kelvin/hpctools
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
6e46c06edd42d1d720f47fcc63845124b2e380cd
hpctools/sunhpc
root 6e46c06edd first commit
2026-04-26 13:25:50 +08:00

1571 lines
48 KiB
Bash
Executable File
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/bin/bash
set -eo pipefail
#set -eou pipefail
R='\e[1;31m'
AR='\e[0;31m'
G='\e[1;92m'
AG='\e[0;92m'
C='\e[1;36m' # 亮青色
AC='\e[0;36m' # 暗青色
Y='\e[1;33m' # 亮黄色
AY='\e[0;33m' # 暗黄色
H='\e[0;90m' # 暗灰色 (推荐)
LH='\e[1;31m' # 亮黑色
AH='\e[1;90m' # 亮灰色
B='\e[0;37m' # 白色
B='\e[0;30m' # 黑色
NC='\e[0m' # No Color
# ====================== 公共工具函数 ======================
# 错误输出函数
error() {
echo -e "${AR}[E] $*${NC}" >&2
exit 1
}
# 信息输出函数
info() {
echo -e "${AG}[I]${NC} ${AC}$*${NC}"
}
# 警告输出函数
warn() {
echo -e "${AY}[W] $*${NC}"
}
# 清理最后一行
cline() {
echo -ne "\r\033[K"
}
COLUMNS=$(( $(tput cols 2>/dev/null || echo 80) - 8 ))
SSEP=$(printf "%${COLUMNS}s" "" | tr " " "-")
DSEP=$(printf "%${COLUMNS}s" "" | tr " " "=")
status_dnf_packages="/tmp/status_dnf_packages"
# ====================== 全局关联数组定义（核心优化） ======================
# 全局选项：verbose/debug/help/nodes 等，支持 GLOBAL_OPTS[v] 判断
declare -A GLOBAL_OPTS=(
[v]="" # -v
[verbose]="" # --verbose
[d]="" # -d
[debug]="" # --debug
[h]="" # -h
[help]="" # --help
[nodes]="" # -n/--nodes 节点值
)
# 子命令栈（数组：存储 exec/list node/serv autofs）
CMD_STACK=()
# 公共参数集合（-开头）
PUBLIC_ARGS=()
# 子命令专属参数（--开头）
PRIVATE_ARGS=()
# 默认节点列表(会自动从 nodes 文件覆盖)
DEFAULT_NODES=("localhost")
# 全局生效节点数组（所有子命令共享）
GLOBAL_NODES=()
declare -A GLOBAL_PORT
declare -A global_env
ask() {
# 基本用法：ask "是否删除文件？" - 无默认值，必须输入 y/yes 或 n/no
# ask "是否继续？" "y" - 默认继续，直接回车相当于 y
# ask "是否继续？" "n" - 默认取消，直接回车相当于 n
# if ask "警告：即将删除所有文件，是否继续？" "n"; then
local prompt="$1"
local default="$2"
local answer
# 构建提示信息
if [ "$default" = "y" ]; then
prompt="$prompt [Y/n]: "
elif [ "$default" = "n" ]; then
prompt="$prompt [y/N]: "
else
prompt="$prompt [y/n]: "
fi
while true; do
read -p "$prompt" answer
# 如果用户直接回车，使用默认值
if [ -z "$answer" ] && [ -n "$default" ]; then
answer="$default"
fi
# 转换为小写进行比较
answer=$(echo "$answer" | tr '[:upper:]' '[:lower:]')
case "$answer" in
y|yes) return 0 ;;
n|no) return 1 ;;
*) echo "请输入 y/yes 或 n/no" ;;
esac
done
}
load_os_envs() {
interfaces=$(ls /sys/class/net | grep -v lo)
for iface in $interfaces; do
readarray -t config <<< "$(nmcli -g IP4.ADDRESS,IP4.GATEWAY,IP4.DNS device show $iface 2>/dev/null)"
# IP地址
ip_addr=$(echo "${config[0]}" | cut -d: -f2 | cut -d'/' -f1)
global_env["${iface}_ip"]="${ip_addr:-}"
# 网关
gateway=$(echo "${config[1]}" | cut -d: -f2)
global_env["${iface}_gateway"]="${gateway:-}"
# DNS
dns=$(echo "${config[2]}" | cut -d: -f2- | tr ' ' ',' | sed 's/,$//')
global_env["${iface}_dns"]="${dns:-223.5.5.5,223.6.6.6}"
# CIDR掩码
cidr=$(nmcli -g IP4.ADDRESS device show $iface 2>/dev/null | head -1 | cut -d'/' -f2)
global_env["${iface}_mask"]="${cidr:-24}"
done
}
getFileArgs(){
# 调用模式: getFileArgs myfile "true" 未指定文件时只返回错误码
# 调用模式: result=$(getFileArgs myfile "true") 返回文件
# 调用模式: getFileArgs myfile 未指定文件时直接退出脚本,默认值: false
local -n filename=$1
local isForce=${2:-"false"} # 默认值为 "false"
# 检查 file 参数是否传入
if [[ -z "${GLOBAL_OPTS[file]-}" ]]; then
local err_msg="未指定文件，请使用 -f 或 --file 参数"
if [[ "$isForce" == "false" ]]; then
error "$err_msg"
else
warn "$err_msg"
return 1 # 返回非0表示失败
fi
fi
# 检查文件是否存在
if [[ -f "${GLOBAL_OPTS[file]}" ]] || [[ -d "${GLOBAL_OPTS[file]}" ]]; then
filename="${GLOBAL_OPTS[file]}"
return 0
else
local err_msg="错误：文件或目录不存在 -> ${GLOBAL_OPTS[file]}"
if [[ "$isForce" == "false" ]]; then
error "$err_msg"
else
warn "$err_msg"
return 1
fi
fi
}
# ====================== 核心函数：加载环境变量信息 ==============
load_conf_envs() {
while read -r key val; do
# 跳过空行、注释行
[[ -z "$key" || "$key" == \#* ]] && continue
global_env["$key"]="$val"
done < "$config_file"
}
# ============================================================
# 辅助函数：打印所有环境变量（调试用）
# ============================================================
print_global_env() {
for key in "${!global_env[@]}"; do
printf "%-20s = %s\n" "$key" "${global_env[$key]}"
done
echo "======================================"
}
# ============================================================
# 辅助函数：获取指定网卡IP
# ============================================================
get_interface_ip() {
local interface="$1"
if ip link show "$interface" &>/dev/null; then
ip addr show "$interface" | grep -oP '(?<=inet\s)\d+(\.\d+){3}' | head -1
else
