354 lines
16 KiB
Python
354 lines
16 KiB
Python
import asyncio
|
||
import logging
|
||
import uuid
|
||
import os
|
||
import sys
|
||
import json
|
||
import time
|
||
from datetime import datetime
|
||
from Apps.YeLiTe.Kit import take_screenshot, clean_station_name, get_image_content_md5, find_template
|
||
from Apps.YeLiTe.ReadImageKit import ReadImageKit
|
||
from Apps.YeLiTe.Service import YiLaiTeService
|
||
from Apps.YeLiTe.Config.Setting import (
|
||
SCROLL_DISTANCE_RATIO, WAIT_AFTER_SCROLL, MAX_STATIONS_COUNT,
|
||
WAIT_DETAIL_PAGE_LOAD, WAIT_BACK_TO_LIST, TEST_CLEAR_REDIS,
|
||
SAFE_EXCLUDE_RATIO, BOTTOM_SAFE_EXCLUDE_RATIO
|
||
)
|
||
from Util.RedisKit import RedisKit
|
||
from Core.BaseCrawler import BaseCrawler
|
||
import uiautomator2 as u2
|
||
|
||
# 项目根目录处理
|
||
project_root = os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
||
if project_root not in sys.path:
|
||
sys.path.append(project_root)
|
||
|
||
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
|
||
logger = logging.getLogger("YiLaiTeCrawler")
|
||
|
||
class YiLaiTeCrawler(BaseCrawler):
|
||
def __init__(self, service=None):
|
||
super().__init__(service or YiLaiTeService())
|
||
self.read_image_kit = ReadImageKit()
|
||
self.redis_kit = RedisKit()
|
||
|
||
async def start(self):
|
||
"""
|
||
实现 BaseCrawler 的启动入口
|
||
"""
|
||
await main(self.service)
|
||
|
||
async def open_app(self):
|
||
"""
|
||
打开驿来特小程序
|
||
"""
|
||
from Apps.YeLiTe import Opener
|
||
return await Opener.open_mini_program()
|
||
|
||
async def crawl_list(self):
|
||
"""
|
||
开始爬取列表页
|
||
"""
|
||
d = u2.connect()
|
||
return await self.crawl_list_logic(d)
|
||
|
||
async def crawl_detail(self, station_info):
|
||
"""
|
||
爬取详情页 (BaseCrawler 要求，此处逻辑已集成在 crawl_list_logic 中)
|
||
"""
|
||
pass
|
||
|
||
async def clean_redis_data(self):
|
||
"""
|
||
清除测试用的 Redis 记录
|
||
"""
|
||
if TEST_CLEAR_REDIS:
|
||
logger.info("清理 Redis 中的场站处理记录...")
|
||
pattern = "crawled:ylt:*"
|
||
keys = await self.redis_kit.keys(pattern)
|
||
if keys:
|
||
await self.redis_kit.delete(*keys)
|
||
|
||
async def crawl_list_logic(self, d):
|
||
processed_count = 0
|
||
no_new_data_count = 0
|
||
last_md5 = None
|
||
background_tasks = []
|
||
|
||
while processed_count < MAX_STATIONS_COUNT:
|
||
# 1. 截图并分析
|
||
screenshot_path = take_screenshot(d, f"list_{int(time.time())}.jpg")
|
||
|
||
# 检测是否滚动到底部
|
||
curr_md5 = get_image_content_md5(screenshot_path, top_ratio=SAFE_EXCLUDE_RATIO, bottom_ratio=BOTTOM_SAFE_EXCLUDE_RATIO)
|
||
if last_md5 == curr_md5:
|
||
logger.info("内容无变化，判定已到底部")
|
||
if os.path.exists(screenshot_path): os.remove(screenshot_path)
|
||
break
|
||
last_md5 = curr_md5
|
||
|
||
