $expected_checksum"
|
||||
log_error " 实际: $actual_checksum"
|
||||
failed_verification=1
|
||||
fi
|
||||
done < "$TEMP_DIR/checksums.txt"
|
||||
fi
|
||||
|
||||
if [[ $failed_verification -eq 1 ]]; then
|
||||
log_error "文件完整性验证失败"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
log_success "所有文件校验通过"
|
||||
}
|
||||
|
||||
# 创建安装目录
|
||||
create_install_dirs() {
|
||||
log_info "创建安装目录..."
|
||||
|
||||
mkdir -p "$INSTALL_DIR"
|
||||
mkdir -p "$TEMP_DIR"
|
||||
|
||||
log_success "安装目录创建完成: $INSTALL_DIR"
|
||||
}
|
||||
|
||||
# 安装系统依赖包
|
||||
install_system_deps() {
|
||||
log_info "检查系统依赖包..."
|
||||
|
||||
local script_dir="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||||
local deps_dir="$script_dir/deps"
|
||||
|
||||
# 检查deps目录是否存在
|
||||
if [[ ! -d "$deps_dir" ]]; then
|
||||
log_info "deps 目录不存在，跳过系统依赖包安装"
|
||||
return 0
|
||||
fi
|
||||
|
||||
# 检查是否有tar.gz文件
|
||||
local deps_count=$(find "$deps_dir" -name "*.tar.gz" | wc -l)
|
||||
if [[ $deps_count -eq 0 ]]; then
|
||||
log_info "deps 目录中没有 tar.gz 文件，跳过系统依赖包安装"
|
||||
return 0
|
||||
fi
|
||||
|
||||
log_info "找到 $deps_count 个系统依赖包，开始安装..."
|
||||
|
||||
# 创建临时目录用于解压依赖包
|
||||
local deps_temp_dir="$TEMP_DIR/deps"
|
||||
mkdir -p "$deps_temp_dir"
|
||||
|
||||
# 处理每个tar.gz文件
|
||||
find "$deps_dir" -name "*.tar.gz" | while read tar_file; do
|
||||
local tar_basename=$(basename "$tar_file")
|
||||
local extract_name="${tar_basename%.tar.gz}"
|
||||
|
||||
log_info "处理依赖包: $tar_basename"
|
||||
|
||||
# 解压到临时目录
|
||||
local extract_dir="$deps_temp_dir/$extract_name"
|
||||
mkdir -p "$extract_dir"
|
||||
|
||||
if tar -xzf "$tar_file" -C "$extract_dir" 2>/dev/null; then
|
||||
log_success " $tar_basename 解压完成"
|
||||
else
|
||||
log_error " $tar_basename 解压失败"
|
||||
continue
|
||||
fi
|
||||
|
||||
# 进入解压目录，查找deb包
|
||||
cd "$extract_dir"
|
||||
local deb_count=$(find . -name "*.deb" | wc -l)
|
||||
|
||||
if [[ $deb_count -gt 0 ]]; then
|
||||
log_info " 找到 $deb_count 个 deb 包，开始安装..."
|
||||
|
||||
# 1. 先尝试安装所有deb包
|
||||
log_info " 第1步：批量安装deb包..."
|
||||
if dpkg -i *.deb 2>/dev/null; then
|
||||
log_success " 所有deb包安装成功"
|
||||
else
|
||||
log_warning " 部分deb包安装失败，可能存在依赖问题"
|
||||
|
||||
# 2. 使用apt-get修复依赖
|
||||
log_info " 第2步：修复依赖关系..."
|
||||
if apt-get install -f -y; then
|
||||
log_success " 依赖关系修复完成"
|
||||
else
|
||||
log_error " 依赖关系修复失败"
|
||||
# 继续处理其他包，不退出
|
||||
fi
|
||||
fi
|
||||
else
|
||||
log_info " $tar_basename 中没有找到deb包，跳过"
|
||||
fi
|
||||
|
||||
# 返回到依赖临时目录
|
||||
cd "$deps_temp_dir"
|
||||
done
|
||||
|
||||
# 检查并启动 cron 服务
|
||||
start_cron_service
|
||||
|
||||
log_success "系统依赖包安装完成"
|
||||
}
|
||||
|
||||
# 启动 cron 服务
|
||||
start_cron_service() {
|
||||
log_info "检查并启动 cron 服务..."