echo ""
fi
}
# ============================================================
# 辅助函数：获取指定网卡MAC地址
# ============================================================
get_interface_mac() {
local interface="$1"
if ip link show "$interface" &>/dev/null; then
ip link show "$interface" | grep -oP '(?<=link/ether\s)[0-9a-f:]+' | head -1
else
echo ""
fi
}
# ============================================================
# 辅助函数：快速获取系统信息摘要
# ============================================================
show_system_summary() {
cat <<EOF
========================================
系统信息摘要
========================================
操作系统: ${global_env[release]} (${global_env[osname]} ${global_env[os_version]})
内核版本: ${global_env[kernel_release]}
主机名: ${global_env[hostname]}
架构: ${global_env[architecture]}
CPU: ${global_env[cpu_model]}
CPU核心数: ${global_env[cpu_cores]}
内存总量: ${global_env[mem_total]} GB
默认网卡: ${global_env[default_interface]}
系统负载: ${global_env[load_1min]} (1min) / ${global_env[load_5min]} (5min)
磁盘使用: ${global_env[disk_root_usage]}% (根分区)
虚拟化: ${global_env[virt_type]}
当前用户: ${global_env[username]} (UID: ${global_env[uid]})
========================================
EOF
}
# ============================================================
# 辅助函数：通用函数
# ============================================================
# 通用命令执行函数，带进度显示和日志记录
# 用法: run_with_progress "命令描述" "要执行的命令" [日志文件]
# 示例: run_with_progress "编译SLURM" "rpmbuild -ta slurm.tar.gz" "slurm_build.log"
run_with_progress() {
local description="$1"
local command="$2"
local log_file="${3:-/tmp/command_$(date +%Y%m%d_%H%M%S).log}"
local count=0
local start_time=$(date +%s)
local exit_code=0
info "$DSEP"
info "开始: $description"
info "日志: $log_file"
info "$DSEP"
# 执行命令，同时保存日志和显示进度
eval "$command" 2>&1 | tee "$log_file" | while IFS= read -r line; do
count=$((count + 1))
local current_time=$(date +%s)
local elapsed=$((current_time - start_time))
local elapsed_min=$((elapsed / 60))
local elapsed_sec=$((elapsed % 60))
# 提取关键信息（去掉路径等冗余信息）
local short_line=$(echo "$line" | sed 's|/home/[^/]*/|~/|g' | cut -c1-60)
printf "\r${AC}[B] [%s]${NR} ${H}[%d]${NR} %02d:%02d - %.60s" \
"$description" $count $elapsed_min $elapsed_sec "$short_line"
done
exit_code=${PIPESTATUS[0]}
# 清除最后一行进度信息
echo -ne "\r\033[K"
if [ $exit_code -eq 0 ]; then
info "[$description] - 完成! (共 $count 步, 耗时 $(($(date +%s) - start_time))秒)"
info "日志已保存: $log_file"
else
warn "❌ $description - 失败! (退出码: $exit_code, 共 $count 步)"
warn "请查看日志: $log_file"
return $exit_code
fi
}
# 过滤安装包
filter_pkgs(){
local -n input_array=$1 # 输入的包数组
local -n output_array=$2 # 输出的包数组
output_array=() # 清空输出数组
for pkg in ${input_array[@]}; do
#grep -q $pkg $status_dnf_packages && continue
if ! rpm -q "$pkg" &>/dev/null; then
#echo "$pkg" >> $status_dnf_packages
output_array+=("$pkg")
else
info "Skip installation, already installed - $pkg"
fi
done
}
cidr_to_netmask() {
local cidr=$1
local masks=(
"0.0.0.0" # 0
"128.0.0.0" # 1
"192.0.0.0" # 2
"224.0.0.0" # 3
"240.0.0.0" # 4
"248.0.0.0" # 5
"252.0.0.0" # 6
"254.0.0.0" # 7
"255.0.0.0" # 8
"255.128.0.0" # 9
"255.192.0.0" # 10
"255.224.0.0" # 11
"255.240.0.0" # 12
"255.248.0.0" # 13
"255.252.0.0" # 14
"255.254.0.0" # 15
"255.255.0.0" # 16
"255.255.128.0" # 17
"255.255.192.0" # 18
"255.255.224.0" # 19
"255.255.240.0" # 20
"255.255.248.0" # 21
"255.255.252.0" # 22
"255.255.254.0" # 23
"255.255.255.0" # 24
"255.255.255.128" # 25
"255.255.255.192" # 26
"255.255.255.224" # 27
"255.255.255.240" # 28
"255.255.255.248" # 29
"255.255.255.252" # 30
"255.255.255.254" # 31
"255.255.255.255" # 32
)
echo "${masks[$cidr]}"
}
get_subnet_address() {
local ip="$1"
local mask="$2"
IFS=. read -r i1 i2 i3 i4 <<< "$ip"
IFS=. read -r m1 m2 m3 m4 <<< "$mask"
n1=$((i1 & m1))
n2=$((i2 & m2))
n3=$((i3 & m3))
n4=$((i4 & m4))
echo "$n1.$n2.$n3.$n4"
}
# ====================== pxe server functions ====================
pxe_dhcp_build() {
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
interface="${GLOBAL_OPTS[interface]}"
ipaddr=${global_env[${interface}_ip]}
netcidr=${global_env[${interface}_mask]}
netmask=$(cidr_to_netmask 24)
subnet=$(get_subnet_address $ipaddr $netmask)
IFS=. read -r i1 i2 i3 i4 <<< "$ipaddr"
start_ip="$i1.$i2.$i3.10"
ender_ip="$i1.$i2.$i3.90"
tests_ip="$i1.$i2.$i3.252"
info "DHCP Interface name: $interface"
info "DHCP IP Address : $ipaddr"
info "DHCP IP Netmask : $netmask"
info "DHCP IP Subnets : $subnet"
info "DHCP Start IPAddr : $start_ip"
