数据迁移一致性测试探索与实践

背景

量级庞大的日志通过mysql不足以支撑业务需求，以前通过任务调度定时跑批从mysql同步到hive存储，这种方式时效性为T+1，也就是说今天的日志，明天才能同步到hive，总而言之时效性不高。为了提高时效性，改为流式计算flink实时同步

• 那么作为测试人员，我们如何保证切换同步方式后的数据正确性呢？通过对比新旧表数据是否一致显然是最简单的方法
• 这次改动涉及600多张表，每一张表的字段数基本在千以上，甚至部分表字段数达万以上，面对如此庞大的数据量，通过人眼一个个去对比显然不太现实

探索与实践

方案一：sql脚本

SELECT column_names, COUNT(*) AS count_diff 
FROM (
	SELECT CONCAT_WS(',',A,B) FROM udc_test.s000 WHERE dt='20230814'
	UNION ALL 
	SELECT CONCAT_WS(',',A,B) FROM test.s000 WHERE dt = '20230814' and rule_log_id in (select rule_log_id from udc_test.s000) 
) AS combined
GROUP BY column_names
HAVING COUNT(column_names) = 1

select * from (
	select 'table1',A,B from udc_test.s000 WHERE dt='20230814' and rule_log_id in ('123456')
	union all 
	select 'table2',A,B from test.s000 WHERE dt='20230814' and rule_log_id in ('123456')
)a order by a.table1 asc

方案二：python脚本

from pyhive import hive
from datetime import datetime

if __name__ == '__main__':
    #换成生产的连接
    conn = hive.Connection(host="xxx", port='xxx', auth="xxx", database='xxx', username='xxx',password='xxx')

    #这里换成需要比较的表名
    tableName1 = 'test.ssc_python_compare_fields1'
    tableName2 = 'test.ssc_python_compare_fields2'

    current_time = datetime.now()
    hash_code = str(hash(current_time))
    # 获取表结构
    query1 = 'desc ' + tableName1
    query2 = 'desc ' + tableName2
    cursor = conn.cursor()
    cursor.execute(query1)
    columns1 = [row[0] for row in cursor.fetchall()]
    cursor.execute(query2)
    columns2 = [row[0] for row in cursor.fetchall()]

    # 去除掉不需要比较的字段
    columns1.remove('# Partition Information')
    columns1.remove('# col_name')
    columns1.remove('dt')

    columns2.remove('# Partition Information')
    columns2.remove('# col_name')
    columns2.remove('dt')

    set1 = set(columns1)
    set2 = set(columns2)
    # 取出来表1特有的字段，可以保存到文件
    diffrence1 = set1 - set2
    print(diffrence1)
    # 取出来表2特有的字段，可以保存到文件
    diffrence2 = set2 - set1
    print(diffrence2)

    # 取表1和表2共有的字段，用于比较差异
    intersection = set1 & set2

    # 生成比较的sql
    sql = 'select  '
    for element in intersection:
        sql = sql + 'if( nvl(t1.' + element + ',' + hash_code + ' )!= nvl( t2.' + element + ',' + hash_code + ') , \'no\',\'yes\') as ' + element + ' , '
    #print(sql)
    sql = sql[:-2]
    #print(sql)
    #sql中的dt可以改成具体需要比较的日期
    sql = sql + ' from ' + tableName1 + ' as t1 left join  ' + tableName2 \
          + ' as t2  on  t1.rule_log_id=t2.rule_log_id ' \
            ' and t1.dt= \'20230815\' and t2.dt = \'20230815\'  and t1.apply_type=t2.apply_type   where  '
    for element in intersection:
        sql = sql + ' t1.' + element + '!=t2.' + element + ' or '

    sql = sql[:-3]
    print(sql)

    sql = sql + ' limit 1 '

    # 执行sql,获取到结果，如果两列不相等的话，值为no，相等的话值为yes
    cursor.execute(sql)
    result = cursor.fetchone()
    # print(result)

    # 获取上述sql的元数据信息
    metadatas = cursor.description
    print('============================================================')
    # 遍历结果集，查找出比较结果不相同的数据，拿到列名
    index = 0
    while index < len(metadatas):
        if (result[index] != 'yes'):
            print(metadatas[index][0])
        index += 1
    print('============================================================')