#!/usr/bin/env bash
# =============================================================================
# 一键启动(防 SIGHUP/外部kill 脱离终端 + 调好参数 + 启动前自检 + 防重复启动)
#
# 两种模式(MODE 环境变量切换):
#   MODE=worker (默认) —— 常驻 worker, 从 COS 自动认领任务(run_worker.py)。不需要本地 text2。
#   MODE=local         —— 旧模式, 处理本地 ./text2 下的任务txt(main.py)。
#
# 用法:
#   bash scripts/启动.sh                       # 默认 worker 模式启动
#   bash scripts/启动.sh 32                     # 临时把 max-workers 改成 32
#   MODE=local bash scripts/启动.sh             # 用旧的本地 text2 模式
#   MAX_WORKERS=20 MEM_GUARD=40 bash scripts/启动.sh
#
# 停止:  bash scripts/stop.sh -y
# =============================================================================
set -u
cd "$(cd "$(dirname "$0")/.." && pwd)"
ROOT="$(pwd)"

# ---- 可调参数(命令行第1个参数 / 环境变量 均可覆盖) ----
# MODE 未显式指定时, 读 config.dispatch.source 决定: local -> 本地text2, 其余 -> COS常驻
if [ -z "${MODE:-}" ]; then
  _src="$(python -c "from config.config_utils import config; print((config.get_dispatch_config() or {}).get('source','cos'))" 2>/dev/null || echo cos)"
  [ "$_src" = "local" ] && MODE="local" || MODE="worker"
fi
MODE="${MODE:-worker}"                      # worker=COS认领常驻 / local=本地text2
MAX_WORKERS="${1:-${MAX_WORKERS:-25}}"      # 0=不传给python,由main.py/run_worker.py自动按本机CPU核数选
                                            # 共享机/跟OCR服务同机部署的, 建议显式传小一点的值(如24), 别用自动
MAX_TASKS="${MAX_TASKS:-0}"                # 0=不回收worker(避免3.11回收缩水把并发缩到0)
MEM_GUARD="${MEM_GUARD:-30}"               # 可用内存低于此(G)暂停启动新视频
SOURCE_LIST="${SOURCE_LIST:-./text2}"      # 仅 local 模式: 任务txt目录
EXECUTOR="${EXECUTOR:-process}"            # process=进程池(绕GIL) / thread
LOG="${LOG:-app.log}"
NICE="${NICE:-15}"                         # 处理进程CPU优先级(0~19,越大越低)
LOG_MAX_MB="${LOG_MAX_MB:-200}"            # $LOG 超过这个大小(MB)就在启动前轮转一次, 0=不轮转

# ---- 启动前轮转一次日志(常驻进程一跑数天/数周, >>追加写不会自己变小) ----
_rotate_log() {
  local f="$1" maxmb="$2"
  [ "$maxmb" -le 0 ] && return 0
  [ -f "$f" ] || return 0
  local sz_mb=$(( $(stat -c%s "$f" 2>/dev/null || echo 0) / 1024 / 1024 ))
  if [ "$sz_mb" -ge "$maxmb" ]; then
    local ts; ts="$(date +%Y%m%d_%H%M%S)"
    mv "$f" "${f}.${ts}" 2>/dev/null && echo "[日志轮转] $f (${sz_mb}M) → ${f}.${ts}, 重新开始写"
    ls -t "${f}".[0-9]* 2>/dev/null | tail -n +6 | xargs -r rm -f   # 只留最近5份轮转备份
  fi
}
_rotate_log "$LOG" "$LOG_MAX_MB"

# 按模式确定入口与进程匹配模式
if [ "$MODE" = "worker" ]; then
  ENTRY="run_worker.py"
  PROC_PAT='python .*run_worker\.py'
else
  ENTRY="main.py"
  PROC_PAT='python .*main\.py'
fi

MAX_WORKERS_DISP="$MAX_WORKERS"
[ "$MAX_WORKERS" = "0" ] && MAX_WORKERS_DISP="0(自动按CPU核数选, 实际值见下面python启动日志)"

echo "=================================================="
echo " 启动 $ENTRY  (MODE=$MODE)"
echo "   max-workers=$MAX_WORKERS_DISP  max-tasks-per-child=$MAX_TASKS  mem-guard=${MEM_GUARD}G"
echo "   executor=$EXECUTOR  nice=$NICE(让OCR服务优先)  日志→$LOG"
[ "$MODE" = "local" ] && echo "   source-list=$SOURCE_LIST"
echo "=================================================="