info "DHCP End IPAddr : $ender_ip"
nopkgs=()
pkgs=("dhcp-server dhcp-common")
filter_pkgs pkgs nopkgs
local cmd=("dnf -y install ${nopkgs[*]}")
[[ -z ${nopkgs[@]} ]] || run_with_progress "DNF 安装DHCP服务..." "$cmd" "$log_file"
dhcp_config="/etc/dhcp/dhcpd.conf"
{
echo "option domain-name \"`hostname -s`\";"
echo "option domain-name-servers 8.8.8.8;"
echo "default-lease-time 600;"
echo "max-lease-time 7200;"
echo "ddns-update-style none;"
echo ""
echo "allow booting;"
echo "allow bootp;"
echo ""
echo "option arch code 93 = unsigned integer 16;"
echo ""
echo "subnet ${subnet} netmask ${netmask} {"
echo " option routers ${ipaddr};"
echo " range $start_ip $ender_ip;"
echo ""
echo " next-server ${ipaddr};"
echo ""
echo ' host test_host_bond {'
echo ' option dhcp-client-identifier "test_host_bond";'
echo " fixed-address $tests_ip;"
echo ' }'
echo ' if option arch= 00:07 or option arch = 00:09 {'
echo ' # UEFI x64'
echo ' filename "/EFI/x86/grub.efi";'
echo ' } else if option arch = 00:0b {'
echo ' # UEFI ARM64'
echo ' filename "/EFI/aarch64/bootaa64.efi";'
echo ' } else {'
echo ' # Legcy BIOS'
echo ' filename "pxelinux.0";'
echo ' }'
echo '}'
} > ${dhcp_config}
# sed -i 's/\$DHCPDARGS/eth0/' /usr/lib/systemd/system/dhcpd.service
dhcp_systemd="/usr/lib/systemd/system/dhcpd.service"
grep -q "DHCPDARGS" $dhcp_systemd &>/dev/null && \
sed -i "s/\$DHCPDARGS/${interface}/" $dhcp_systemd &>/dev/null
systemctl daemon-reload && systemctl restart dhcpd.service
systemctl is-enabled dhcpd &>/dev/null || systemctl enable dhcpd.service &>/dev/null
systemctl is-active dhcpd &>/dev/null && info "DHCP 服务已经配置成功..." \
|| warn "DHCP 服务部署失败,请查看服务信息 systemctl status dhcpd.service"
}
pxe_tftp_build() {
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
nopkgs=()
pkgs=("tftp-server tftp syslinux")
filter_pkgs pkgs nopkgs
local cmd=("dnf -y install ${nopkgs[*]}")
[[ -z ${nopkgs[@]} ]] || run_with_progress "DNF 安装TFTP服务..." "$cmd" "$log_file"
tftp_server_conf="/usr/lib/systemd/system/tftp.service"
tftp_socket_conf="/usr/lib/systemd/system/tftp.socket"
tftp_pxe_dirs="/var/lib/tftpboot"
[[ -e "$tftp_pxe_dirs" ]] || mkdir -p "$tftp_pxe_dirs/pxelinux.cfg"
chmod -R 755 $tftp_pxe_dirs
# Pxe 启动文件配置
src_pxelinux="/usr/share/syslinux"
src_files=("pxelinux.0 menu.c32 ldlinux.c32")
for fname in ${src_files[@]}; do
src_file="$src_pxelinux/$fname"
dst_file="$tftp_pxe_dirs/$fname"
[[ -f $src_file ]] || warn "$src_file not found!"
if [[ -f $dst_file ]]; then
info "The $fname already exists, Skipt shis copy..."
else
info "Copying the $fname to $dst_file"
/usr/bin/cp $src_file $dst_file
fi
done
# 文件配置
if ! getFileArgs myfile "true"; then
warn " 必须指定系统ISO文件或挂载位置,例如: "
warn " -f /root/Rocky-9.7-x86_64-dvd.iso 或 -f /mnt/cdrom/"
exit 1
fi
echo "------------> $myfile"
exit 1
# -c 允许创建新文件(上传) -p 使用普通系统权限检查
{
echo '[Unit]'
echo 'Description=Tftp Server'
echo 'Requires=tftp.socket'
echo 'Documentation=man:in.tftpd'
echo ''
echo '[Service]'
echo "ExecStart=/usr/sbin/in.tftpd -c -p -s "$tftp_pxe_dirs""
echo 'StandardInput=socket'
echo ''
echo '[Install]'
echo 'WantedBy=multi-user.target'
echo 'Also=tftp-server.socket'
} > ${tftp_server_conf}
systemctl daemon-reload && systemctl restart tftp.socket
systemctl is-enabled tftp.socket &>/dev/null || systemctl enable tftp.socket &>/dev/null
info "TFTP 服务端口: 69"
systemctl is-active tftp.socket &>/dev/null && info "TFTP 服务已经配置成功..." \
|| warn "TFTP 服务部署失败,请查看服务信息 systemctl status tftp.socket"
}
pxe_http_build() {
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
nopkgs=()
pkgs=("httpd")
filter_pkgs pkgs nopkgs
local cmd=("dnf -y install ${nopkgs[*]}")
[[ -z ${nopkgs[@]} ]] || run_with_progress "DNF 安装HTTPD服务..." "$cmd" "$log_file"
http_root_conf="/etc/httpd"
http_html_dirs="/var/www/html"
http_serv_conf="/usr/lib/systemd/system/httpd.service"
custom_conf="$http_root_conf/conf.d/sunhpc.conf"
{
echo 'Alias /rocky97 /var/www/html/rocky97'
echo ''
echo 'UseCanonicalName Off'
echo "ServerName `hostname -f`"
echo ''
echo "<Directory \"$rocky_dirs\">"
echo " Options Indexes FollowSymLinks"
echo " AllowOverride None"
echo " Require all granted"
echo ""
echo " EnableSendfile on"
echo "</Directory>"
} > ${custom_conf}
#systemctl daemon-reload && systemctl restart httpd
#systemctl is-enabled httpd &>/dev/null || systemctl enable httpd &>/dev/null
info "HTTP 服务端口: 80,443"
if systemctl is-active httpd &>/dev/null ; then
info "HTTP 服务已经配置成功..."