|
||||
|
||||
# 检查 cron 是否已经在运行
|
||||
if pgrep -x "cron" > /dev/null; then
|
||||
log_success "cron 服务已在运行"
|
||||
return 0
|
||||
fi
|
||||
|
||||
# 检查 /usr/sbin/cron 是否存在
|
||||
if [[ ! -f "/usr/sbin/cron" ]]; then
|
||||
log_warning "cron 可执行文件不存在，跳过启动"
|
||||
return 1
|
||||
fi
|
||||
|
||||
# 启动 cron 服务
|
||||
log_info "启动 cron 服务..."
|
||||
if /usr/sbin/cron start 2>/dev/null || /usr/sbin/cron 2>/dev/null; then
|
||||
log_success "cron 服务启动成功"
|
||||
|
||||
sleep 2
|
||||
|
||||
if pgrep -x "cron" > /dev/null; then
|
||||
log_success "cron 服务运行正常"
|
||||
else
|
||||
log_warning "cron 服务可能未正常启动"
|
||||
fi
|
||||
else
|
||||
log_error "cron 服务启动失败"
|
||||
return 1
|
||||
fi
|
||||
}
|
||||
|
||||
# 安装组件
|
||||
install_components() {
|
||||
log_info "开始安装组件..."
|
||||
|
||||
artifact_dir=$(dirname "$VERSION_FILE_PATH")
|
||||
log_info "Artifact 目录: $artifact_dir"
|
||||
install_count=0
|
||||
total_count=0
|
||||
|
||||
if [[ -f "$TEMP_DIR/install_order.txt" ]]; then
|
||||
total_count=$(wc -l < "$TEMP_DIR/install_order.txt")
|
||||
fi
|
||||
|
||||
if [[ -f "$TEMP_DIR/install_order.txt" ]]; then
|
||||
while IFS= read -r filename; do
|
||||
install_count=$((install_count + 1))
|
||||
|
||||
# 从文件名中提取组件名（去掉时间戳后缀）
|
||||
component=$(echo "$filename" | sed 's/-[0-9]\{8\}-[0-9]\{6\}\.tar\.gz$//')
|
||||
|
||||
log_info "[$install_count/$total_count] 安装 $component..."
|
||||
log_info " 文件名: $filename"
|
||||
|
||||
# 直接使用完整的文件名
|
||||
tar_file="$artifact_dir/$filename"
|
||||
|
||||
if [[ ! -f "$tar_file" ]]; then
|
||||
log_error "找不到组件文件: $filename"
|
||||
log_info " 期望路径: $tar_file"
|
||||
log_info " 当前目录: $(pwd)"
|
||||
log_info " 目录内容:"
|
||||
ls -la "$artifact_dir" | while read line; do
|
||||
log_info " $line"
|
||||
done
|
||||
exit 1
|
||||
fi
|
||||
|
||||
log_info " 找到文件: $tar_file"
|
||||
|
||||
# 解压到临时目录
|
||||
component_temp_dir="$TEMP_DIR/$component"
|
||||
mkdir -p "$component_temp_dir"
|
||||
|
||||
if tar -xzf "$tar_file" -C "$component_temp_dir" 2>/dev/null; then
|
||||
log_success " $component 解压完成"
|
||||
else
|
||||
log_error " $component 解压失败"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 查找解压后的目录
|
||||
extracted_dir=""
|
||||
for dir in "$component_temp_dir"/*; do
|
||||
if [[ -d "$dir" ]]; then
|
||||
extracted_dir="$dir"
|
||||
break
|
||||
fi
|
||||
done
|
||||
|
||||
if [[ -z "$extracted_dir" ]]; then
|
||||
log_error " $component 解压后未找到目录"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 执行安装脚本
|
||||
if [[ -f "$extracted_dir/install.sh" ]]; then
|
||||
log_info " 执行 $component 安装脚本..."
