sbatch.sh

#!/bin/bash
# ==========================================================
# 用法：
#   sbatch submit_job.sh <gpu_num> <job_name> [debug|long]
#
# 参数说明：
#   <gpu_num>    : GPU 数量（必填，如 0 / 1 / 2）
#   <job_name>   : 任务名称（必填）
#   [debug|long] : 可选，debug=2小时，long=24小时，默认12小时
#
# 资源规则：
#   每块 GPU 分配 16 个 CPU 核
#   每个 CPU 核分配 8G 内存
#   若 GPU=0，则默认申请 16C + 128G
#
# 示例：
#   sbatch submit_job.sh 1 train-job
#   sbatch submit_job.sh 2 bigjob long
#   sbatch submit_job.sh 1 debug-test debug
#   sbatch submit_job.sh 0 cpuonly
# ==========================================================

GPU_NUM="$1"
JOB_NAME="$2"
MODE="$3"

# 参数检查
if [[ -z "$GPU_NUM" || -z "$JOB_NAME" ]]; then
    echo "❌ 参数错误！"
    echo "用法: sbatch submit_job.sh <gpu_num> <job_name> [debug|long]"
    exit 1
fi

if ! [[ "$GPU_NUM" =~ ^[0-9]+$ ]]; then
    echo "❌ 错误：gpu_num 必须是非负整数"
    exit 1
fi

# 默认参数（12小时）
PARTITION="q-dyvm6xra"
QOS="qos_g5_12hrs"
TIME_LIMIT="12:00:00"

if [[ "$MODE" == "debug" ]]; then
    PARTITION="q-g7hdnrvb"
    QOS=""
    TIME_LIMIT="2:00:00"
elif [[ "$MODE" == "long" ]]; then
    PARTITION="q-dyvm6xra"
    QOS="qos_g5_24hrs"
    TIME_LIMIT="24:00:00"
elif [[ -n "$MODE" ]]; then
    echo "❌ 错误：第三个参数只能是 debug 或 long"
    exit 1
fi

# 资源配置
CPU_PER_GPU=16
MEM_PER_CPU=8G

if [[ "$GPU_NUM" -eq 0 ]]; then
    TOTAL_CPU=16
    TOTAL_MEM="128G"
    GPU_DIRECTIVE=""
else
    TOTAL_CPU=$(( GPU_NUM * CPU_PER_GPU ))
    TOTAL_MEM="$(( TOTAL_CPU * 8 ))G"
    GPU_DIRECTIVE="#SBATCH --gres=gpu:${GPU_NUM}"
fi

# 输出信息
echo "=========================================================="
echo "✅ 提交 sbatch 参数："
echo "GPU 数量     : ${GPU_NUM}"
echo "CPU 数量     : ${TOTAL_CPU}"
echo "总内存       : ${TOTAL_MEM}"
echo "任务名称     : ${JOB_NAME}"
echo "分区         : ${PARTITION}"
echo "时间限制     : ${TIME_LIMIT}"
echo "QOS          : ${QOS:-"-"}"
echo "=========================================================="

# 生成临时 sbatch 脚本并提交
TMP_SCRIPT=$(mktemp /tmp/${JOB_NAME}.XXXXXX.sbatch)

cat > "$TMP_SCRIPT" <<EOF
#!/bin/bash
#SBATCH -p ${PARTITION}
#SBATCH -A dyvm6xra
#SBATCH -J ${JOB_NAME}
#SBATCH -c ${TOTAL_CPU}
#SBATCH --mem=${TOTAL_MEM}
#SBATCH -t ${TIME_LIMIT}
#SBATCH -o ${JOB_NAME}.%j.out
#SBATCH -e ${JOB_NAME}.%j.err
EOF

if [[ -n "$QOS" ]]; then
    echo "#SBATCH -q ${QOS}" >> "$TMP_SCRIPT"
fi

if [[ -n "$GPU_DIRECTIVE" ]]; then
    echo "${GPU_DIRECTIVE}" >> "$TMP_SCRIPT"
fi

cat >> "$TMP_SCRIPT" <<'EOF'

echo "Job started on $(hostname)"
echo "Start time: $(date)"
echo "CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES"

# ===== 在这里写你的实际任务命令 =====
# 例子：
# cd /path/to/your/project
# source ~/.bashrc
# conda activate your_env
# python train.py

cd /home/dyvm6xra/dyvm6xrauser11/workspace/cz/FlowCache/FlowCache4MAGI-1
source ~/.bashrc
conda activate magi
bash scripts/sample/flowcache_vbench.sh

echo "请把实际运行命令写到脚本里"
sleep 60

echo "End time: $(date)"
EOF

sbatch "$TMP_SCRIPT"