177 lines
4.3 KiB
Bash
177 lines
4.3 KiB
Bash
#!/bin/bash
|
||
set -u
|
||
|
||
# Usage: gsub <jobname>
|
||
|
||
job=${1:-}
|
||
if [[ -z "$job" ]]; then
|
||
echo "Usage: $0 <jobname-without-extension>"
|
||
exit 1
|
||
fi
|
||
|
||
# ==========================================
|
||
# 0. 安全检测函数 (Safety Check)
|
||
# ==========================================
|
||
check_dangerous_path() {
|
||
local path="${1:-}"
|
||
|
||
# 1. Empty check
|
||
if [[ -z "$path" ]]; then
|
||
echo "Error: Empty path is dangerous for deletion." >&2
|
||
return 1
|
||
fi
|
||
|
||
# 2. Root check
|
||
if [[ "$path" == "/" ]]; then
|
||
echo "Error: Root path '/' is dangerous for deletion." >&2
|
||
return 1
|
||
fi
|
||
|
||
# 3. Space check (optional, but good for safety)
|
||
if [[ "$path" =~ ^[[:space:]]+$ ]]; then
|
||
echo "Error: Whitespace path is dangerous." >&2
|
||
return 1
|
||
fi
|
||
|
||
return 0
|
||
}
|
||
|
||
# ==========================================
|
||
# 1. 检查运行环境 (Check Host)
|
||
# ==========================================
|
||
# 如果不是 cluster，尝试通过 SSH 远程调用
|
||
host_short=$(hostname -s 2>/dev/null || hostname)
|
||
if [[ "$host_short" != "cluster" ]]; then
|
||
# 假设本地挂载路径 /mnt/home 对应远程 /home (根据原脚本逻辑调整)
|
||
cur_dir=$(pwd)
|
||
remote_dir="${cur_dir//\/mnt\/home/\/home}"
|
||
|
||
# 定位当前脚本并转换为远程路径
|
||
# 获取脚本所在目录的绝对路径
|
||
script_dir=$(cd "$(dirname "$0")" && pwd)
|
||
script_name=$(basename "$0")
|
||
local_script="$script_dir/$script_name"
|
||
|
||
# 同样对脚本路径进行替换
|
||
remote_script="${local_script//\/mnt\/home/\/home}"
|
||
|
||
# 尝试在远程执行自己
|
||
echo "Running remotely on cluster: $remote_script" >&2
|
||
ssh cluster "cd '$remote_dir' && '$remote_script' '$job'"
|
||
exit $?
|
||
fi
|
||
|
||
# ==========================================
|
||
# 2. 准备作业 (Prepare Job)
|
||
# ==========================================
|
||
|
||
gin_file="$job.gin"
|
||
if [[ ! -f "$gin_file" ]]; then
|
||
echo "Error: $gin_file not found in $(pwd)"
|
||
exit 2
|
||
fi
|
||
|
||
# 解析配置确定资源 (Parse Proc)
|
||
# 查找 %NProcShared=XX
|
||
proc=$(sed -n 's/^%NProcShared=\([0-9]\+\).*$/\1/pI' "$gin_file" | head -n 1)
|
||
|
||
queue=""
|
||
ppn=""
|
||
|
||
if [[ "$proc" == "32" ]]; then
|
||
queue="n32"
|
||
ppn="32"
|
||
elif [[ "$proc" == "20" ]]; then
|
||
queue="n20"
|
||
ppn="20"
|
||
else
|
||
echo "Error: Unsupported NProcShared=$proc in $gin_file. Only 20 or 32 allowed."
|
||
exit 1
|
||
fi
|
||
|
||
# 清理旧文件 (Clean up old output)
|
||
if [[ -f "$job.out" ]]; then
|
||
# 原脚本逻辑：休眠并删除
|
||
# echo "Warning: $job.out exists. Deleting..." >&2
|
||
# 使用安全检查
|
||
if check_dangerous_path "$job.out"; then
|
||
rm "$job.out"
|
||
else
|
||
echo "Skipping deletion of unsafe path: $job.out" >&2
|
||
exit 1
|
||
fi
|
||
fi
|
||
|
||
# ==========================================
|
||
# 3. 生成作业脚本 (.job)
|
||
# ==========================================
|
||
job_file="$job.job"
|
||
|
||
# 使用 heredoc 动态生成 PBS 脚本
|
||
# 整合了原 g16_32.pbs 的内容和 gsub32 的追加内容
|
||
cat > "$job_file" <<EOF
|
||
#!/bin/sh
|
||
#PBS -l nodes=1:ppn=$ppn
|
||
#PBS -q $queue
|
||
#PBS -j oe
|
||
#PBS -N $job
|
||
|
||
cd \$PBS_O_WORKDIR
|
||
|
||
# Define Safety Check Function in Job Script
|
||
check_rm_path() {
|
||
p="\$1"
|
||
# Empty check
|
||
if [ -z "\$p" ]; then
|
||
echo "Refusing to delete empty path"
|
||
return 1
|
||
fi
|
||
# Root check
|
||
if [ "\$p" = "/" ]; then
|
||
echo "Refusing to delete root path"
|
||
return 1
|
||
fi
|
||
return 0
|
||
}
|
||
|
||
export g16root=/share/apps/soft
|
||
source \$g16root/g16/bsd/g16.profile
|
||
|
||
# Create Scratch Directory
|
||
if [ -n "\$USER" ] && [ -n "\$PBS_JOBID" ]; then
|
||
mkdir -p /scr/\$USER/\$PBS_JOBID
|
||
export GAUSS_SCRDIR=/scr/\$USER/\$PBS_JOBID
|
||
else
|
||
echo "Error: USER or PBS_JOBID not set. Cannot setup scratch."
|
||
exit 1
|
||
fi
|
||
|
||
NODES=\`cat \$PBS_NODEFILE | uniq\`
|
||
echo "--------------------------------------------------------"
|
||
echo " JOBID: \$PBS_JOBID"
|
||
echo " The job was started at \`date\`"
|
||
echo " The job was running at \$NODES."
|
||
echo "--------------------------------------------------------"
|
||
|
||
# Run G16 Job
|
||
echo "Executing: g16 < $gin_file > $job.out"
|
||
g16 < $gin_file > $job.out
|
||
|
||
echo "--------------------------------------------------------"
|
||
echo " The job was finished at \`date\`"
|
||
echo "--------------------------------------------------------"
|
||
|
||
# Delete the tmp File (Cleanup Scratch)
|
||
echo "Cleaning up \$GAUSS_SCRDIR"
|
||
if check_rm_path "\$GAUSS_SCRDIR"; then
|
||
rm -rf "\$GAUSS_SCRDIR"
|
||
fi
|
||
|
||
EOF
|
||
|
||
# ==========================================
|
||
# 4. 提交作业 (Submit)
|
||
# ==========================================
|
||
# qsub 会输出 Job ID，例如 12345.cluster
|
||
qsub "$job_file"
|