|
||||
if (cd "$extracted_dir" && ./install.sh "$INSTALL_DIR"); then
|
||||
log_success " $component 安装完成"
|
||||
else
|
||||
log_error " $component 安装失败"
|
||||
exit 1
|
||||
fi
|
||||
else
|
||||
log_error " $component 缺少 install.sh 文件"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 将解压后的目录移动到安装目录，保留组件目录
|
||||
component_install_dir="$INSTALL_DIR/$component"
|
||||
# 简化安装逻辑：直接删除旧目录，不进行备份
|
||||
if [[ -d "$component_install_dir" ]]; then
|
||||
log_info " 组件目录已存在，删除旧版本: $component_install_dir"
|
||||
rm -rf "$component_install_dir"
|
||||
# log_info " 组件目录已存在，备份后更新: $component_install_dir"
|
||||
# mv "$component_install_dir" "${component_install_dir}.backup.$(date +%Y%m%d_%H%M%S)"
|
||||
fi
|
||||
mv "$extracted_dir" "$component_install_dir"
|
||||
log_success " 组件目录已保存: $component_install_dir"
|
||||
|
||||
# 清理临时文件
|
||||
rm -rf "$component_temp_dir"
|
||||
done < "$TEMP_DIR/install_order.txt"
|
||||
fi
|
||||
|
||||
log_success "所有组件安装完成"
|
||||
}
|
||||
|
||||
# 创建安装记录
|
||||
create_install_record() {
|
||||
log_info "创建安装记录..."
|
||||
|
||||
# 等待一段时间确保所有进程都已启动
|
||||
log_info "等待进程启动..."
|
||||
sleep 3
|
||||
|
||||
local install_time=$(date -u +"%Y-%m-%dT%H:%M:%SZ")
|
||||
local install_record_file="$INSTALL_DIR/.install_record"
|
||||
|
||||
# 创建 JSON 格式的安装记录
|
||||
cat > "$install_record_file" << EOF
|
||||
{
|
||||
"version": "$VERSION",
|
||||
"build_time": "$BUILD_TIME",
|
||||
"install_time": "$install_time",
|
||||
"install_dir": "$INSTALL_DIR",
|
||||
"install_pid": $$,
|
||||
"components": {
|
||||
EOF
|
||||
|
||||
# 添加组件信息
|
||||
local first_component=true
|
||||
if [[ -f "$TEMP_DIR/components.txt" ]]; then
|
||||
while IFS= read -r line; do
|
||||
component=$(echo "$line" | cut -d':' -f1)
|
||||
version=$(echo "$line" | cut -d':' -f2)
|
||||
|
||||
# 获取组件的进程信息
|
||||
local component_pid=""
|
||||
|
||||
# 根据组件名查找进程，使用多种方法确保能找到PID
|
||||
case "$component" in
|
||||
"node-exporter")
|
||||
# 尝试多种方式查找node_exporter进程
|
||||
component_pid=$(pgrep -f "node_exporter" | head -1)
|
||||
if [[ -z "$component_pid" ]]; then
|
||||
component_pid=$(pgrep -f "node-exporter" | head -1)
|
||||
fi
|
||||
if [[ -z "$component_pid" ]]; then
|
||||
component_pid=$(ps aux | grep -v grep | grep "node_exporter" | awk '{print $2}' | head -1)
|
||||
fi
|
||||
;;
|
||||
"dcgm-exporter")
|
||||
# 查找dcgm-exporter进程
|
||||
component_pid=$(pgrep -f "dcgm-exporter" | head -1)
|
||||
if [[ -z "$component_pid" ]]; then
|
||||
component_pid=$(pgrep -f "dcgm_exporter" | head -1)
|
||||
fi
|
||||
if [[ -z "$component_pid" ]]; then
|
||||
component_pid=$(ps aux | grep -v grep | grep "dcgm-exporter" | awk '{print $2}' | head -1)
|
||||
fi
|
||||
;;
|
||||
"fluent-bit")
|
||||
# 查找fluent-bit进程
|
||||
component_pid=$(pgrep -f "fluent-bit" | head -1)
|
||||
if [[ -z "$component_pid" ]]; then
|
||||
component_pid=$(pgrep -f "fluent_bit" | head -1)
|
||||
fi
|
||||
if [[ -z "$component_pid" ]]; then
|
||||
component_pid=$(ps aux | grep -v grep | grep "fluent-bit" | awk '{print $2}' | head -1)
|
||||
fi
|
||||
;;
|
||||
"argus-agent")
|
||||
# 查找argus-agent进程
|
||||
component_pid=$(pgrep -f "argus-agent" | head -1)
|
||||
if [[ -z "$component_pid" ]]; then
|
||||
component_pid=$(ps aux | grep -v grep | grep "argus-agent" | awk '{print $2}' | head -1)
|
||||
fi
|
||||
;;
|
||||
esac
|
||||
|
||||
# 记录找到的PID信息
|
||||
if [[ -n "$component_pid" ]]; then
|
||||
log_info " 找到 $component 进程 PID: $component_pid"
|
||||
else
|
||||
log_warning " 未找到 $component 进程"
|
||||
fi
|
||||
|
||||
# 添加逗号分隔符
|
||||
if [[ "$first_component" == "true" ]]; then
|
||||
first_component=false
|
||||
else
|
||||
echo "," >> "$install_record_file"
|
||||
fi
|
||||
|
||||
# 添加组件信息
|
||||
cat >> "$install_record_file" << EOF
|
||||
"$component": {
|
||||
"version": "$version",
|
||||
"pid": "$component_pid",
|
||||
"install_dir": "$INSTALL_DIR/$component"
|
||||
}
|
||||
EOF
|
||||
done < "$TEMP_DIR/components.txt"
|
||||
fi
|
||||
|
||||
# 结束 JSON
|
||||
cat >> "$install_record_file" << EOF
|
||||
}
|
||||
}
|
||||
EOF
|
||||
|
||||
log_success "安装记录已创建: $install_record_file"
|
||||
}
|
||||
|
||||
# 设置健康检查定时任务
|
||||
setup_health_check_cron() {
|
||||
log_info "设置健康检查定时任务..."
