#!/usr/bin/env bash
# deploy/qa-db/replace_db.sh —— PlantAI QA 裸进程数据库整库替换（用户高频操作）
# 用法（需 root；数据目录换新后属主会重对齐，按提示执行即可）：
#   sudo -S bash replace_db.sh qdrant <新storage目录>   # 替换 Qdrant 数据
#   sudo -S bash replace_db.sh neo4j   <新data目录>     # 替换 Neo4j 数据
# 行为：停 unit → 旧目录 mv 为 .bak.<时间戳> → 新目录就位+属主对齐 → 启 unit → 探活
#   → 通过后提示删除 .bak 与回滚命令。失败自动回滚。
# 配置默认与 install_qdrant.sh / install_neo4j.sh 一致：
#   QA_DB_ROOT 默认 <安装根>/qa-db；Qdrant HTTP 27433；Neo4j bolt 19888
set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
INSTALL_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
QA_DB_ROOT="${QA_DB_ROOT:-$INSTALL_ROOT/qa-db}"
HTTP_PORT="${QDRANT_HTTP_PORT:-27433}"
BOLT_PORT="${NEO4J_BOLT_PORT:-19888}"
KIND="${1:-}"
NEW_DIR="${2:-}"

[[ $EUID -eq 0 ]] || { echo "需要 root：sudo -S bash $0 $*" >&2; exit 1; }
case "$KIND" in
  qdrant) UNIT=qdrant-plantai;    CUR="$QA_DB_ROOT/qdrant/storage"; OWNER="${SUDO_USER:-xzq}";;
  neo4j)  UNIT=neo4j-plantai;     CUR="$QA_DB_ROOT/neo4j/data";    OWNER="${SUDO_USER:-xzq}";;
  *) echo "用法: $0 qdrant|neo4j <新目录>"; exit 2;;
esac
[[ -n "$NEW_DIR" && -d "$NEW_DIR" ]] || { echo "新目录不存在：$NEW_DIR" >&2; exit 2; }
systemctl is-active --quiet "$UNIT" || { echo "unit $UNIT 未运行，请先启动或检查"; exit 2; }

TS="$(date +%Y%m%dT%H%M%S)"
BAK="${CUR}.bak.$TS"

echo "== 整库替换 $KIND =="
echo "  旧数据：$CUR"
echo "  新数据：$NEW_DIR"
echo "  备份为：$BAK"
systemctl stop "$UNIT"
sleep 3

rollback() {
  echo "[回滚] 恢复旧数据并重启"
  rm -rf "$CUR"
  mv "$BAK" "$CUR"
  chown -R "$OWNER:$OWNER" "$(dirname "$CUR")" 2>/dev/null || true
  systemctl start "$UNIT"
  exit 1
}

if [[ -e "$CUR" ]]; then mv "$CUR" "$BAK"; fi
mkdir -p "$(dirname "$CUR")"
mv "$NEW_DIR" "$CUR"
chown -R "$OWNER:$OWNER" "$CUR"

systemctl start "$UNIT" || { echo "[失败] 启动失败，自动回滚"; rollback; }

probe_ok=0
for _ in $(seq 1 90); do
  if [[ "$KIND" == qdrant ]] && curl -sf --max-time 2 "http://127.0.0.1:$HTTP_PORT/collections" >/dev/null 2>&1; then
    probe_ok=1; break
  elif [[ "$KIND" == neo4j ]] && (exec 3<>"/dev/tcp/127.0.0.1/$BOLT_PORT") 2>/dev/null; then
    exec 3>&- 2>/dev/null || true; probe_ok=1; break
  fi
  sleep 2
done

if [[ $probe_ok -eq 0 ]]; then
  echo "[失败] 新库探活未通过，自动回滚"
  rollback
fi

echo "== 替换成功 =="
echo "  新库在线：$CUR"
echo "  旧库备份：$BAK"
echo "  确认稳定后删除备份：rm -rf $BAK"
echo "  如需回滚：rm -rf $CUR && mv $BAK $CUR && systemctl restart $UNIT"