info "后续配置: "
info " 拷贝系统ISO文件到 $rocky_repo_src"
info " # rsync -ah --info=progress2 /mnt/ /srv/repos/rocky/9.7/"
else
warn "HTTP 服务部署失败,请查看服务信息 systemctl status httpd.service"
fi
if getFileArgs myfile "true"; then
local mnt_iso_dirs=$myfile
if [[ -f $myfile ]]; then
mnt_dst_dirs="/mnt/cdrom"
[[ -e "$mnt_dst_dirs" ]] && umount $mnt_dst_dirs &>/dev/null || mkdir -p $mnt_dst_dirs
info "Mount path: $mnt_dst_dirs"
mount $myfile $mnt_dst_dirs &>/dev/null
mnt_iso_dirs=$mnt_dst_dirs
fi
rpmname=$(find $mnt_iso_dirs -iname "rocky-release-*.rpm")
filename=$(basename "$rpmname" .rpm)
distro=$(echo "$filename" | cut -d'-' -f1) # rocky
version=$(echo "$filename" | cut -d'-' -f3) # 9.7
distro=${distro:-"linux"}
version=${version:-"1.0"}
rocky_repo_distro="/srv/repos/$distro"
rocky_repo_version="$rocky_repo_distro/$version"
rocky_html_dst="$http_html_dirs/$distro"
[[ ! -e $rocky_repo_version ]] && mkdir -p $rocky_repo_version
[[ ! -e $rocky_html_dst ]] && ln -s $rocky_repo_distro $rocky_html_dst
info "拷贝系统镜像到本地: $rocky_repo_src"
info " 系统发行版: $distro"
info " 系统版本 : $version"
info " 源地址 : $mnt_iso_dirs"
info " 目的地址 : $rocky_repo_version"
rsync -ah --info=progress2 "$mnt_iso_dirs" "$rocky_repo_version/"
info "测试访问: curl -L http://`hostname -s`/$distro/$version/"
else
info "后续配置: "
info " 拷贝系统ISO文件到 $rocky_repo_src"
info " # rsync -ah --info=progress2 /mnt/ /srv/repos/rocky/9.7/"
fi
}
pxe_dnsmasq_build() {
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
nopkgs=()
pkgs=("syslinux dnsmasq dnsmasq-utils ipxe-bootimgs")
filter_pkgs pkgs nopkgs
local cmd=("dnf -y install ${nopkgs[*]}")
[[ -z ${nopkgs[@]} ]] || run_with_progress "DNF 安装 dnsmasq 服务..." "$cmd" "$log_file"
# -n 非空True，-z 为空True
[[ -n ${global_env[interface]} ]] && iface=${global_env[interface]} || \
error "请指定DHCP监听接口命名, 编辑 $config_file 文件, 例如: interface eth1"
[[ -n ${global_env[server]} ]] && ipaddr=${global_env[server]} || \
error "请指定监听接口IP地址, 编辑 $config_file 文件, 例如: server 172.16.9.254"
[[ -n ${global_env[dns]} ]] && dnsip=${global_env[dns]} || dnsip="223.5.5.5"
[[ -n ${global_env[range]} ]] && iprange=${global_env[range]} || \
error "请指定DHCP网络IP地址范围, 编辑 $config_file 文件, 例如: range 172.16.9.1,172.16.9.100"
vmlinuz="http://$ipaddr/rocky/isolinux/vmlinuz"
[[ -n ${global_env[vmlinuz]} ]] && vmlinuz=${global_env[vmlinuz]} || \
error "请指定vmlinuz web路径, 编辑 $config_file 文件, 例如: vmlinuz http://$ipaddr/rocky/isolinux/vmlinuz"
initrd="http://$ipaddr/rocky/isolinux/initrd.img"
[[ -n ${global_env[initrd]} ]] && initrd=${global_env[initrd]} || \
error "请指定initrd web路径, 编辑 $config_file 文件, 例如: initrd http://$ipaddr/rocky/isolinux/initrd.img"
[[ -n ${global_env[ks]} ]] && ks=${global_env[ks]} || \
error "请指定ks web路径, 编辑 $config_file 文件, 例如: ks http://$ipaddr/ks/ks.ks"
[[ -n ${global_env[repo]} ]] && repo=${global_env[repo]} || \
error "请指定ks web路径, 编辑 $config_file 文件, 例如: repo http://$ipaddr/rocky/"
kargs="net.ifnames=0 biosdevname=0"
[[ -n ${global_env[kargs]} ]] && kargs="$kargs ${global_env[kargs]}"
ipxeboot="http://$ipaddr/ks/boot.ipxe"
[[ -n ${global_env[bootipxe]} ]] && ipxeboot=${global_env[bootipxe]}
htmlroot="/var/www/html"
[[ -n ${global_env[htmlroot]} ]] && htmlroot=${global_env[htmlroot]}
boot_root="/srv/pxelinux/tftpboot"
dnsmasq_conf="/etc/dnsmasq.d/sunhpc.conf"
[[ -e "$boot_root/boot" ]] || mkdir -p "$boot_root/boot"
{
echo "interface=$interface"
echo "bind-interfaces"
echo ""
echo "enable-tftp"
echo "tftp-root=${boot_root}"
echo ""
echo "# Gateway:3, DNS:6"
echo "dhcp-range=$iprange,12h"
echo "dhcp-option=3,${ipaddr}"
echo "dhcp-option=6,$dnsip"
echo ""
echo "dhcp-match=set:bios,option:client-arch,0"
echo "dhcp-match=set:uefi,option:client-arch,7"
echo "dhcp-match=set:uefi,option:client-arch,9"
echo ""
echo "dhcp-boot=tag:bios,boot/undionly.kpxe"
echo "dhcp-boot=tag:uefi,boot/bootx64.efi"
echo "dhcp-boot=tag:ipxe,$ipxeboot"
} > ${dnsmasq_conf}
ipxeconf="$htmlroot$(echo "$htmlroot" | cut -d'/' -f4-)"
echo "----ipxeconf----> $ipxeconf"
}
# ====================== server functions ========================
slurm_build() {
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
FILE_CONF="slurm_build.conf"
REPOS_DIRS="/etc/yum.repos.d/*.repo"
SEARCH_STR="mirrors.ustc.edu.cn"
log_file="slurm_build.log"
EPEL_C="/etc/yum.repos.d/epel.repo"
REPO_C="/etc/yum.repos.d/slurm.repo"
DST_DIR="/srv/repos/slurm/packages"
STATUS="$DST_DIR/.installed"
init_db_flags="/var/lib/mysql/.initialized"
slurmctld_flags="/etc/slurm/.slurmctld"
slurmdbd_flags="/etc/slurm/.slurmdbd"
slurmd_flags="/etc/slurm/.slurmd"
# 检查 file 参数是否传入
if [[ -z "${GLOBAL_OPTS[file]-}" ]]; then
echo "错误：未指定文件，请使用 -f 或 --file 参数"
exit 1
fi
# 检查文件是否存在
if [[ ! -f "${GLOBAL_OPTS[file]}" ]]; then
echo "错误：文件不存在 → ${GLOBAL_OPTS[file]}"
exit 1
fi
info "${DSEP}"
info "🔨 开始 Slurm 构建安装（断点续跑模式）"
info "${DSEP}"
dnf config-manager --enable crb
[ -f "$EPEL_C" ] && info "EPEL Repo is enabled" || dnf -y install epel-release
local cmd="dnf -y install @development yum-utils dnf-utils rpm-build"
[[ -f $STATUS ]] || run_with_progress "DNF 安装系统开发依赖包..." "$cmd" "$log_file"
local pkgs="munge munge-devel mariadb mariadb-devel \
gtk2 gtk2-devel gtk3 gtk3-devel http-parser http-parser-devel \
json-c json-c-devel libyaml libyaml-devel libjwt libjwt-devel \
wget python3 readline-devel pam-devel perl-ExtUtils-MakeMaker \
perl-devel perl-JSON-PP createrepo_c hdf5 hdf5-devel man2html \
man2html-core pam pam-devel freeipmi freeipmi-devel numactl \
numactl-devel pmix pmix-devel hwloc hwloc-devel lua lua-devel \
ucx ucx-devel jq lz4-devel mariadb-server mariadb-devel"
nopkgs=()
filter_pkgs pkgs nopkgs
cmd=("dnf -y install ${nopkgs[*]}")
[[ -z ${nopkgs[@]} ]] || run_with_progress "DNF 安装系统基础依赖包..." "$cmd" "$log_file"
export CPPFLAGS="-I$PWD/deps_pkgs/cuda-13.0/targets/x86_64-linux/include ${CPPFLAGS}"
export LDFLAGS="-L$PWD/deps_pkgs/cuda-13.0/targets/x86_64-linux/lib/stubs ${LDFLAGS}"
[[ -f $STATUS ]] && info "Slurm 已经存在 ..." || info "开始制作 Slurm rpm 安装包 ..."
