248 lines
10 KiB
Python
248 lines
10 KiB
Python
import asyncio
|
|
import os
|
|
import random
|
|
import time
|
|
from asyncio import Task
|
|
from typing import Dict, List, Optional, Tuple
|
|
|
|
from playwright.async_api import (BrowserContext, BrowserType, Page,
|
|
async_playwright)
|
|
|
|
import config
|
|
from base.base_crawler import AbstractCrawler
|
|
from models import kuaishou
|
|
from proxy.proxy_account_pool import AccountPool
|
|
from tools import utils
|
|
from var import comment_tasks_var, crawler_type_var
|
|
|
|
from .client import KuaiShouClient
|
|
from .exception import DataFetchError
|
|
from .login import KuaishouLogin
|
|
|
|
|
|
class KuaishouCrawler(AbstractCrawler):
|
|
platform: str
|
|
login_type: str
|
|
crawler_type: str
|
|
context_page: Page
|
|
ks_client: KuaiShouClient
|
|
account_pool: AccountPool
|
|
browser_context: BrowserContext
|
|
|
|
def __init__(self):
|
|
self.index_url = "https://www.kuaishou.com"
|
|
self.user_agent = utils.get_user_agent()
|
|
|
|
def init_config(self, platform: str, login_type: str, account_pool: AccountPool, crawler_type: str):
|
|
self.platform = platform
|
|
self.login_type = login_type
|
|
self.account_pool = account_pool
|
|
self.crawler_type = crawler_type
|
|
|
|
async def start(self):
|
|
account_phone, playwright_proxy, httpx_proxy = self.create_proxy_info()
|
|
async with async_playwright() as playwright:
|
|
# Launch a browser context.
|
|
chromium = playwright.chromium
|
|
self.browser_context = await self.launch_browser(
|
|
chromium,
|
|
playwright_proxy,
|
|
self.user_agent,
|
|
headless=config.HEADLESS
|
|
)
|
|
# stealth.min.js is a js script to prevent the website from detecting the crawler.
|
|
await self.browser_context.add_init_script(path="libs/stealth.min.js")
|
|
self.context_page = await self.browser_context.new_page()
|
|
await self.context_page.goto(f"{self.index_url}?isHome=1")
|
|
|
|
# Create a client to interact with the kuaishou website.
|
|
self.ks_client = await self.create_ks_client(httpx_proxy)
|
|
if not await self.ks_client.pong():
|
|
login_obj = KuaishouLogin(
|
|
login_type=self.login_type,
|
|
login_phone=account_phone,
|
|
browser_context=self.browser_context,
|
|
context_page=self.context_page,
|
|
cookie_str=config.COOKIES
|
|
)
|
|
await login_obj.begin()
|
|
await self.ks_client.update_cookies(browser_context=self.browser_context)
|
|
|
|
crawler_type_var.set(self.crawler_type)
|
|
if self.crawler_type == "search":
|
|
# Search for notes and retrieve their comment information.
|
|
await self.search()
|
|
elif self.crawler_type == "detail":
|
|
# Get the information and comments of the specified post
|
|
await self.get_specified_videos()
|
|
else:
|
|
pass
|
|
|
|
utils.logger.info("Kuaishou Crawler finished ...")
|
|
|
|
async def search(self):
|
|
utils.logger.info("Begin search kuaishou keywords")
|
|
ks_limit_count = 20 # kuaishou limit page fixed value
|
|
for keyword in config.KEYWORDS.split(","):
|
|
utils.logger.info(f"Current search keyword: {keyword}")
|
|
page = 1
|
|
while page * ks_limit_count <= config.CRAWLER_MAX_NOTES_COUNT:
|
|
video_id_list: List[str] = []
|
|
videos_res = await self.ks_client.search_info_by_keyword(
|
|
keyword=keyword,
|
|
pcursor=str(page),
|
|
)
|
|
if not videos_res:
|
|
utils.logger.error(f"search info by keyword:{keyword} not found data")
|
|
continue
|
|
|
|
vision_search_photo: Dict = videos_res.get("visionSearchPhoto")
|
|
if vision_search_photo.get("result") != 1:
|
|
utils.logger.error(f"search info by keyword:{keyword} not found data ")
|
|
continue
|
|
|
|
for video_detail in vision_search_photo.get("feeds"):
|
|
video_id_list.append(video_detail.get("photo", {}).get("id"))
|
|
await kuaishou.update_kuaishou_video(video_item=video_detail)
|
|
|
|
# batch fetch video comments
|
|
page += 1
|
|
await self.batch_get_video_comments(video_id_list)
|
|
|
|
async def get_specified_videos(self):
|
|
"""Get the information and comments of the specified post"""
|
|
semaphore = asyncio.Semaphore(config.MAX_CONCURRENCY_NUM)
|
|
task_list = [
|
|
self.get_video_info_task(video_id=video_id, semaphore=semaphore) for video_id in config.KS_SPECIFIED_ID_LIST
|
|
]
|
|
video_details = await asyncio.gather(*task_list)
|
|
for video_detail in video_details:
|
|
if video_detail is not None:
|
|
await kuaishou.update_kuaishou_video(video_detail)
|
|
await self.batch_get_video_comments(config.KS_SPECIFIED_ID_LIST)
|
|
|
|
async def get_video_info_task(self, video_id: str, semaphore: asyncio.Semaphore) -> Optional[Dict]:
|
|
"""Get video detail task"""
|
|
async with semaphore:
|
|
try:
|
|
result = await self.ks_client.get_video_info(video_id)
|
|
utils.logger.info(f"Get video_id:{video_id} info result: {result} ...")
