• python做本地数据与数据库的校验


    # coding=utf-8
    from pymongo import MongoClient
    #连接mongo数据库
    env = {
        'dev': {
            'mongodb_host': 'mongodb://账号:密码@服务地址:端口/数据库',
            'mongodb_db': 'titan1',
        },
        'online': {
            'mongodb_host': 'mongodb://账号:密码@服务名:端口/数据库',
            'mongodb_db': 'titan',
        }
    }
    current_env = 'dev'
    client = MongoClient(env[current_env]['mongodb_host'])
    database = client[env[current_env]['mongodb_db']]
    my_collection=database['car']#连接车辆集合
    #需要迁移的店铺
    def moveback_shop():
        file = open('/Users/huzhihui/Desktop/change_result.txt','a')
        with open('/Users/huzhihui/Desktop/moveback_shop.txt','r') as f:
            lines = f.readlines()#读取回迁的店铺数据[回迁店铺code 新site_id 新sites 裂变后的大风车店铺B]
        n=0
        for line in lines:
            n+=1
            print  '执行第'+str(n)+''
            file.write('执行第'+str(n)+'')
            # 拿到回迁店铺code
            shop_code = line.strip('
    ').split()[0]
            b_shop_code = line.strip('
    ').split()[3]
            # 筛选出工具化二手车和工具化新车集合
            data1 = my_collection.find({'store':"%s"%shop_code})
            #大风车商家发的二手车和新车已经变成新的code,老的code查不到
            if data1.count()!= 0 :
                # 拿到数据库变更后单个车辆的site_id和sites
                x=0
                for single_data in data1:
                    x+=1
                    print '执行第'+str(x)+'个车辆
    '
                    file.write('执行第'+str(x)+'个车辆
    ')
                    # 获取site_id
                    changed_site_id = single_data.get('site_id')
                    # 获取site数组
                    changed_sites = single_data.get('sites')
                    #获取source
                    source=single_data.get('source')
                    if source == 'stock_item':
                        #判断是不是集采新车
                        # 拿到文件中基础服务提供的新的site_id
                        new_site_id=line.strip('
    ').split()[1]
                        # 拿到文件中基础服务提供的新的sites数组
                        new_sites=line.strip('
    ').split()[2].split(',')
                        new_sites=map(int,new_sites)
                        #对比数据库中新的site_id和店铺服务提供的新的site_id是否相等
                        if changed_site_id == new_site_id:
                            print shop_code+'车辆id:'+str(single_data['_id'])+':site_id成功变更为:'+new_site_id
                            file.write(shop_code+'车辆id:'+str(single_data['_id'])+':site_id成功变更为:'+new_site_id+'
    ')
                        else:
                            print shop_code+'车辆id:'+str(single_data['_id'])+':site_id变更失败'
                            file.write(shop_code+'车辆id:'+str(single_data['_id'])+':site_id变更失败'+'
    ')
                        if changed_sites == new_sites:
                            print shop_code+'车辆id:'+str(single_data['_id'])+ ':sites成功变更'
                            file.write(shop_code+'车辆id:' +str(single_data['_id'])+ ':sites成功变更' + '
    ')
                        else:
                            print shop_code+'车辆id:' +str(single_data['_id'])+ ':sites变更失败'
                            file.write(shop_code+'车辆id:'+str(single_data['_id']) + ':sites变更失败' + '
    ')
                    else:
                        print shop_code+'车辆id:'+str(single_data['_id'])+'变更失败,应该变更店铺code为:'+b_shop_code+'
    '
                        file.write(shop_code+'车辆id:'+str(single_data['_id'])+'变更失败,应该变更店铺code为:'+b_shop_code+'
    ')
            else:
                print '查询失败,无法查询到集采新车'+'
    '
                file.write('查询失败,无法查询到集采新车'+'
    ')
            t=0
            data2 = my_collection.find({'store': "%s" % b_shop_code})
            if data2.count()!=0:
                for car_data in data2:
                    t+=1
                    print '执行第' + str(t) + '个车辆
    '
                    file.write('执行第' + str(t) + '个车辆
    ')