TARFILE=${GLOBAL_OPTS[file]}
local cmd="rpmbuild -ta \
--with slurmrestd --with hdf5 --with hwloc --with muma \
--with pmix --with nvml --with lua --with ucx --with jwt --with freeipmi \
$TARFILE"
[[ -f $STATUS ]] || run_with_progress "Slurm RPM 构建..." "$cmd" "$log_file"
mkdir -pv $DST_DIR
find $HOME/rpmbuild/RPMS/ -iname "*.rpm" -exec mv {} $DST_DIR \;
{
echo "[slurm]"
echo "name=Slurm Local Repository"
echo "baseurl=file:///$DST_DIR"
echo "gpgcheck=0"
echo "enabled=1"
} > "$REPO_C"
touch $DST_DIR/.installed
#createrepo_c $DST_DIR 2>&1 | tee $log_file
info "SLURM 部署完成, Repos: $DST_DIR"
local pkgs="pcp-pmda-slurm slurm slurm-perlapi \
slurm-contribs slurm-devel slurm-libpmi slurm-pam_slurm \
slurm-sackd slurm-slurmctld slurm-slurmd \
slurm-slurmdbd slurm-slurmrestd \
slurm-example-configs"
nopkgs=()
filter_pkgs pkgs nopkgs
cmd=("dnf -y install ${nopkgs[*]}")
[[ -z ${nopkgs[@]} ]] || run_with_progress "Slurm 服务安装..." "$cmd" "$log_file"
# 开始配置 MariaDB 数据库
slurm_db_conf="/etc/my.cnf.d/slurm.cnf"
{
echo "[mysqld]"
echo "innodb_buffer_pool_size = 2G"
echo "innodb_log_file_size = 64M"
echo "innodb_lock_wait_timeout = 900"
} > "$slurm_db_conf"
systemctl is-active mariadb &>/dev/null || systemctl start mariadb &>/dev/null
systemctl is-enabled mariadb &>/dev/null || systemctl enable mariadb &>/dev/null
DB_ROOT_PASSWORD="admin_b101"
# 静默执行初始化数据库: mysql_secure_installation
[[ -f $init_db_flags ]] && info "Database already initialized, skipping mysql_secure_installation" \
|| mysql_secure_installation <<EOF
n
y
${DB_ROOT_PASSWORD}
${DB_ROOT_PASSWORD}
y
y
y
y
EOF
if mysql -u root -p${DB_ROOT_PASSWORD} -e "select 1" >/dev/null 2>&1; then
info "MariaDB initialization and securing completed successfully!"
else
warn "Error: Authentication failed. Please check the logs."
fi
info "数据库初始化密码: $DB_ROOT_PASSWORD"
info "重新设置密码: "
info " systemctl stop mariadb"
info " mysqld_safe --skip-grant-tables &"
info " mysql -u root"
info " flush privileges;"
info " alter user 'root'@'localhost' identified by 'new_pass'"
info " exit"
info "killall mysqld"
info "systemctl start mariadb"
SLURM_DB_PW="admin_b101"
SLURM_DB_NM="slurm_acct_db"
info "开始创建 Slurm 基础数据库: ${SLURM_DB_NM}"
info "Slurm Database User: slurm"
info "Slurm Database Pass: $SLURM_DB_PW"
[[ -f $init_db_flags ]] || mysql -u root -p${DB_ROOT_PASSWORD} <<-EOSQL
CREATE DATABASE IF NOT EXISTS ${SLURM_DB_NM};
DROP USER IF EXISTS 'slurm'@'localhost';
CREATE USER IF NOT EXISTS 'slurm'@'localhost' identified by '${SLURM_DB_PW}';
GRANT ALL PRIVILEGES ON ${SLURM_DB_NM}.* TO 'slurm'@'localhost' WITH GRANT OPTION;
FLUSH PRIVILEGES;
EOSQL
touch ${init_db_flags}
# 配置 slurmdbd 服务
slurm_root="/etc/slurm"
slurm_dbdc="$slurm_root/slurmdbd.conf"
slurm_ctld="$slurm_root/slurm.conf"
getent passwd slurm &>/dev/null || useradd -r -b /var/lib -s /sbin/nologin slurm
getent passwd slurm &>/dev/null || useradd -r -b /var/lib -s /sbin/nologin slurmrestd
touch ${slurm_dbdc}
chown slurm:slurm ${slurm_dbdc}
chmod 600 ${slurm_dbdc}
mkdir -p "/var/log/slurm"
mkdir -p "/var/run/slurmdbd"
chown -R slurm:slurm "/var/log/slurm"
chown -R slurm:slurm "/var/run/slurmdbd"
[[ -f $slurmdbd_flags ]] && info "SlurmDBD Service is configuration finisheld..." \
|| cat > ${slurm_dbdc} <<-EOSLURMDBD
AuthType=auth/munge
LogFile=/var/log/slurm/slurmdbd.log
PidFile=/var/run/slurmdbd/slurmdbd.pid
DbdHost=`hostname -s`
DbdPort=6819
SlurmUser=slurm
DebugLevel=info
StorageType=accounting_storage/mysql
StorageHost=localhost
StoragePort=3306
StorageUser=slurm
StoragePass=${SLURM_DB_PW}
StorageLoc=${SLURM_DB_NM}
ArchiveEvents=yes
ArchiveJobs=yes
ArchiveResvs=yes
ArchiveSteps=no
ArchiveSuspend=no
ArchiveTXN=no
ArchiveUsage=no
PurgeEventAfter=1month
PurgeJobAfter=12month
PurgeResvAfter=1month
PurgeStepAfter=1month
PurgeSuspendAfter=1month
PurgeTXNAfter=12month
PurgeUsageAfter=24month
EOSLURMDBD
touch $slurmdbd_flags
if [[ ! -f $slurmdbd_flags ]]; then
systemctl restart slurmdbd &>/dev/null
systemctl is-enabled slurmdbd &>/dev/null || systemctl enable slurmdbd &>/dev/null
fi
# 配置 slurmctl 服务
nodeinfo=$(slurmd -C | head -1)
[[ -f $slurmctld_flags ]] && info "SlurmCTLD Service is configuration finisheld..." \
|| cat > ${slurm_ctld} <<-EOSLURMCTLD
ClusterName=cluster
ControlMachine=`hostname -s`
AuthType=auth/munge
SlurmUser=slurm
SlurmdUser=root
SlurmctldPort=6817
SlurmdPort=6818
StateSaveLocation=/var/spool/slurm/ctld
SlurmdSpoolDir=/var/spool/slurm/d
SlurmctldPidFile=/var/run/slurmctld.pid
SlurmdPidFile=/var/run/slurmd.pid
ProctrackType=proctrack/linuxproc
SchedulerType=sched/backfill
SelectType=select/cons_tres
SelectTypeParameters=CR_Core
AccountingStorageType=accounting_storage/slurmdbd
AccountingStorageHost=localhost
AccountingStoragePort=6819
MailProg=/bin/true
${nodeinfo} State=UNKNOWN
PartitionName=control Nodes=cluster Default=YES MaxTime=INFINITE State=UP
EOSLURMCTLD
touch $slurmctld_flags
if [[ ! -f $slurmctld_flags ]]; then
systemctl restart slurmctld &>/dev/null
systemctl is-enabled slurmctld &>/dev/null || systemctl enable slurmctld &>/dev/null
fi
echo "$SSEP"
/usr/bin/sacctmgr show clusters format=cluster,controlhost,controlport,qos
