Python3实现的爬虫爬取数据并存入mysql数据库操作示例

本文实例讲述了Python3实现的爬虫爬取数据并存入mysql数据库操作。分享给大家供大家参考,具体如下:

爬一个电脑客户端的订单。罗总推荐,抓包工具用的是HttpAnalyzerStdV7,与chrome自带的F12类似。客户端有接单大厅,罗列所有订单的简要信息。当单子被接了,就不存在了。我要做的是新出订单就爬取记录到我的数据库zyc里。

设置每10s爬一次。

抓包工具页面如图:

Python3实现的爬虫爬取数据并存入mysql数据库操作示例

首先是爬虫,先找到数据存储的页面,再用正则爬出。

# -*- coding:utf-8 -*-
import re
import requests
import pymysql #Python3的mysql模块,Python2 是mysqldb
import datetime
import time
def GetResults():
  requests.adapters.DEFAULT_RETRIES = 5 #有时候报错,我在网上找的不知道啥意思,好像也没用。
  reg = [r'"id":(.*?),',
      r'"order_no":"(.*?)",',
      r'"order_title":"(.*?)",',
      r'"publish_desc":"(.*?)",',
      r'"game_area":"(.*?)\\/(.*?)\\/(.*?)",',
      r'"order_current":"(.*?)",',
      r'"order_content":"(.*?)",',
      r'"order_hours":(.*?),',
      r'"order_price":"(.*?)",',
      r'"add_price":"(.*?)",',
      r'"safe_money":"(.*?)",',
      r'"speed_money":"(.*?)",',
      r'"order_status_desc":"(.*?)",',
      r'"order_lock_desc":"(.*?)",',
      r'"cancel_type_desc":"(.*?)",',
      r'"kf_status_desc":"(.*?)",',
      r'"is_show_pwd":(.*?),',
      r'"game_pwd":"(.*?)",',
      r'"game_account":"(.*?)",',
      r'"game_actor":"(.*?)",',
      r'"left_hours":"(.*?)",',
      r'"created_at":"(.*?)",',
      r'"account_id":"(.*?)",',
      r'"mobile":"(.*?)",',
      r'"contact":"(.*?)",',
      r'"qq":"(.*?)"},']
  results=[]
  try:
    for l in range(1,2):   #页码
      proxy = {'HTTP':'61.135.155.82:443'} #代理ip
      html = requests.get('https://www.dianjingbaozi.com/api/dailian/soldier/hall?access_token=3ef3abbea1f6cf16b2420eb962cf1c9a&dan_end=&dan_start=&game_id=2&kw=&order=price_desc&page=%d'%l+'&pagesize=30&price_end=0&price_start=0&server_code=000200000000&sign=ca19072ea0acb55a2ed2486d6ff6c5256c7a0773&timestamp=1511235791&type=public&type_id=%20HTTP/1.1',proxies=proxy) # 用get的方式访问。网页解码成中文。接单大厅页。
      #
      html=html.content.decode('utf-8')
      outcome_reg_order_no = re.findall(r'"order_no":"(.*?)","game_area"', html)  #获取订单编号,因为订单详情页url与订单编号有关。
      for j in range(len(outcome_reg_order_no)):
        html_order = requests.get('http://www.lpergame.com/api/dailian/order/detail?access_token=eb547a14bad97e1ee5d835b32cb83ff1&order_no=' +outcome_reg_order_no[j] + '&sign=c9b503c0e4e8786c2945dc0dca0fabfa1ca4a870&timestamp=1511146154 HTTP/1.1',proxies=proxy)  #订单详细页
        html_order=html_order.content.decode('utf-8')
        # print(html_order)
        outcome_reg = []
        for i in range(len(reg)):#每条订单
          outcome = re.findall(reg[i], html_order)
          if i == 4:
            for k in range(len(outcome)):
              outcome_reg.extend(outcome[k])
          else:
            outcome_reg.extend(outcome)
        results.append(outcome_reg) #结果集
    return results
  except:
    time.sleep(5)  #有时太频繁会报错。
    print("失败")
    pass