# ---- 1. 防重复启动 ----
if pgrep -af "$PROC_PAT" 2>/dev/null | grep -v 'multiprocessing' | grep -v 'grep' >/dev/null; then
  echo "✗ 已经有 $ENTRY 在运行了, 不重复启动。"
  echo "  查看: pgrep -af '$PROC_PAT' | grep -v multiprocessing"
  echo "  如要重启: 先 bash scripts/stop.sh -y"
  exit 1
fi

# ---- 2. 启动前自检 ----
fail=0
need() { command -v "$1" >/dev/null 2>&1 && echo "  ✓ $1" || { echo "  ✗ 缺少 $1 (需在 PATH 或项目目录)"; fail=1; }; }
echo "[依赖检查]"
need python
need ffmpeg
need ffprobe
if command -v coscli >/dev/null 2>&1; then echo "  ✓ coscli"
elif [ -x "./coscli-linux-amd64" ]; then echo "  ✓ ./coscli-linux-amd64"
else echo "  ✗ 找不到 coscli / ./coscli-linux-amd64"; fail=1; fi

echo "[配置/任务检查]"
[ -f config/config.json ] && echo "  ✓ config/config.json" || { echo "  ✗ 缺 config/config.json"; fail=1; }

if [ "$MODE" = "worker" ]; then
  # worker 模式: 任务来自 COS, 不需要本地 text2; 检查 dispatch 配置与机器ID + COS 队列可达
  chk=$(python - <<'PY' 2>&1
import sys
try:
    from config.config_utils import config
    from tool import job_cos
    dp = config.get_dispatch_config() or {}
    if not dp.get("jobs_root"):
        print("NO_JOBS_ROOT"); sys.exit(0)
    mid = job_cos.get_machine_id()
    todo = job_cos.list_todo()       # 顺带验证 COS 队列可达
    print("OK\t%s\t%d" % (mid, len(todo)))
except Exception as e:
    print("ERR\t%s" % e)
PY
)
  case "$chk" in
    OK*)   mid=$(echo "$chk" | cut -f2); n=$(echo "$chk" | cut -f3)
           echo "  ✓ dispatch 配置OK  机器ID=$mid  todo剩余分片=$n"
           [ "$n" = "0" ] && echo "  ⚠ todo 队列为空(worker会空闲等待; 先 python tools/seed_jobs.py 灌任务)";;
    NO_JOBS_ROOT) echo "  ✗ config.dispatch.jobs_root 未配置(取任务地址)"; fail=1;;
    *)     echo "  ✗ 访问 COS 任务队列失败: ${chk#ERR	}"; fail=1;;
  esac
else
  if ls "$SOURCE_LIST"/*.txt >/dev/null 2>&1; then
    n=$(ls "$SOURCE_LIST"/*.txt 2>/dev/null | wc -l)
    echo "  ✓ $SOURCE_LIST 有 $n 个任务txt"
  else
    echo "  ✗ $SOURCE_LIST 下没有 .txt 任务文件"; fail=1
  fi
fi

if [ "$fail" -ne 0 ]; then
  echo ""
  echo "✗ 自检未通过, 已中止(没启动)。修好上面 ✗ 的项再跑。"
  exit 1
fi

# ---- 3. 启动(setsid+nohup+disown 三保险, SSH断开也杀不到) ----
echo ""
echo "[启动中...]"
if [ "$MODE" = "worker" ]; then
  setsid nohup nice -n "$NICE" python run_worker.py \
    --max-workers "$MAX_WORKERS" \
    --max-tasks-per-child "$MAX_TASKS" \
    --mem-guard-gb "$MEM_GUARD" \
    --executor "$EXECUTOR" >> "$LOG" 2>&1 &
else
  setsid nohup nice -n "$NICE" python main.py \
    --source-list "$SOURCE_LIST" \
    --max-workers "$MAX_WORKERS" \
    --max-tasks-per-child "$MAX_TASKS" \
    --mem-guard-gb "$MEM_GUARD" \
    --executor "$EXECUTOR" >> "$LOG" 2>&1 &
fi
disown 2>/dev/null || true

# ---- 4. 等几秒确认起来了 ----
sleep 6
if pgrep -af "$PROC_PAT" 2>/dev/null | grep -v 'multiprocessing' | grep -v 'grep' >/dev/null; then
  echo "✓ 已启动。日志末尾:"
  tail -n 5 "$LOG" 2>/dev/null | sed 's/^/    /'
  echo ""
  echo "  实时看日志:  tail -f $LOG"
  echo "  数worker:    pgrep -af multiprocessing | wc -l   (稍等会接近 $MAX_WORKERS_DISP)"
  echo "  停止:        bash scripts/stop.sh -y"
else
  echo "✗ 启动后没检测到进程, 查日志看原因:"
  tail -n 20 "$LOG" 2>/dev/null | sed 's/^/    /'
  exit 1
fi