                    source = car_data.get('source')
                    if  source =='stock_item':
                        print shop_code +'车辆id:'+str(car_data['_id'])+'变更失败,不应该变更店铺code为:'+b_shop_code+'
    '
                        file.write(shop_code +'车辆id:'+str(car_data['_id'])+'变更失败,不应该变更店铺code为:'+b_shop_code+'
    ')
                    else:
                        print shop_code +'车辆id:'+str(car_data['_id'])+ '成功变更为:' + b_shop_code + '
    '
                        file.write(shop_code +'车辆id:'+str(car_data['_id'])+ '成功变更为:' + b_shop_code + '
    ')
            else:
                print '查询失败,新老code无法查询到任何车辆'
        file.close()
        f.close()
    if __name__=='__main__': moveback_shop() #!/usr/bin/env python # -*- coding: utf-8 -*- import pymysql from pymongo import MongoClient import psycopg2 def sql1(): db = pymysql.connect(host="", user="", password="", db="", port=3306, charset='utf8') #已二进制格式打开文件,用于读取 with open("/Users/huzhihui/Desktop/shop.txt", "rb") as f: lines = f.readlines() #记录错误数据 rf = open("/Users/huzhihui/Desktop/错误数据1", "wb") #记录错误次数 ere = 0 #记录比对的数据数量 n = 0 for line in lines: line = line.strip(' ') cur = db.cursor() print line # print isinstance(line,str) try: cur.execute('select * from baseshop WHERE code="%s"'%line) # cur.execute('select * from baseshop WHERE code=" 13088303331"') # print 'select * from baseshop WHERE code="%s"'%line # 获取所有记录列表 results = cur.fetchall() assert results[0][0]==1954 print results[0][0] print results[0][1] print results[0][2] except Exception,e: ere += 1 rf.writelines(str(e)+ "------" + line) finally: n += 1 print "run:" + "" + str(n) + "错误" + str(ere) def psycopg1(): conn = psycopg2.connect(database="cheniu_trade_dev1", user="dev1_trade", password="dev1_trade_pwd", host="", port="5432") cur = conn.cursor() cur.execute('select * from cheniu_orders where id = 113395') rows = cur.fetchall() for row in rows: print "ID = ", row[0] print "NAME = ", row[1] print "ADDRESS = ", row[2] print "SALARY = ", row[3], " " def readFromFile(): f = open("/Users/zwq/Desktop/shop_old.txt", "rb") lines = f.readlines() for line in lines: shopcode, oldcode = line.strip(' ').split(' ') print "shopcode:" + shopcode + " oldcode:" + oldcode f.close() def writeToFile(): rf = open("/Users/zwq/Desktop/shop_add.txt", "w+") f = open("/Users/zwq/Desktop/shop.txt", "r+") #从shop.txt读数据,生成两条数据并写入shop_add.txt lines = f.readlines() for line in lines: line = line.strip(' ') rf.write(line + ' ' + line + "old" + " ") #去除最后一行的换行 rf.seek(-1, 2) if rf.read(1) == " ": rf.seek(-1, 2) rf.truncate() rf.close() f.close() def readFromFile1(): f = open("/Users/zwq/Desktop/shop_old.txt", "rb") n = 0 f.seek(-1,2) while True: n += 1 print str(n) + "-------" lines = f.read(1) print lines if n ==30: break if lines == " ": print "oooooooooooooo" def writeToFile1(): rf = open("/Users/zwq/Desktop/shop_old.txt", "r+") f = open("/Users/zwq/Desktop/shop.txt", "r+") lines = f.readlines() for line in lines: line = line.strip(' ') rf.write(line + ' ' + line + "old" + " ") rf.seek(-1, 2) if rf.read(1) == " ": rf.seek(-1, 2) rf.truncate() rf.close() def mongo2(): conn = MongoClient('115.29.204.169', 27017) db = conn.gaea # 连接mydb数据库,没有则自动创建 db.authenticate("gaea", "gaea") my_collection = db.user #连接user集合 data = my_collection.find({"iid":1111049}) for item in data: print item #print data if __name__ == '__main__': # sql1() # readFromFile()#writeToFile() # writeToFile1() # readFromFile1() # psycopg1() #print ret
  • 相关阅读:
    Linux知识(7)----远程登录 和远程拷贝
    ROS知识(9)----安装Turtlebot2和远程控制Turtlebot2
    几张图理解Roll, Pitch, Yaw的含义
    GIT(1)----更新代码和上传代码操作的步骤
    ROS知识(8)----CMakeLists.txt文件编写的理解
    ROS知识(7)----ROS命令中的单引号`和‘的混淆问题
    ROS知识(6)----卸载ROS系统
    python知识(2)----python的多态
    DWR3.0框架入门(3) —— ScriptSession的维护及优化
    log4j.properties配置详解
  • 原文地址:https://www.cnblogs.com/hzh1028/p/9444096.html
Copyright © 2020-2023  润新知