o2o-castad-backend/app/utils/nvMapPwScraper.py

438 lines
17 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters!

This file contains ambiguous Unicode characters that may be confused with others in your current locale. If your use case is intentional and legitimate, you can safely ignore this warning. Use the Escape button to highlight these characters.

import asyncio
import re
from html import unescape
from difflib import SequenceMatcher
from playwright.async_api import async_playwright
from urllib import parse
import time
from app.utils.logger import get_logger
# 로거 설정
logger = get_logger("pwscraper")
class NvMapPwScraper():
# cls vars
is_ready = False
_playwright = None
_browser = None
_context = None
_win_width = 1280
_win_height = 720
_max_retry = 3
_timeout = 60 # place id timeout threshold seconds
# instance var
page = None
@classmethod
def default_context_builder(cls):
context_builder_dict = {}
context_builder_dict['viewport'] = {
'width' : cls._win_width,
'height' : cls._win_height
}
context_builder_dict['screen'] = {
'width' : cls._win_width,
'height' : cls._win_height
}
context_builder_dict['user_agent'] = "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/130.0.0.0 Safari/537.36"
context_builder_dict['locale'] = 'ko-KR'
context_builder_dict['timezone_id']='Asia/Seoul'
return context_builder_dict
@classmethod
async def initiate_scraper(cls):
if not cls._playwright:
cls._playwright = await async_playwright().start()
if not cls._browser:
cls._browser = await cls._playwright.chromium.launch(headless=True)
if not cls._context:
cls._context = await cls._browser.new_context(**cls.default_context_builder())
cls.is_ready = True
GRAPHQL_URL = "https://pcmap-api.place.naver.com/graphql"
@classmethod
async def fetch_graphql(
cls,
place_id: str,
payloads: list[dict],
) -> list[dict | None] | None:
"""실제 브라우저로 네이버 WTM 안티봇 캡차를 통과해 GraphQL 쿼리를 실행한다.
네이버 pcmap GraphQL은 두 헤더를 검사한다:
- x-wtm-graphql : base64({"arg": place_id, "type": ..., "source": "place"})
- x-wtm-ncaptcha-token : 캡차 JS가 생성 (요청마다 발급, 직접 생성 불가)
둘 다 없으면 405(캡차)로 막힌다. 따라서 place 페이지를 실제로 로드해
페이지가 자연 발생시키는 GraphQL 요청에서 두 헤더를 캡처한 뒤,
같은 토큰으로 우리 쿼리들을 in-page fetch로 실행한다.
Args:
place_id: 네이버 place ID
payloads: GraphQL POST 본문 목록
Returns:
payload별 파싱 JSON 목록(실패 항목은 None). 토큰 캡처 실패 시 None.
"""
if not cls.is_ready:
logger.warning("[NvMapPwScraper] fetch_graphql: scraper가 초기화되지 않았습니다")
return None
page = await cls._context.new_page()
captured: dict = {}
def on_request(req):
if "/graphql" in req.url and req.method == "POST" and "tok" not in captured:
tok = req.headers.get("x-wtm-ncaptcha-token")
if tok:
captured["tok"] = tok
captured["wtm"] = req.headers.get("x-wtm-graphql")
page.on("request", on_request)
try:
for t in ("place", "restaurant", "accommodation"):
try:
await page.goto(
f"https://pcmap.place.naver.com/{t}/{place_id}/home",
wait_until="domcontentloaded",
timeout=30000,
)
except Exception as e:
logger.debug(f"[NvMapPwScraper] goto {t} 실패: {e}")
continue
# 페이지가 GraphQL 요청을 보내며 토큰이 헤더에 실릴 때까지 대기 (최대 ~9초)
for _ in range(30):
if captured.get("tok"):
break
await page.wait_for_timeout(300)
if captured.get("tok"):
logger.info(f"[NvMapPwScraper] WTM 토큰 캡처 성공 (type={t})")
break
if not captured.get("tok"):
logger.warning("[NvMapPwScraper] WTM 토큰 캡처 실패")
return None
headers = {
"Content-Type": "application/json",
"x-wtm-graphql": captured["wtm"],
"x-wtm-ncaptcha-token": captured["tok"],
}
results: list[dict | None] = []
for idx, payload in enumerate(payloads, start=1):
r = await page.evaluate(
"""async (a) => {
const [url, body, hdr] = a;
try {
const r = await fetch(url, {
method: 'POST',
headers: hdr,
body: JSON.stringify(body),
credentials: 'include',
});
if (r.status !== 200) return {__err: r.status};
return await r.json();
} catch (e) { return {__err: String(e)}; }
}""",
[cls.GRAPHQL_URL, payload, headers],
)
if isinstance(r, dict) and "__err" in r:
op = payload.get("operationName", "?")