stations = await self.read_image_kit.analyze_station_list(screenshot_path)
|
||
if not stations:
|
||
no_new_data_count += 1
|
||
if no_new_data_count >= 5:
|
||
logger.info("连续 5 页无新数据，停止")
|
||
break
|
||
# 滑动
|
||
d.swipe_ext("up", scale=SCROLL_DISTANCE_RATIO)
|
||
await asyncio.sleep(WAIT_AFTER_SCROLL)
|
||
continue
|
||
|
||
no_new_data_count = 0
|
||
new_stations_in_page = 0
|
||
for station in stations:
|
||
if processed_count >= MAX_STATIONS_COUNT:
|
||
break
|
||
|
||
name = station.get('name')
|
||
address = station.get('address')
|
||
point = station.get('point')
|
||
|
||
if not name or not point:
|
||
continue
|
||
|
||
# 去重
|
||
redis_key = f"crawled:ylt:{clean_station_name(name)}"
|
||
if await self.redis_kit.get_data(redis_key):
|
||
logger.info(f"场站 {name} 已处理过，跳过")
|
||
continue
|
||
|
||
# --- 新增逻辑：立即中断当前页面的后续遍历，先处理当前这个场站 ---
|
||
# 因为一旦点击进入二级页面再返回，列表页的状态可能会变（如滚动位置微调），
|
||
# 或者之前的坐标已经失效。为了稳健，每次处理完一个点击跳转，都应该重新截图分析列表页。
|
||
# 但重新截图分析成本太高，所以至少应该确保处理完这一个后，不会盲目点击后续的旧坐标。
|
||
#
|
||
# 然而，当前逻辑是遍历 `stations` 列表。如果第一个场站点击并返回了，
|
||
# 后面的 `stations` 中的坐标可能已经不准了（如果返回时页面刷新了）。
|
||
#
|
||
# 观察日志发现："发现第一场站后，没有进去查看二级，而是直接退回了小程序的搜索页面"
|
||
# 这说明可能是 `d.press("back")` 的逻辑有问题，或者页面结构导致 `back` 行为不一致。
|
||
#
|
||
# 针对用户反馈："发现第一场站后，没有进去查看二级，而是直接退回了小程序的搜索页面"
|
||
# 这通常是因为在列表页点击后，并没有真正进入二级页（也许点偏了，或者加载慢），
|
||
# 然后程序认为进去了，执行了 `back`，结果把列表页关了退回了搜索页。
|
||
#
|
||
# 现在的代码已经加了 MD5 校验 `if before_md5 != after_md5`，理论上能防住这个问题。
|
||
# 但用户最新的反馈是："没有马上进入第一个卡片对应的二级，而是向下继续滚动了去获取更多的场站了"
|
||
# 这意味着 `before_md5 != after_md5` 判断为 False，认为没进去，所以跳过了处理，继续循环。
|
||
#
|
||
# 为什么明明点击了却没进去？
|
||
# 1. 坐标点偏了？测试脚本显示点是红心，应该是准的。
|
||
# 2. 点击操作被某些透明层拦截了？
|
||
# 3. 页面还没加载完全？
|
||
#
|
||
# 让我们增加点击后的重试机制，或者微调点击位置。
|
||
# 另外，如果点击没进去，不应该直接 continue，应该尝试 debug 或者稍微移动一下位置再点。
|
||
#
|
||
# 这里我们加一个重试点击逻辑
|
||
|
||
# 点击进入前截图，用于对比是否成功进入二级页
|
||
before_click_path = take_screenshot(d, f"before_{clean_station_name(name)}")
|
||
before_md5 = get_image_content_md5(before_click_path, top_ratio=SAFE_EXCLUDE_RATIO, bottom_ratio=BOTTOM_SAFE_EXCLUDE_RATIO)
|
||
|
||
# 点击进入
|
||
logger.info(f">>> 发现新场站: {name} 坐标: {point}")
|
||
# 稍微等待一下，确保 UI 稳定
|
||
await asyncio.sleep(0.5)
|
||
# 使用 input tap 替代 d.click，提高点击成功率 (部分小程序对 click 响应不佳)
|
||
d.shell(f"input tap {int(point[0])} {int(point[1])}")
|
||
await asyncio.sleep(WAIT_DETAIL_PAGE_LOAD)
|
||
|
||
# 分析详情页 (采用异步后台模式)
|
||
detail_shot = take_screenshot(d, f"detail_{clean_station_name(name)}_{int(time.time())}")
|
||
after_md5 = get_image_content_md5(detail_shot, top_ratio=SAFE_EXCLUDE_RATIO, bottom_ratio=BOTTOM_SAFE_EXCLUDE_RATIO)
|
||
|
||
# 清理临时对比图
|
||
if os.path.exists(before_click_path): os.remove(before_click_path)
|
||
|
||
if before_md5 == after_md5:
|
||
logger.warning(f"首次点击 {name} 未跳转，尝试稍微偏移位置重试...")