#/usr/bin/sacctmgr show clusters format=cluster,controlhost,controlport --noheader
echo "$SSEP"
/usr/bin/sinfo -N -l | tail -n +2
echo "$SSEP"
}
# ====================== 核心函数：节点解析 ======================
load_nodes_from_file() {
local nodes_file="./nodes"
# 如果文件不存在，使用默认内置节点
if [[ ! -f "${nodes_file}" ]]; then
log_debug "节点文件 ${nodes_file} 不存在，使用内置默认节点"
return
fi
# 读取文件，去空行、去注释、去重、去前后空格
mapfile -t lines < "${nodes_file}"
DEFAULT_NODES=()
for line in "${lines[@]}"; do
# 跳过空行/空白行
[ -z "${line// /}" ] && continue
# 把一行切成数组(按空格分割)
cols=($line)
node="${cols[0]}" # 节点名称(必须)
port="${cols[1]:-22}" # 端口不存在 -> 默认 22
addr="${cols[2]:-}" # IP 不存在 -> 默认空字符串
# 按空格分割成数组
GLOBAL_PORT["$node"]=${port}
# 去掉空格 + 跳过空行/注释行
clean_line=$(echo "${node}" | xargs)
[[ -z "${clean_line}" || "${clean_line}" =~ ^# ]] && continue
DEFAULT_NODES+=("${node}")
done
log_info "成功从 nodes 文件加载节点：${#DEFAULT_NODES[@]} 个"
log_debug "nodes 文件节点列表：${DEFAULT_NODES[*]}"
}
parse_nodes() {
local node_val="$1"
GLOBAL_NODES=()
if [[ "$node_val" == "all" ]]; then
GLOBAL_NODES=("${DEFAULT_NODES[@]}")
else
IFS=',' read -ra tmp <<< "$node_val"
GLOBAL_NODES=("${tmp[@]}")
fi
GLOBAL_OPTS[nodes]="$node_val"
}
# 检查节点状态（online/offline/auth/pass）
check_node_status() {
local node="$1"
port=${GLOBAL_PORT[$node]}
# 检查网络连通性
if ! ping -c 1 -W 1 "$node" &>/dev/null; then
echo "offline"
return
fi
# 检查是否是本机
localhost=$(hostname -s)
if [[ "$node" == "localhost" || "$node" == $localhost ]]; then
echo "localhost"
return
fi
# 检查SSH免密登录
if ssh -o ConnectTimeout=1 -o BatchMode=yes -o StrictHostKeyChecking=accept-new -p $port "$node" true &>/dev/null; then
echo "online,auth"
else
echo "online,pass"
fi
}
list_node() {
echo "$SSEP"
printf "${G}%-12s %-15s %-8s %s${NC}\n" "Nodename" "Status" "Port" "Days"
echo "$SSEP"
for node in "${GLOBAL_NODES[@]}"; do
status=$(check_node_status "$node")
port=${GLOBAL_PORT[$node]}
if [[ "$status" == "offline" ]]; then
uptime_days="---"
else
uptime_days=$(ssh -o ConnectTimeout=1 -o BatchMode=yes -p$port "$node" "uptime \
| awk -F'up|days' '{print \$2}' | tr -d ',' | xargs" 2>/dev/null || echo "---")
fi
printf "%-12s %-15s %-8s %s\n" "$node" "$status" "$port" "$uptime_days"
done
echo "$SSEP"
}
run_cmd() {
local node="$1"
local cmds="$2"
local output=""
local exit_code=0
# 获取本机主机名（自动识别管理节点）
local localhost=$(hostname -s)
if [[ "$node" == "$localhost" || "$node" == "localhost" ]]; then
# ====================== 本机：直接执行 ======================
output=$(eval "$cmds" 2>&1)
exit_code=$?
else
# ====================== 远程：SSH 执行 ======================
output=$(ssh -o ConnectTimeout=1 -o BatchMode=yes "$node" "$cmds" 2>&1)
exit_code=$?
fi
if [[ $exit_code -eq 0 ]]; then
info "执行成功: ${node}"
else
warn "执行失败: ${node} (${R}Code${NC}: ${R}$exit_code${NC})"
fi
echo "$DSEP"
echo "$output"
}
# ====================== 核心函数: 同步文件路径处理 ======================
sync_file_path() {
local src_file="$1"
# ======================== 核心规则 ========================
# 1. 如果源是 绝对路径（/开头），直接返回原路径（系统目录）
if [[ "$src_file" == /* ]]; then
echo "$src_file"
return
fi
# 2. 如果源是 相对路径（linux/alls/* 或 linux/节点/*）
# 自动去掉 linux/xxx/ 前缀，剩余部分自动变成 绝对路径（前面加 /）
# 支持：etc usr opt var home root 任意目录，无限扩展
# ==================================================
local rel_path
rel_path=$(echo "$src_file" | sed -E 's#^linux/(alls|[-_a-zA-Z0-9]+)/##')
# 拼接为绝对路径
echo "/$rel_path"
}
sync_file_all() {
load_nodes "$NODE_ARG"
# ===================== 美化打印节点列表（固定缩进 + 自动换行） =====================
local indent=" " # 11个空格
local line=""
local count=0
local max_per_line=5 # 每行显示几个节点
info "开始同步所有节点文件，节点列表："
for node in "${NODES[@]}"; do
line+="$node "
count=$((count + 1))
if [[ $count -ge $max_per_line ]]; then
echo -e "${indent}${line}"
line=""
count=0
fi
done
# 打印剩余不足一行的节点
if [[ -n "$line" ]]; then
echo -e "${indent}${line}"
fi
# ================================================================================
for node_dir in linux/*; do
[[ -d "$node_dir" ]] || continue
for sub_dir in "$node_dir"/*; do
[[ -d "$sub_dir" ]] || continue
info "处理源目录：$sub_dir"
for node in "${NODES[@]}"; do
if [[ "$node_dir" == "linux/alls" || "$node_dir" == "linux/$node" ]]; then
target=$(sync_file_path "$sub_dir")
info "同步 ${Y}$sub_dir${NC} -> ${R}$node${NC}:${Y}$target${NC}"
rsync -avzAP --info=progress2 --exclude='.*' "$sub_dir"/ "$node":"$target"/
fi
done
done
done
info "同步完成！"
}
sync_file_single() {
local src_file="$1"
[[ -z "$src_file" ]] && error "请使用 -f 指定文件路径"
[[ -e "$src_file" ]] || error "文件不存在：$src_file"
load_nodes "$NODE_ARG"
local indent=" " # 11个空格
local line=""
local count=0
local max_per_line=5 # 每行显示几个节点
info "开始同步所有节点文件，节点列表："
for node in "${NODES[@]}"; do
line+="$node "
count=$((count + 1))
if [[ $count -ge $max_per_line ]]; then
echo -e "${indent}${line}"
line=""
count=0
fi
done
# 打印剩余不足一行的节点
if [[ -n "$line" ]]; then
echo -e "${indent}${line}"
fi
target=$(sync_file_path "$src_file")
for node in "${NODES[@]}"; do
info "同步 ${Y}$src_file${NC} -> ${R}$node${NC}:${Y}$target${NC}"
rsync -avzAP --info=progress2 "$src_file" "$node":"$target"
done
info "同步完成！"
}
