python批量读取excel csv文件插入mysql数据库


#python批量读取excel csv文件插入mysql数据库

import os
import csv
import argparse
import pymysql
import sys



class ConnectionDatabase(object):
    # 连接mysql数据库
    def __init__(self, ip, user_name, passwd, db, char='utf8'):
        self.ip = ip
        # self.port = port
        self.username = user_name
        self.passwd = passwd
        self.mysqldb = db
        self.char = char

        self.MySQL_db = pymysql.connect(
            host=self.ip,
            user=self.username,
            password=self.passwd,
            database=self.mysqldb,
            charset=self.char)

    def mysql_findList(self, sql):
        cursor = self.MySQL_db.cursor()
        MySQL_sql = sql
        results = None
        if not cursor:
            raise (NameError, "数据库连接失败")
        try:
            # 执行SQL语句
            cursor.execute(MySQL_sql)
            # 获取所有记录列表
            results = cursor.fetchall()
        except Exception as e:
            print(e)
            self.MySQL_db.close()
        if results:
            return results
        else:
            return None

    # 数据增删改查(sqlserver)
    def mysql_exe_sql(self, sql, params):
        cursor = self.MySQL_db.cursor()
        MySQL_sql = sql
        result = 0
        if not cursor:
            raise (NameError, "数据库连接失败")
        try:
            # 执行SQL语句
            self.MySQL_db.ping(True)
            cursor.execute(MySQL_sql, params)
            result = cursor.rowcount
        except Exception as e:
            print(e)
            self.MySQL_db.rollback()
            self.MySQL_db.close()

        return result > 0

    '''
        提交数据
    '''
    def commitData(self):
        try:
            self.MySQL_db.commit()
        except Exception as e:
            print(e)

    '''
        关闭数据库连接
    '''
    def closeConn(self):
        if self.MySQL_db:
            self.MySQL_db.close()


'''
    读取文件夹下的csv文件
'''
def readAllFiles(filePath):
    fileList = os.listdir(filePath)
    i = 0
    for file in fileList:
        path = os.path.join(filePath, file)
        if os.path.isfile(path):
            file = open(path, 'r', encoding='utf-8')
            print(path)
            i += 1
            print("插入第>>>>", i, ">>>>数据表")
            analysisWorkflowCsv(file)
            pass
        else:
            readAllFiles(path)


def analysisWorkflowCsv(file):
    csvFile = csv.reader(file)
    # 读取一行,下面的reader中已经没有该行了
    # head_row = next(csvFile)
    # print(head_row)
    __conn = ConnectionDatabase(ip="localhost", user_name="root", passwd="", db="mydb", char="utf8")
    counter = 0
    for row in csvFile:
        data = {}
        # 获取excel内需要的数据,从0开始,根据导入的csv文件的列数来决定data的容量
        data['a'] = row[0]
        data['b'] = row[1]
        data['c'] = row[2]
        # data['d'] = row[3]
        # data['e'] = row[4]
        # data['f'] = row[5]
        # data['g'] = row[6]
        # data['h'] = row[7]
        # data['i'] = row[8]
        # data['j'] = row[9]
        if insert_data(__conn, data):
            counter += 1
        if counter % 1000 == 0:
            __conn.commitData()
    print("已经插入工作流数据: %d 条。" % counter)
    __conn.commitData()
    __conn.closeConn()


'''
    插入工作流程数据
'''
def insert_data(__conn, data):
    # 在mysql建立表,字段名可以根据需要设置,也可以按a,b,c这样的简单记录也可以。跟data容量的一一对应。
    # __sql = '''
    #     INSERT INTO `mydb`.`tbl_test` (
    #   `a`,
    #   `b`,
    #   `c`,
    #   `d`,
    #   `e`,
    #   `f`,
    #   `g`,
    #   `h`,
    #   `i`,
    #   `j`
    # )
    # VALUES
    #   (
    #     %s,
    #     %s,
    #     %s,
    #     %s,
    #     %s,
    #     %s,
    #     %s,
    #     %s,
    #     %s,
    #     %s
    #   )
    # '''

    __params = (
        data['a'], data['b'],
        data['c'],
        data['d'],
        data['e'],data['f'],data['g'],data['h'],data['i'],data['j']
    )
    # print(__sql % __params)
    return __conn.mysql_exe_sql(__sql, __params)


if __name__ == "__main__":
    # 文件所在的文件夹父路径,按文件夹下面的文件批量导入
    testFilePath = "F:/excel2Mysql/"
    readAllFiles(testFilePath)
    

 

原创文章,作者:ItWorker,如若转载,请注明出处:https://blog.ytso.com/280679.html

(0)
上一篇 2022年8月15日
下一篇 2022年8月15日

相关推荐

发表回复

登录后才能评论