logger.warning(
f"[NvMapPwScraper] graphql fetch 실패 "
f"({idx}/{len(payloads)} {op}): {r['__err']}"
)
results.append(None)
else:
results.append(r)
return results
except Exception as e:
logger.error(f"[NvMapPwScraper] fetch_graphql 오류: {e}")
return None
finally:
await page.close()
def __init__(self):
if not self.is_ready:
raise Exception("nvMapScraper is not initiated")
async def __aenter__(self):
await self.create_page()
return self
async def __aexit__(self, exc_type, exc, tb):
await self.page.close()
async def create_page(self):
self.page = await self._context.new_page()
await self.page.add_init_script(
'''const defaultGetter = Object.getOwnPropertyDescriptor(
Navigator.prototype,
"webdriver"
).get;
defaultGetter.apply(navigator);
defaultGetter.toString();
Object.defineProperty(Navigator.prototype, "webdriver", {
set: undefined,
enumerable: true,
configurable: true,
get: new Proxy(defaultGetter, {
apply: (target, thisArg, args) => {
Reflect.apply(target, thisArg, args);
return false;
},
}),
});
const patchedGetter = Object.getOwnPropertyDescriptor(
Navigator.prototype,
"webdriver"
).get;
patchedGetter.apply(navigator);
patchedGetter.toString();''')
await self.page.set_extra_http_headers({
'sec-ch-ua': '\"Not?A_Brand\";v=\"99\", \"Chromium\";v=\"130\"'
})
await self.page.goto("http://google.com")
async def goto_url(self, url, wait_until="domcontentloaded", timeout=20000):
page = self.page
await page.goto(url, wait_until=wait_until, timeout=timeout)
@staticmethod
def _clean_title(text: str) -> str:
text = unescape(text) # HTML 엔티티 디코딩 (& → &)
text = re.sub(r"<.*?>", "", text) # HTML 태그 제거
return text.strip()
@staticmethod
def _similarity(a: str, b: str) -> float:
return SequenceMatcher(None, a, b).ratio()
@staticmethod
def _refine_address(address: str) -> str:
"""한국 주소 패턴에서 첫 번째 유효한 주소만 추출한다."""
patterns = [
# 도로명 (정식): 경기도 가평군 운악로 278
re.compile(
r'[가-힣]+(?:특별시|광역시|특별자치시|도|특별자치도|시)\s+'
r'[가-힣\s]+?(?:로|길|대로)\s+\d+(?:-\d+)?'
),
# 지번 (정식): 경기도 가평군 조종면 운악리 278
re.compile(
r'[가-힣]+(?:특별시|광역시|특별자치시|도|특별자치도|시)\s+'
r'[가-힣\s]+?(?:읍|면|동|리|가)\s+\d+(?:-\d+)?'
),
# 도로명 (축약): 경기 가평 운악로 278
re.compile(
r'[가-힣]{1,4}\s+[가-힣]{1,6}\s+'
r'[가-힣\s]+?(?:로|길|대로)\s+\d+(?:-\d+)?'
),
# 지번 (축약): 경기 가평 조종면 운악리 278
re.compile(
r'[가-힣]{1,4}\s+[가-힣]{1,6}\s+'
r'[가-힣\s]+?(?:읍|면|동|리|가)\s+\d+(?:-\d+)?'
),
]
for pattern in patterns:
m = pattern.search(address)
if m:
return m.group().strip()
return address
async def _extract_candidates_from_list_page(self) -> list[dict]:
"""pcmap.place.naver.com iframe HTML에서 place ID와 업체명을 추출한다."""
pcmap_frame = None
for frame in self.page.frames:
if "pcmap.place.naver.com" in frame.url:
pcmap_frame = frame
logger.debug(f"[DEBUG] pcmap frame 발견: {frame.url[:80]}")
break
if not pcmap_frame:
logger.debug("[DEBUG] pcmap frame 없음")
return []
try:
html = await pcmap_frame.content()
except Exception as e:
logger.debug(f"[DEBUG] pcmap frame content 추출 실패: {e}")
return []
# {"id":"11659052","name":"프레지던트 호텔",...} 형태의 JSON 쌍 추출
pair_pattern = re.compile(
r'"id"\s*:\s*"(\d{5,})"[^}]{0,200}?"name"\s*:\s*"([^"]{1,60})"'
r'|"name"\s*:\s*"([^"]{1,60})"[^}]{0,200}?"id"\s*:\s*"(\d{5,})"'
)
seen = {} # place_id → title (순서 보존)
for m in pair_pattern.finditer(html):
if m.group(1): # id 먼저
pid, title = m.group(1), m.group(2)
else: # name 먼저
pid, title = m.group(4), m.group(3)
if pid not in seen:
seen[pid] = title
candidates = [
{"title": title, "place_url": f"https://map.naver.com/p/entry/place/{pid}"}
for pid, title in list(seen.items())[:10]
]
for i, c in enumerate(candidates):
logger.debug(f"[DEBUG] 후보 {i+1}: {c['title']} / {c['place_url']}")
logger.debug(f"[DEBUG] 목록 후보 {len(candidates)}개 추출")
return candidates
@staticmethod
def _parse_allsearch_candidates(body: dict) -> list[dict]:
"""allSearch 응답 JSON에서 후보 목록을 추출한다."""
place_list = (((body.get("result") or {}).get("place") or {}).get("list")) or []
return [
{
"title": p.get("name") or "",
"place_url": f"https://map.naver.com/p/entry/place/{p['id']}",
"roadAddress": p.get("roadAddress") or "",
"address": p.get("address") or "",
}
for p in place_list
if p.get("id")
]
def _select_best_candidate(self, candidates: list[dict], title: str, address: str) -> dict | None:
"""이름 유사도(70%)와 주소 유사도(30%)를 함께 고려해 최적 후보를 선택한다.