|
||||
|
||||
# 直接使用当前安装目录，不依赖current软链接
|
||||
# INSTALL_DIR 是 /opt/argus-metric/versions/1.34.0
|
||||
local check_health_script="$INSTALL_DIR/check_health.sh"
|
||||
|
||||
# 检查健康检查脚本是否存在
|
||||
if [[ ! -f "$check_health_script" ]]; then
|
||||
log_error "健康检查脚本不存在: $check_health_script"
|
||||
return 1
|
||||
fi
|
||||
|
||||
# 确保脚本有执行权限
|
||||
chmod +x "$check_health_script"
|
||||
|
||||
# 创建临时crontab文件
|
||||
local temp_cron="/tmp/crontab_$$"
|
||||
|
||||
# 获取当前用户的crontab（如果存在）
|
||||
crontab -l 2>/dev/null > "$temp_cron" || touch "$temp_cron"
|
||||
|
||||
# 检查并删除旧的健康检查任务
|
||||
if grep -q "check_health.sh" "$temp_cron"; then
|
||||
log_info "发现旧的健康检查定时任务，正在更新..."
|
||||
# 删除所有包含check_health.sh的行
|
||||
grep -v "check_health.sh" "$temp_cron" > "$temp_cron.new"
|
||||
mv "$temp_cron.new" "$temp_cron"
|
||||
log_info "旧的健康检查定时任务已删除"
|
||||
fi
|
||||
|
||||
# 添加新的定时任务（每5分钟执行一次）
|
||||
echo "# Argus-Metrics 健康检查定时任务" >> "$temp_cron"
|
||||
echo "*/5 * * * * $check_health_script >> $INSTALL_DIR/.health_cron.log 2>&1" >> "$temp_cron"
|
||||
|
||||
# 安装新的crontab
|
||||
if crontab "$temp_cron"; then
|
||||
log_success "健康检查定时任务设置成功"
|
||||
log_info " 执行频率: 每5分钟"
|
||||
log_info " 日志文件: $INSTALL_DIR/.health_cron.log"
|
||||
log_info " 查看定时任务: crontab -l"
|
||||
log_info " 删除定时任务: crontab -e"
|
||||
else
|
||||
log_error "健康检查定时任务设置失败"
|
||||
rm -f "$temp_cron"
|
||||
return 1
|
||||
fi
|
||||
|
||||
# 清理临时文件
|
||||
rm -f "$temp_cron"
|
||||
|
||||
log_info "健康检查通过crontab自动执行"
|
||||
}
|
||||
|
||||
# 设置 DNS 同步定时任务
|
||||
setup_dns_sync_cron() {
|
||||
log_info "设置 DNS 同步定时任务..."