# ====================== 帮助文档 ======================
show_help() {
cat <<EOF
SunHPC 命令行工具
Usage: sunhpc <COMMAND> [OPTIONS]
公共参数：
-v, --verbose 详细输出
-d, --debug 调试模式
-h, --help 帮助信息
-n, --nodes <list> 指定节点（all 或 node1,node2,node3）
支持命令：
sunhpc pxe dhcp 部署dhcp服务
sunhpc pxe tftp 部署tftp服务
sunhpc pxe http 部署http服务
sunhpc pxe dnsmasq 部署pxe服务
sunhpc exec 批量执行命令
sunhpc sync file 执行同步任务
sunhpc list node 查看节点列表
sunhpc show vimrc 显示Vimrc变量
sunhpc show system 显示系统信息
sunhpc serv slurm 构建 slurm 服务
sunhpc serv autofs 构建 autofs 服务
sunhpc create syncfile 显示系统信息
Report bugs to: <support@sunhpc.com>
Htool URL: https://gitea.sunhpc.com
Gitea URL: https://gitea.sunhpc.com/htool.git
EOF
}
show_help_sync() {
cat <<EOF
SunHPC 命令行工具
Usage: sunhpc <COMMAND> [OPTIONS]
公共参数：
-v, --verbose 详细输出
-d, --debug 调试模式
-h, --help 帮助信息
-n, --nodes <list> 指定节点（all 或 node1,node2,node3）
命令示例:
sunhpc sync file -f/--file <filepath> 同步单个文件
sunhpc sync file -a/--all 同步所有文件
Report bugs to: <support@sunhpc.com>
Htool URL: https://gitea.sunhpc.com
Gitea URL: https://gitea.sunhpc.com/htool.git
EOF
}
show_help_slurm() {
cat <<EOF
SunHPC 命令行工具
Usage: sunhpc <COMMAND> [OPTIONS]
公共参数：
-v, --verbose 详细输出
-d, --debug 调试模式
-h, --help 帮助信息
-n, --nodes <list> 指定节点（all 或 node1,node2,node3）
命令示例:
sunhpc serv slurm -b/--build 构建Slurm安装包
sunhpc serv slurm -i/--install -m/master 安装Slurm到管理节点
sunhpc serv slurm -i/--install -c/compute 安装Slurm到计算节点
Report bugs to: <support@sunhpc.com>
Htool URL: https://gitea.sunhpc.com
Gitea URL: https://gitea.sunhpc.com/htool.git
EOF
}
# ====================== Verbose/Debug 输出工具 ======================
log_info() {
if [[ -n "${GLOBAL_OPTS[v]}" || -n "${GLOBAL_OPTS[verbose]}" ]]; then
echo -e "\033[32m[I]\033[0m $1"
fi
}
log_debug() {
if [[ -n "${GLOBAL_OPTS[d]}" || -n "${GLOBAL_OPTS[debug]}" ]]; then
echo -e "\033[33m[D]\033[0m $1"
fi
}
# ====================== 子命令执行 ======================
run_pxe_dhcp() {
log_debug "--------- run_pxe_dhcp----------------"
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
# -n 非空True，-z 为空True
[[ -n ${GLOBAL_OPTS[interface]} ]] && pxe_dhcp_build || \
warn "请指定DHCP监听接口命名, 例如: sunhpc pxe dhcp -i eth0"
}
run_pxe_tftp() {
log_debug "--------- run_pxe_tftp----------------"
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
pxe_tftp_build
}
run_pxe_http() {
log_debug "--------- run_pxe_http----------------"
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
pxe_http_build
}
run_init() {
[[ -n "${GLOBAL_OPTS[interface]}" ]] && iface="${GLOBAL_OPTS[interface]}" || \
error "必须指定一个网络接口,例如: -i/--interface eth1"
[[ -n "${GLOBAL_OPTS[html]-}" ]] && html="${GLOBAL_OPTS[html]}" || \
error "必须指定 html 路径,例如: --html /var/www/html"
ipaddr=${global_env[${iface}_ip]}
netcidr=${global_env[${iface}_mask]}
netmask=$(cidr_to_netmask 24)
subnet=$(get_subnet_address $ipaddr $netmask)
IFS=. read -r i1 i2 i3 i4 <<< "$ipaddr"
start_ip="$i1.$i2.$i3.10"
ender_ip="$i1.$i2.$i3.90"
tmpdir=$(find -L $html -name ".treeinfo")
tmpdir=$(dirname ${tmpdir#$html})
echo "iface---------> $iface"
echo "addr----------> $ipaddr"
echo "startip-------> $start_ip"
echo "enderip-------> $ender_ip"
echo "htmlroot------> $html"
echo "reporoot------> $tmpdir"
exit 1
{
echo "interface $iface"
echo "server $ipaddr"
echo "range $start_ip,$ender_ip"
echo "htmlroot $html"
echo "reporoot $tmpdir"
} > ${config_file}
#gen_config_file
}
run_exec() {
log_debug "--------- run_exec----------------"
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
local cmds=$1
[[ -z "$cmds" ]] && error "请指定要执行的命令，例如：sunhpc exec \"df -h\""
echo "$DSEP"
info "执行命令: $cmds"
localhost=$(hostname -s)
for node in "${GLOBAL_NODES[@]}"; do
if [[ "$node" != "$localhost" || "$node" != "localhost" ]]; then
status=$(check_node_status "$node")
if [[ "$status" == "offline" ]]; then
warn "$node 节点离线，跳过执行"
continue
fi
if [[ "$status" == "online,pass" ]]; then
warn "$node 需要密码登录，跳过执行"
continue
fi
fi
# ========== 核心：自动本机/远程执行 ==========
run_cmd "$node" "$cmds"
done
echo "$SSEP"
info "命令执行完成！"
echo "$SSEP"
}
run_sync() {
log_debug "--------- run_sync----------------"
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
case "${PRIVATE_ARGS[@]}" in
-a|--all) sync_file_all ;;
-f|--file) sync_file_single ;;
*) show_help_sync ;;
esac
}
run_list_node() {
log_debug "--------- run_list_node----------------"
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
log_info "执行 list node"
log_info "当前节点列表：${GLOBAL_NODES[*]}"
list_node
}
run_show_vimrc() {
cat <<-EOF
" 开启语法高亮
syntax enable
" 设置缩进为4个空格
set tabstop=4
set softtabstop=4
set shiftwidth=4
" 用空格替代Tab键（禁止输入真实Tab）
set expandtab
" 自动缩进（写代码更方便）
set autoindent
set smartindent
EOF
}
run_serv_slurm() {
log_debug "--------- run_serv_slurm------------------"
case "${PRIVATE_ARGS[@]}" in
-b|--build) slurm_build ;;
*) show_help_slurm ;;
esac
}
run_serv_autofs() {
log_debug "--------- run_serv_autofs----------------"
log_debug "当前节点：${GLOBAL_NODES[*]}"
log_debug "公共参数：${PUBLIC_ARGS[*]}"
log_debug "私有参数：${PRIVATE_ARGS[*]}"
log_info "执行 server autofs"
if [[ " ${PRIVATE_ARGS[@]} " =~ " --build " ]]; then
log_debug "检测到 --build 参数，开始构建..."
