#!/bin/bash
# 优化版：监控 gunicorn 主进程及其子进程资源变化（表格+彩色输出）

PROC_NAME="gunicorn"
LOG_DIR="/data/sh/monitor/logs"
LOG_FILE="$LOG_DIR/proc_monitor.log"
DATA_FILE="$LOG_DIR/last_snapshot.txt"

mkdir -p "$LOG_DIR"

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

# 获取系统内存概况
get_sys_mem() {
  free -h | awk '/Mem:/ {printf "系统内存概况：总 %s | 已用 %s | 空闲 %s | 可用 %s", $2, $3, $4, $7}'
}

# 获取时间
format_time() {
  date "+%Y-%m-%d %H:%M:%S"
}

# 获取 gunicorn 主进程和子进程信息
get_proc_info() {
  ps -eo pid,ppid,comm,%mem,%cpu,rss --no-headers | awk -v pname="$PROC_NAME" '$3 == pname {print $1, $2, $3, $4, $5, $6}'
}

# 保存当前状态快照
save_snapshot() {
  get_proc_info | awk '{print $1,$3,$4,$5}' > "$DATA_FILE"
}

# 打印表格头
print_table_header() {
  printf "%-10s %-12s %-12s %-12s %-12s\n" "PID" "进程名" "内存(Gi)" "CPU(%)" "备注"
}

# 打印分隔线
print_line() {
  printf "%s\n" "---------------------------------------------------------------"
}

# 获取主进程 PID
MAIN_PID=$(pgrep -f "$PROC_NAME" | head -n 1)
if [ -z "$MAIN_PID" ]; then
  echo -e "${RED}未找到进程：$PROC_NAME${NC}"
  exit 1
fi

TIMESTAMP=$(format_time)
echo "==== $TIMESTAMP ====" >> "$LOG_FILE"
echo "$(get_sys_mem)" >> "$LOG_FILE"
echo "主进程名：$PROC_NAME | PID: $MAIN_PID" >> "$LOG_FILE"

# 写表格头
print_table_header >> "$LOG_FILE"
print_line >> "$LOG_FILE"

# 获取当前进程信息
CUR_INFO=$(get_proc_info)

echo "$CUR_INFO" | while read -r pid ppid name mem cpu rss; do
  mem_gi=$(awk "BEGIN {printf \"%.2f\", $rss/1024/1024}") # KB -> Gi
  printf "%-10s %-12s %-12s %-12s\n" "$pid" "$name" "${mem_gi}Gi" "${cpu}%" >> "$LOG_FILE"
done

print_line >> "$LOG_FILE"

# 若无上次记录，跳过变化分析
if [ ! -f "$DATA_FILE" ]; then
  echo "暂无上次记录，跳过变化分析。" >> "$LOG_FILE"
  save_snapshot
  echo "==== $TIMESTAMP ====" >> "$LOG_FILE"
  exit 0
fi

# 对比分析部分
declare -A CUR_MEM CUR_CPU PRE_MEM PRE_CPU

while read -r pid name mem cpu; do
  CUR_MEM["$pid"]=$mem
  CUR_CPU["$pid"]=$cpu
done < <(get_proc_info | awk '{print $1,$3,$4,$5}')

while read -r pid name mem cpu; do
  PRE_MEM["$pid"]=$mem
  PRE_CPU["$pid"]=$cpu
done < "$DATA_FILE"

CUR_PIDS=$(printf "%s\n" "${!CUR_MEM[@]}" | sort)
PRE_PIDS=$(printf "%s\n" "${!PRE_MEM[@]}" | sort)

NEW_PIDS=$(comm -23 <(echo "$CUR_PIDS") <(echo "$PRE_PIDS"))
OLD_PIDS=$(comm -13 <(echo "$CUR_PIDS") <(echo "$PRE_PIDS"))

echo "分析变化：" >> "$LOG_FILE"

# 新旧进程分析
if [ -n "$NEW_PIDS" ]; then
  echo -e "  ${GREEN}新出现的子进程:${NC}" >> "$LOG_FILE"
  for pid in $NEW_PIDS; do
    echo "    $PROC_NAME (PID:$pid)" >> "$LOG_FILE"
  done
fi

if [ -n "$OLD_PIDS" ]; then
  echo -e "  ${YELLOW}已消失的子进程:${NC}" >> "$LOG_FILE"
  for pid in $OLD_PIDS; do
    echo "    $PROC_NAME (PID:$pid)" >> "$LOG_FILE"
  done
fi

# 资源变化分析
echo "  子进程资源变化：" >> "$LOG_FILE"
for pid in $CUR_PIDS; do
  if [ -n "${PRE_MEM[$pid]}" ]; then
    cpu_diff=$(awk "BEGIN {printf \"%.1f\", ${CUR_CPU[$pid]} - ${PRE_CPU[$pid]}}")
    mem_diff=$(awk "BEGIN {printf \"%.2f\", ${CUR_MEM[$pid]} - ${PRE_MEM[$pid]}}")

    color_start=""
    color_end="$NC"
    if (( $(echo "$mem_diff > 0.2" | bc -l) )); then
      color_start=$RED
    elif (( $(echo "$mem_diff < -0.2" | bc -l) )); then
      color_start=$GREEN
    else
      color_start=$YELLOW
    fi

    echo -e "    ${color_start}$PROC_NAME (PID:$pid): CPU变化 ${cpu_diff}% | 内存变化 ${mem_diff}%${color_end}" >> "$LOG_FILE"
  fi
done

# 当前最高占用进程
top_cpu_pid=$(ps -eo pid,comm,%cpu --no-headers | grep "$PROC_NAME" | sort -k3 -nr | head -n 1 | awk '{print $1}')
top_mem_pid=$(ps -eo pid,comm,%mem --no-headers | grep "$PROC_NAME" | sort -k3 -nr | head -n 1 | awk '{print $1}')

if [ -n "$top_cpu_pid" ]; then
  echo -e "  ${BLUE}当前占用最高CPU的进程:${NC} $PROC_NAME (PID:$top_cpu_pid)" >> "$LOG_FILE"
fi
if [ -n "$top_mem_pid" ]; then
  echo -e "  ${BLUE}当前占用最高内存的进程:${NC} $PROC_NAME (PID:$top_mem_pid)" >> "$LOG_FILE"
fi

save_snapshot
echo "==== $TIMESTAMP ====" >> "$LOG_FILE"