|
||||
|
||||
# 使用当前版本目录中的 DNS 同步脚本
|
||||
local sync_dns_script="$INSTALL_DIR/sync_dns.sh"
|
||||
|
||||
# 检查 DNS 同步脚本是否存在
|
||||
if [[ ! -f "$sync_dns_script" ]]; then
|
||||
log_warning "DNS 同步脚本不存在: $sync_dns_script"
|
||||
log_warning "跳过 DNS 同步定时任务设置"
|
||||
return 0
|
||||
fi
|
||||
|
||||
# 确保脚本有执行权限
|
||||
chmod +x "$sync_dns_script"
|
||||
|
||||
# 创建临时crontab文件
|
||||
local temp_cron="/tmp/crontab_$$"
|
||||
|
||||
# 获取当前用户的crontab（如果存在）
|
||||
crontab -l 2>/dev/null > "$temp_cron" || touch "$temp_cron"
|
||||
|
||||
# 检查并删除旧的 DNS 同步任务
|
||||
if grep -q "sync_dns.sh" "$temp_cron"; then
|
||||
log_info "发现旧的 DNS 同步定时任务，正在更新..."
|
||||
# 删除所有包含sync_dns.sh的行
|
||||
grep -v "sync_dns.sh" "$temp_cron" > "$temp_cron.new"
|
||||
mv "$temp_cron.new" "$temp_cron"
|
||||
log_info "旧的 DNS 同步定时任务已删除"
|
||||
fi
|
||||
|
||||
# 添加新的定时任务（每30秒执行一次）
|
||||
# 直接使用版本目录中的 DNS 同步脚本
|
||||
echo "# Argus-Metrics DNS 同步定时任务" >> "$temp_cron"
|
||||
echo "* * * * * $sync_dns_script >> $INSTALL_DIR/.dns_sync.log 2>&1" >> "$temp_cron"
|
||||
echo "* * * * * sleep 30; $sync_dns_script >> $INSTALL_DIR/.dns_sync.log 2>&1" >> "$temp_cron"
|
||||
|
||||
# 安装新的crontab
|
||||
if crontab "$temp_cron"; then
|
||||
log_success "DNS 同步定时任务设置成功"
|
||||
log_info " 执行频率: 每30秒"
|
||||
log_info " 日志文件: $INSTALL_DIR/.dns_sync.log"
|
||||
log_info " 查看定时任务: crontab -l"
|
||||
log_info " 删除定时任务: crontab -e"
|
||||
else
|
||||
log_error "DNS 同步定时任务设置失败"
|
||||
rm -f "$temp_cron"
|
||||
return 1
|
||||
fi
|
||||
|
||||
# 清理临时文件
|
||||
rm -f "$temp_cron"
|
||||
|
||||
log_info "DNS 同步通过crontab自动执行"
|
||||
}
|
||||
|
||||
# 设置版本校验定时任务
|
||||
setup_version_check_cron() {
|
||||
log_info "设置版本校验定时任务..."
|
||||
|
||||
# 使用当前版本目录中的版本校验脚本
|
||||
local check_version_script="$INSTALL_DIR/check_version.sh"
|
||||
|
||||
# 检查脚本是否存在
|
||||
if [[ ! -f "$check_version_script" ]]; then
|
||||
log_warning "版本校验脚本不存在: $check_version_script"
|
||||
log_info "跳过版本校验定时任务设置"
|
||||
return 0
|
||||
fi
|
||||
|
||||
# 确保脚本可执行
|
||||
chmod +x "$check_version_script"
|
||||
|
||||
# 创建临时crontab文件
|
||||
local temp_cron="/tmp/crontab_$$"
|
||||
crontab -l > "$temp_cron" 2>/dev/null || touch "$temp_cron"
|
||||
|
||||
# 检查是否已存在版本校验定时任务
|
||||
if grep -q "check_version.sh" "$temp_cron"; then
|
||||
log_info "发现旧的版本校验定时任务，正在更新..."
|
||||
# 删除所有包含check_version.sh的行
|
||||
grep -v "check_version.sh" "$temp_cron" > "$temp_cron.new"
|
||||
mv "$temp_cron.new" "$temp_cron"
|
||||
log_info "旧的版本校验定时任务已删除"
|
||||
fi
|
||||
|
||||
# 添加新的定时任务（每30分钟执行一次）
|
||||
echo "# Argus-Metrics 版本校验定时任务" >> "$temp_cron"
|
||||
echo "*/1 * * * * $check_version_script >> $INSTALL_DIR/.version_check.log 2>&1" >> "$temp_cron"
|
||||
|
||||
# 安装新的crontab
|
||||
if crontab "$temp_cron"; then
|
||||
log_success "版本校验定时任务设置成功"
|
||||
log_info " 执行频率: 每1分钟"
|
||||
log_info " 日志文件: $INSTALL_DIR/.version_check.log"
|
||||
log_info " 查看定时任务: crontab -l"
|
||||
log_info " 删除定时任务: crontab -e"
|
||||
else
|
||||
log_error "版本校验定时任务设置失败"
|
||||
rm -f "$temp_cron"
|
||||
return 1
|
||||
fi
|
||||
|
||||
# 清理临时文件
|
||||
rm -f "$temp_cron"
|
||||
|
||||
log_info "版本校验通过crontab自动执行"
|
||||
}
|
||||
|
||||
# 设置自动重启定时任务
|
||||
setup_restart_cron() {
|
||||
log_info "设置自动重启定时任务..."