|
|
return result.get("visionVideoDetail")
|
|
except DataFetchError as ex:
|
|
utils.logger.error(f"Get video detail error: {ex}")
|
|
return None
|
|
except KeyError as ex:
|
|
utils.logger.error(f"have not fund note detail video_id:{video_id}, err: {ex}")
|
|
return None
|
|
|
|
async def batch_get_video_comments(self, video_id_list: List[str]):
|
|
"""
|
|
batch get video comments
|
|
:param video_id_list:
|
|
:return:
|
|
"""
|
|
utils.logger.info(f"[batch_get_video_comments] video ids:{video_id_list}")
|
|
semaphore = asyncio.Semaphore(config.MAX_CONCURRENCY_NUM)
|
|
task_list: List[Task] = []
|
|
for video_id in video_id_list:
|
|
task = asyncio.create_task(self.get_comments(video_id, semaphore), name=video_id)
|
|
task_list.append(task)
|
|
|
|
comment_tasks_var.set(task_list)
|
|
await asyncio.gather(*task_list)
|
|
|
|
async def get_comments(self, video_id: str, semaphore: asyncio.Semaphore):
|
|
"""
|
|
get comment for video id
|
|
:param video_id:
|
|
:param semaphore:
|
|
:return:
|
|
"""
|
|
async with semaphore:
|
|
try:
|
|
utils.logger.info(f"[get_comments] bengin get video_id: {video_id} comments ...")
|
|
await self.ks_client.get_video_all_comments(
|
|
photo_id=video_id,
|
|
crawl_interval=random.random(),
|
|
callback=kuaishou.batch_update_ks_video_comments
|
|
)
|
|
except DataFetchError as ex:
|
|
utils.logger.error(f"[get_comments] get video_id: {video_id} comment error: {ex}")
|
|
except Exception as e:
|
|
utils.logger.error(f"[get_comments] may be been blocked, err:", e)
|
|
# use time.sleeep block main coroutine instead of asyncio.sleep and cacel running comment task
|
|
# maybe kuaishou block our request, we will take a nap and update the cookie again
|
|
current_running_tasks = comment_tasks_var.get()
|
|
for task in current_running_tasks:
|
|
task.cancel()
|
|
time.sleep(20)
|
|
await self.context_page.goto(f"{self.index_url}?isHome=1")
|
|
await self.ks_client.update_cookies(browser_context=self.browser_context)
|
|
|
|
def create_proxy_info(self) -> Tuple[Optional[str], Optional[Dict], Optional[str]]:
|
|
"""Create proxy info for playwright and httpx"""
|
|
# phone: 13012345671 ip_proxy: 111.122.xx.xx1:8888
|
|
phone, ip_proxy = self.account_pool.get_account()
|
|
if not config.ENABLE_IP_PROXY:
|
|
return phone, None, None
|
|
utils.logger.info("Begin proxy info for playwright and httpx ...")
|
|
playwright_proxy = {
|
|
"server": f"{config.IP_PROXY_PROTOCOL}{ip_proxy}",
|
|
"username": config.IP_PROXY_USER,
|
|
"password": config.IP_PROXY_PASSWORD,
|
|
}
|
|
httpx_proxy = f"{config.IP_PROXY_PROTOCOL}{config.IP_PROXY_USER}:{config.IP_PROXY_PASSWORD}@{ip_proxy}"
|
|
return phone, playwright_proxy, httpx_proxy
|
|
|
|
async def create_ks_client(self, httpx_proxy: Optional[str]) -> KuaiShouClient:
|
|
"""Create xhs client"""
|
|
utils.logger.info("Begin create kuaishou API client ...")
|
|
cookie_str, cookie_dict = utils.convert_cookies(await self.browser_context.cookies())
|
|
xhs_client_obj = KuaiShouClient(
|
|
proxies=httpx_proxy,
|
|
headers={
|
|
"User-Agent": self.user_agent,
|
|
"Cookie": cookie_str,
|
|
"Origin": self.index_url,
|
|
"Referer": self.index_url,
|
|
"Content-Type": "application/json;charset=UTF-8"
|
|
},
|
|
playwright_page=self.context_page,
|
|
cookie_dict=cookie_dict,
|
|
)
|
|
return xhs_client_obj
|
|
|
|
async def launch_browser(
|
|
self,
|
|
chromium: BrowserType,
|
|
playwright_proxy: Optional[Dict],
|
|
user_agent: Optional[str],
|
|
headless: bool = True
|
|
) -> BrowserContext:
|
|
"""Launch browser and create browser context"""
|
|
utils.logger.info("Begin create browser context ...")
|
|
if config.SAVE_LOGIN_STATE:
|
|
user_data_dir = os.path.join(os.getcwd(), "browser_data",
|
|
config.USER_DATA_DIR % self.platform) # type: ignore
|
|
browser_context = await chromium.launch_persistent_context(
|
|
user_data_dir=user_data_dir,
|
|
accept_downloads=True,
|
|
headless=headless,
|
|
proxy=playwright_proxy, # type: ignore
|
|
viewport={"width": 1920, "height": 1080},
|
|
user_agent=user_agent
|
|
)
|
|
return browser_context
|
|
else:
|
|
browser = await chromium.launch(headless=headless, proxy=playwright_proxy) # type: ignore
|
|
browser_context = await browser.new_context(
|
|
viewport={"width": 1920, "height": 1080},
|
|
user_agent=user_agent
|
|
)
|
|
return browser_context
|
|
|
|
async def close(self):
|
|
"""Close browser context"""
|
|
await self.browser_context.close()
|
|
utils.logger.info("Browser context closed ...")
|