echo "✅ 正在构建 autofs 服务"
fi
}
create_sync_file() {
dirspath=(
"/etc/passwd"
"/etc/shadow"
"/etc/group"
"/etc/slurm"
"/etc/systemd/system"
"/usr/lib/systemd/system"
)
local base="${1:-.}"
local name="${2:-linux}"
local all_dir="${base}/${name}/all"
mkdir -p "${all_dir}" || return 1
for d in "${dirspath}"; do
if [[ -f $d ]]; then
src_dir=$(dirname "$d")
dst_dir="${all_dir}/$src_dir"
mkdir -p "${dst_dir}"
/usr/bin/cp $d ${dst_dir}
else
mkdir -p "${all_dir}/${d}"
fi
done
for n in ${GLOBAL_NODES[@]}; do
mkdir -p "${base}/${name}/${n}"
done
}
# ====================== 命令路由 ======================
dispatch() {
# 优先显示帮助
if [[ -n "${GLOBAL_OPTS[h]}" || -n "${GLOBAL_OPTS[help]}" || ${#CMD_STACK[@]} -eq 0 ]]; then
show_help
exit 0
fi
log_debug "$DSEP"
log_debug "解析子命令：${CMD_STACK[*]}"
log_debug "专属参数 ：${PRIVATE_ARGS[*]}"
case "${CMD_STACK[0]}" in
pxe)
case "${CMD_STACK[1]}" in
dhcp) run_pxe_dhcp ;;
tftp) run_pxe_tftp;;
http) run_pxe_http;;
dnsmasq) pxe_dnsmasq_build;;
*) show_help;;
esac
;;
exec)
run_exec "${CMD_STACK[1]}"
;;
sync)
case "${CMD_STACK[1]}" in
file) run_sync ;;
*) show_help ;;
esac
;;
list)
case "${CMD_STACK[1]}" in
node) run_list_node ;;
*) show_help ;;
esac
;;
show)
case "${CMD_STACK[1]}" in
vimrc) run_show_vimrc ;;
system) show_system_summary ;;
*) show_help;;
esac
;;
serv)
case "${CMD_STACK[1]}" in
slurm) run_serv_slurm ;;
autofs) run_serv_autofs ;;
*) show_help;;
esac
;;
create)
case "${CMD_STACK[1]}" in
syncfile) create_sync_file ;;
*) show_help ;;
esac
;;
*) show_help ;;
esac
}
# ====================== 核心函数：参数解析 ======================
parse_args() {
while [[ $# -gt 0 ]]; do
local arg="$1"
case "$arg" in
# 公共短参数：-v -d -h -n
-v) GLOBAL_OPTS[v]=1; GLOBAL_OPTS[verbose]=1; PUBLIC_ARGS+=("$arg"); shift ;;
-d) GLOBAL_OPTS[d]=1; GLOBAL_OPTS[debug]=1; PUBLIC_ARGS+=("$arg"); shift ;;
-h) GLOBAL_OPTS[h]=1; GLOBAL_OPTS[help]=1; PUBLIC_ARGS+=("$arg"); shift ;;
-f) GLOBAL_OPTS[file]=$2; PUBLIC_ARGS+=("$arg" "$2"); shift 2 ;;
-i) GLOBAL_OPTS[interface]=$2; PUBLIC_ARGS+=("$arg" "$2"); shift 2 ;;
-n) parse_nodes "$2"; PUBLIC_ARGS+=("$arg" "$2"); shift 2 ;;
# 公共长参数：--verbose --debug --help --nodes
--verbose) GLOBAL_OPTS[verbose]=1; GLOBAL_OPTS[v]=1; PUBLIC_ARGS+=("$arg"); shift ;;
--debug) GLOBAL_OPTS[debug]=1; GLOBAL_OPTS[d]=1; PUBLIC_ARGS+=("$arg"); shift ;;
--help) GLOBAL_OPTS[help]=1; GLOBAL_OPTS[h]=1; PUBLIC_ARGS+=("$arg"); shift ;;
--file) GLOBAL_OPTS[file]=$2; PUBLIC_ARGS+=("$arg" "$2"); shift 2 ;;
--html) GLOBAL_OPTS[html]=$2; PUBLIC_ARGS+=("$arg" "$2"); shift 2 ;;
--interface) GLOBAL_OPTS[interface]=$2; PUBLIC_ARGS+=("$arg" "$2"); shift 2 ;;
--nodes) parse_nodes "$2"; PUBLIC_ARGS+=("$arg" "$2"); shift 2 ;;
# -- 专属短参数（子命令私有）
-*) PRIVATE_ARGS+=("$arg"); shift ;;
# -- 专属长参数（子命令私有）
--*) PRIVATE_ARGS+=("$arg"); shift ;;
# 子命令（无 - / --）
*) CMD_STACK+=("$arg"); shift ;;
esac
done
}
# ====================== 主入口 ======================
main() {
# 配置文件
config_file="htool.conf"
# load system env
load_os_envs
# 启动时先加载 nodes 文件
load_nodes_from_file
# 默认使用文件中的节点
GLOBAL_NODES=("${DEFAULT_NODES[@]}")
parse_args "$@"
# 获取基础信息存储变量
if [[ -e $config_file ]]; then
load_conf_envs
else
run_init
fi
dispatch
}
main "$@"
Reference in New Issue View Git Blame Copy Permalink