|
||
# 尝试向下偏移 20px 点击
|
||
offset_y = point[1] + 20
|
||
d.shell(f"input tap {int(point[0])} {int(offset_y)}")
|
||
await asyncio.sleep(WAIT_DETAIL_PAGE_LOAD)
|
||
|
||
# 再次截图检查
|
||
if os.path.exists(detail_shot): os.remove(detail_shot)
|
||
detail_shot = take_screenshot(d, f"detail_{clean_station_name(name)}_{int(time.time())}")
|
||
after_md5 = get_image_content_md5(detail_shot, top_ratio=SAFE_EXCLUDE_RATIO, bottom_ratio=BOTTOM_SAFE_EXCLUDE_RATIO)
|
||
|
||
if before_md5 != after_md5:
|
||
logger.info(f"成功进入详情页: {name}")
|
||
|
||
# --- 新增：点击“阶段性电价”按钮以获取完整电价列表 ---
|
||
dqdf_pos = find_template(detail_shot, "dqdf.jpg")
|
||
detail_shots = []
|
||
|
||
if dqdf_pos:
|
||
logger.info(f"发现‘阶段性电价’按钮 {dqdf_pos}，点击进入...")
|
||
d.click(dqdf_pos[0], dqdf_pos[1])
|
||
await asyncio.sleep(2) # 等待列表加载
|
||
# 删除旧的详情页截图
|
||
if os.path.exists(detail_shot): os.remove(detail_shot)
|
||
|
||
# 截图1 (顶部)
|
||
shot1 = take_screenshot(d, f"detail_price_{clean_station_name(name)}_{int(time.time())}_1")
|
||
detail_shots.append(shot1)
|
||
|
||
# 向下滚动以获取完整价格表
|
||
logger.info("向下滚动以获取更多电价信息(从底部向上滑)...")
|
||
# User feedback: 必须放在下方，向上滚动
|
||
w, h = d.window_size()
|
||
# 调整滑动参数：避开底部导航栏，增加滑动距离，放慢速度
|
||
# start_y=0.75, end_y=0.25
|
||
d.swipe(w * 0.5, h * 0.75, w * 0.5, h * 0.25, 0.8)
|
||
await asyncio.sleep(2.0) # 增加等待时间，确保滚动停止
|
||
|
||
# 截图2 (底部)
|
||
shot2 = take_screenshot(d, f"detail_price_{clean_station_name(name)}_{int(time.time())}_2")
|
||
detail_shots.append(shot2)
|
||
|
||
# 关闭分时段定价列表 (点击屏幕最顶部空白处)
|
||
logger.info("点击屏幕上部空白处以关闭定价列表...")
|
||
d.click(w * 0.5, h * 0.1)
|
||
await asyncio.sleep(1.0)
|
||
else:
|
||
logger.info("未发现‘阶段性电价’按钮，直接分析当前页")
|
||
detail_shots.append(detail_shot)
|
||
# --------------------------------------------------
|
||
|
||
# 启动后台任务处理详情页
|
||
task = asyncio.create_task(self.analyze_detail_background(name, detail_shots, address=address))
|
||
background_tasks.append(task)
|
||
|
||
processed_count += 1
|
||
new_stations_in_page += 1
|
||
await self.redis_kit.set_data(redis_key, "1", expire=86400*7)
|
||
|
||
# 返回列表页 (现在应该已经回到了详情页主界面)
|
||
d.press("back")
|
||
await asyncio.sleep(WAIT_BACK_TO_LIST)
|
||
|
||
# 再次检查是否回到了列表页，如果没回，再点一次 back
|
||
after_back_shot = take_screenshot(d, "check_back")
|
||
after_back_md5 = get_image_content_md5(after_back_shot, top_ratio=SAFE_EXCLUDE_RATIO, bottom_ratio=BOTTOM_SAFE_EXCLUDE_RATIO)
|
||
if os.path.exists(after_back_shot): os.remove(after_back_shot)
|
||
|
||
if after_back_md5 == after_md5: # 还在详情页/电费页
|
||
logger.info("似乎还在二级页面，尝试再次返回...")