주소 없이 업체명만으로 검색한 경우(3차 폴백)는 이름 유사도만 사용한다.
"""
if not candidates:
return None
for c in candidates:
name_score = self._similarity(title, self._clean_title(c["title"]))
cand_addr = c.get("roadAddress") or c.get("address") or ""
addr_score = self._similarity(address, cand_addr) if address and cand_addr else 0.0
c["_name_score"] = name_score
c["_addr_score"] = addr_score
c["_total_score"] = name_score * 0.7 + addr_score * 0.3 if address else name_score
return max(candidates, key=lambda c: c["_total_score"])
async def _capture_allsearch(self, url: str, timeout_s: float = 8.0) -> list[dict]:
"""검색 페이지가 자연 발생시키는 allSearch API 응답을 캡처해 후보 목록으로 변환한다.
pcmap iframe 렌더링을 기다릴 필요가 없어 경쟁 조건이 없고, 업체명 단독
검색처럼 iframe 자체가 생성되지 않는 케이스에서도 동작한다.
"""
captured: list[dict] = []
def on_response(resp):
# GET 요청의 실제 응답만 캡처 (OPTIONS preflight 등 오탐 방지)
if "allSearch" in resp.url and resp.request.method == "GET":
async def _consume():
try:
captured.append(await resp.json())
except Exception:
pass
asyncio.create_task(_consume())
self.page.on("response", on_response)
try:
try:
await self.goto_url(url, wait_until="domcontentloaded", timeout=self._timeout * 1000)
except Exception:
logger.error("[ERROR] Can't Finish domcontentloaded")
deadline = time.monotonic() + timeout_s
while time.monotonic() < deadline:
if "/place/" in self.page.url or captured:
break
await self.page.wait_for_timeout(200)
finally:
self.page.remove_listener("response", on_response)
if not captured:
return []
return self._parse_allsearch_candidates(captured[0])
async def _try_search(self, address: str, title: str) -> str | None:
"""주어진 주소+업체명으로 검색해서 place URL을 반환한다. 실패 시 None."""
encoded_query = parse.quote(f"{address} {title}".strip())
url = f"https://map.naver.com/p/search/{encoded_query}"
# 1순위: allSearch API 응답 캡처 (iframe 렌더링 대기 불필요, 업체명 단독 검색도 지원)
candidates = await self._capture_allsearch(url)
if "/place/" in self.page.url:
return self.page.url
if candidates:
best = self._select_best_candidate(candidates, title, address)
logger.info(
f"[AUTO-SELECT] '{title}''{best['title']}' "
f"(name={best['_name_score']:.2f}, addr={best['_addr_score']:.2f}) {best['place_url']}"
)
return best['place_url']
# 2순위(안전망): allSearch 캡처 실패 시 기존 iframe HTML 파싱 방식으로 폴백
logger.warning("[FALLBACK] allSearch 캡처 실패 → iframe 파싱 방식 시도")
iframe_candidates = []
for _ in range(3): # 500ms × 3 = 최대 1.5초
if "/place/" in self.page.url:
return self.page.url
iframe_candidates = await self._extract_candidates_from_list_page()
if iframe_candidates:
break
await self.page.wait_for_timeout(500)
if iframe_candidates:
best = self._select_best_candidate(iframe_candidates, title, address)
logger.info(
f"[AUTO-SELECT-IFRAME] '{title}''{best['title']}' "
f"(name={best['_name_score']:.2f}, addr={best['_addr_score']:.2f}) {best['place_url']}"
)
return best['place_url']
# isCorrectAnswer=true 로 강제 단일결과 재시도 (원본 로직 유지)
correct_url = self.page.url.replace("?", "?isCorrectAnswer=true&")
try:
await self.goto_url(correct_url, wait_until="networkidle", timeout=self._timeout * 1000)
except:
if "/place/" in self.page.url:
return self.page.url
logger.error("[ERROR] Can't Finish networkidle (isCorrectAnswer)")
if "/place/" in self.page.url:
return self.page.url
return None
async def get_place_id_url(self, selected):
title = self._clean_title(selected['title'])
address = self._clean_title(selected.get('roadAddress', selected['address']))
# 1차 시도: 원본 주소 + 업체명
logger.debug(f"[DEBUG] 1차 시도 - address: {address}")
result = await self._try_search(address, title)
if result:
return result
# 2차 시도: 정제 주소 + 업체명
refined = self._refine_address(address)
if refined != address:
logger.info(f"[REFINE] 주소 정제: '{address}''{refined}'")
result = await self._try_search(refined, title)
if result:
return result
# 3차 시도: 업체명만으로 검색
logger.info(f"[RETRY] 업체명만으로 재시도: '{title}'")
result = await self._try_search("", title)
if result:
return result
logger.error(f"[ERROR] Not found url for {selected}")
return None