|
||||
|
||||
# 使用当前版本目录中的重启脚本
|
||||
local restart_script="$INSTALL_DIR/restart_unhealthy.sh"
|
||||
|
||||
# 检查脚本是否存在
|
||||
if [[ ! -f "$restart_script" ]]; then
|
||||
log_warning "重启脚本不存在: $restart_script"
|
||||
log_info "跳过自动重启定时任务设置"
|
||||
return 0
|
||||
fi
|
||||
|
||||
# 确保脚本可执行
|
||||
chmod +x "$restart_script"
|
||||
|
||||
# 创建临时crontab文件
|
||||
local temp_cron="/tmp/crontab_$$"
|
||||
crontab -l > "$temp_cron" 2>/dev/null || touch "$temp_cron"
|
||||
|
||||
# 检查是否已存在自动重启定时任务
|
||||
if grep -q "restart_unhealthy.sh" "$temp_cron"; then
|
||||
log_info "发现旧的自动重启定时任务，正在更新..."
|
||||
# 删除所有包含restart_unhealthy.sh的行
|
||||
grep -v "restart_unhealthy.sh" "$temp_cron" > "$temp_cron.new"
|
||||
mv "$temp_cron.new" "$temp_cron"
|
||||
log_info "旧的自动重启定时任务已删除"
|
||||
fi
|
||||
|
||||
# 添加新的定时任务（每2分钟执行一次）
|
||||
echo "# Argus-Metrics 自动重启定时任务" >> "$temp_cron"
|
||||
echo "*/2 * * * * $restart_script >> $INSTALL_DIR/.restart.log 2>&1" >> "$temp_cron"
|
||||
|
||||
# 安装新的crontab
|
||||
if crontab "$temp_cron"; then
|
||||
log_success "自动重启定时任务设置成功"
|
||||
log_info " 执行频率: 每2分钟"
|
||||
log_info " 日志文件: $INSTALL_DIR/.restart.log"
|
||||
log_info " 查看定时任务: crontab -l"
|
||||
log_info " 删除定时任务: crontab -e"
|
||||
else
|
||||
log_error "自动重启定时任务设置失败"
|
||||
rm -f "$temp_cron"
|
||||
return 1
|
||||
fi
|
||||
|
||||
# 清理临时文件
|
||||
rm -f "$temp_cron"
|
||||
|
||||
log_info "自动重启检查通过crontab自动执行"
|
||||
}
|
||||
|
||||
# 显示安装信息
|
||||
show_install_info() {
|
||||
log_success "Argus-Metrics All-in-One 安装完成！"
|
||||
}
|
||||
|
||||
cleanup() {
|
||||
if [[ -d "$TEMP_DIR" ]]; then
|
||||
rm -rf "$TEMP_DIR"
|
||||
fi
|
||||
}
|
||||
|
||||
trap cleanup EXIT
|
||||
|
||||
# 主函数
|
||||
main() {
|
||||
echo "=========================================="
|
||||
echo " Argus-Metrics All-in-One 安装脚本 v1.0"
|
||||
echo "=========================================="
|
||||
echo
|
||||
|
||||
# 加载配置文件
|
||||
load_config
|
||||
|
||||
log_info "安装目录: $INSTALL_DIR"
|
||||
echo
|
||||
|
||||
check_root
|
||||
check_system
|
||||
find_version_file
|
||||
create_install_dirs
|
||||
parse_version_info
|
||||
verify_checksums
|
||||
install_system_deps
|
||||
install_components
|
||||
copy_config_files
|
||||
create_install_record
|
||||
setup_health_check_cron
|
||||
setup_dns_sync_cron
|
||||
setup_version_check_cron
|
||||
setup_restart_cron
|
||||
show_install_info
|
||||
}
|
||||
|
||||
if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
|
||||
main "$@"
|
||||
fi
|
||||
Reference in New Issue
Block a user