|
||
d.press("back")
|
||
await asyncio.sleep(WAIT_BACK_TO_LIST)
|
||
else:
|
||
logger.warning(f"点击场站 {name} 后页面似乎未跳转，跳过返回操作")
|
||
if os.path.exists(detail_shot): os.remove(detail_shot)
|
||
# 即使没进去，也记录一下，避免短时间内重复尝试
|
||
await self.redis_kit.set_data(redis_key, "1", expire=3600)
|
||
|
||
if new_stations_in_page == 0 and stations:
|
||
no_new_data_count += 1
|
||
if no_new_data_count >= 5:
|
||
logger.info("连续 5 页均无新场站，停止")
|
||
break
|
||
else:
|
||
no_new_data_count = 0
|
||
|
||
# 滑动翻页
|
||
d.swipe_ext("up", scale=SCROLL_DISTANCE_RATIO)
|
||
await asyncio.sleep(WAIT_AFTER_SCROLL)
|
||
|
||
# 等待后台任务完成
|
||
if background_tasks:
|
||
logger.info(f"等待 {len(background_tasks)} 个后台分析任务完成...")
|
||
await asyncio.gather(*background_tasks, return_exceptions=True)
|
||
|
||
return processed_count
|
||
|
||
async def analyze_detail_background(self, station_name, image_paths, address=None):
|
||
"""
|
||
后台异步分析详情页 (支持多张截图合并)
|
||
"""
|
||
try:
|
||
if isinstance(image_paths, str):
|
||
image_paths = [image_paths]
|
||
|
||
logger.info(f"开始后台分析场站: {station_name} (图片数: {len(image_paths)})")
|
||
|
||
all_prices = []
|
||
for img_path in image_paths:
|
||
prices = await self.read_image_kit.analyze_detail_price(img_path)
|
||
if prices:
|
||
all_prices.extend(prices)
|
||
|
||
# 去重合并
|
||
unique_prices = []
|
||
seen_periods = set()
|
||
|
||
for p in all_prices:
|
||
start = str(p.get('start', '')).strip()
|
||
end = str(p.get('end', '')).strip()
|
||
|
||
if not start or not end:
|
||
continue
|
||
|
||
key = f"{start}-{end}"
|
||
if key not in seen_periods:
|
||
seen_periods.add(key)
|
||
unique_prices.append(p)
|
||
|
||
# 按开始时间排序
|
||
unique_prices.sort(key=lambda x: x.get('start', '00:00'))
|
||
|
||
if unique_prices:
|
||
await self.service.process_price_detail_data(station_name, unique_prices, address=address)
|
||
logger.info(f"场站 {station_name} 价格分析完成并入库 (记录数: {len(unique_prices)}, 地址: {address})")
|
||
else:
|
||
logger.warning(f"场站 {station_name} 未识别到价格信息")
|
||
|
||
except Exception as e:
|
||
logger.error(f"后台分析 {station_name} 失败: {e}")
|
||
finally:
|
||
# 调试阶段暂时不删除截图，方便排查 VLM 识别失败原因
|
||
pass
|
||
# if isinstance(image_paths, list):
|
||
# for p in image_paths:
|
||
# if os.path.exists(p): os.remove(p)
|
||
# elif isinstance(image_paths, str) and os.path.exists(image_paths):
|
||
# os.remove(image_paths)
|
||
|
||
async def main(service=None):
|
||
if service is None:
|
||
service = YiLaiTeService()
|
||
await service.init_db()
|
||
|
||
crawler = YiLaiTeCrawler(service)
|
||
d = u2.connect()
|
||
|
||
# 清理 Redis
|
||
await crawler.clean_redis_data()
|
||
|
||
# 记录任务日志
|
||
task_id = str(uuid.uuid4())
|
||
await service.log_task_start(task_id)
|
||
|
||
total_count = 0
|
||
status = "success"
|
||
error_msg = None
|
||
|
||
try:
|
||
total_count = await crawler.crawl_list_logic(d)
|
||
except Exception as e:
|
||
logger.error(f"主流程执行失败: {e}")
|
||
status = "failed"
|
||
error_msg = str(e)
|
||
finally:
|
||
await service.log_task_end(task_id, total_count, status, error_msg)
|
||
# 如果是内部初始化的 service，则关闭
|
||
if service and not isinstance(service, YiLaiTeService):
|
||
await service.close_db()
|
||
|
||
async def get_image_md5_async(path):
|
||
return get_image_md5(path)
|
||
|
||
if __name__ == "__main__":
|
||
asyncio.run(main())
|