根据爬虫结果建表,这里变量名要准确。并且要设置唯一索引,使每次爬的只有新订单入库。

def mysql_create():
  mysql_host = ''
  mysql_db = 'zyc'
  mysql_user = 'zyc'
  mysql_password = ''
  mysql_port = 3306
  db = pymysql.connect(host=mysql_host, port=mysql_port, user=mysql_user, password=mysql_password, db=mysql_db,charset='utf8') # 连接数据库编码注意是utf8,不然中文结果输出会乱码
  sql_create = "CREATE TABLE DUMPLINGS (id CHAR(10),order_no CHAR(50),order_title VARCHAR(265),publish_desc VARCHAR(265),game_name VARCHAR(265),"\
         "game_area VARCHAR(265),game_area_distinct VARCHAR(265),order_current VARCHAR(3908),order_content VARCHAR(3908),order_hours CHAR(10)," \
         "order_price FLOAT(10),add_price FLOAT(10),safe_money FLOAT(10),speed_money FLOAT(10),order_status_desc VARCHAR(265),"\
         "order_lock_desc VARCHAR(265),cancel_type_desc VARCHAR(265),kf_status_desc VARCHAR(265),is_show_pwd TINYINT,game_pwd CHAR(50),"\
         "game_account VARCHAR(265),game_actor VARCHAR(265),left_hours VARCHAR(265),created_at VARCHAR(265),account_id CHAR(50),"\
         "mobile VARCHAR(265),mobile2 VARCHAR(265),contact VARCHAR(265),contact2 VARCHAR(265),qq VARCHAR(265),"\
         "PRIMARY KEY (`id`),UNIQUE KEY `no`(`order_no`))ENGINE=InnoDB AUTO_INCREMENT=12 DEFAULT CHARSET=utf8"
  sql_key="CREATE UNIQUE INDEX id ON DUMPLINGS(id)"
  cursor = db.cursor()
  cursor.execute("DROP TABLE IF EXISTS DUMPLINGS")
  cursor.execute(sql_create)# 执行SQL语句
  cursor.execute(sql_key)
  db.close() # 关闭数据库连

把数据导入Mysql,注意编码和字段之间的匹配。

def IntoMysql(results):
  mysql_host = ''
  mysql_db = 'zyc'
  mysql_user = 'zyc'
  mysql_password = ''
  mysql_port = 3306
  db = pymysql.connect(host=mysql_host, port=mysql_port, user=mysql_user, password=mysql_password, db=mysql_db,charset='utf8') # 连接数据库编码注意是utf8,不然中文结果输出会乱码
  cursor = db.cursor()
  for j in range(len(results)):
    try:
      sql = "INSERT INTO DUMPLINGS(id,order_no,order_title,publish_desc ,game_name," \
         "game_area,game_area_distinct,order_current,order_content,order_hours," \
         "order_price,add_price,safe_money,speed_money,order_status_desc," \
         "order_lock_desc,cancel_type_desc,kf_status_desc,is_show_pwd,game_pwd," \
         "game_account,game_actor,left_hours,created_at,account_id," \
         "mobile,mobile2,contact,contact2,qq) VALUES ("
      for i in range(len(results[j])):
        sql = sql + "'" + results[j][i] + "',"
      sql = sql[:-1] + ")"
      sql = sql.encode('utf-8')
      cursor.execute(sql)
      db.commit()
    except:pass
  db.close()

每十秒运行一次。

mysql_create()
i=0
while True:
  results = GetResults()
  IntoMysql(results)
  i=i+1
  print("爬虫次数:",i)
  time.sleep(10)

结果如图:

 Python3实现的爬虫爬取数据并存入mysql数据库操作示例

更多关于Python相关内容可查看本站专题:《Python Socket编程技巧总结》、《Python正则表达式用法总结》、《Python数据结构与算法教程》、《Python函数使用技巧总结》、《Python字符串操作技巧汇总》、《Python+MySQL数据库程序设计入门教程》及《Python常见数据库操作技巧汇总》

希望本文所述对大家Python程序设计有所帮助。

相关推荐