MaxCompute 实现增量数据推送(全量比对增量逻辑)
ODPS 2.0 支持了很多新的集合命令(专有云升级到3版本后陆续支持),简化了日常工作中求集合操作的繁琐程度。增加的SQL语法包括:UNOIN ALL、UNION DISTINCT并集,INTERSECT ALL、INTERSECTDISTINCT交集,EXCEPT ALL、EXCEPT DISTINCT补集。语法格式如下: select_statement UNION ALL select_statement; select_statement UNION [DISTINCT] select_statement; select_statement INTERSECT ALL select_statement; select_statement INTERSECT [DISTINCT] select_statement; select_statement EXCEPT ALL select_statement; select_statement EXCEPT [DISTINCT] select_statement; select_statement MINUS ALL select_statement; select_statement MINUS [DISTINCT] select_statement; 用途:分别求两个数据集的并集、交集以及求第二个数据集在第一个数据集中的补集。参数说明:• UNION: 求两个数据集的并集,即将两个数据集合并成一个数据集。• INTERSECT:求两个数据集的交集。即输出两个数据集均包含的记录。• EXCEPT: 求第二个数据集在第一个数据集中的补集。即输出第一个数据集包含而第二个数据集不包含的记录。• MINUS: 等同于EXCEPT。 具体语法参考:https://help.aliyun.com/document_detail/73782.html?spm=5176.11065259.1996646101.searchclickresult.718d3520fmmOJ0 实际项目中有一个利用两日全量数据,比对出增量的需求(推送全量数据速度很慢,ADB/DRDS等产品数据量超过1亿,建议试用增量同步)。我按照旧的JOIN方法和新的集合方法做了下比对验证,试用了下新的集合命令EXCEPT ALL。测试 -- 方法一:JOIN -- other_columns 代表很多列 create table tmp_opcode1 as select * from( select uuid,other_columns,opcode2 from( -- 今日新增+今日变化 select t1.uuid ,t1.other_columns ,case when t2.uuid is null then 'I' else 'U' end AS opcode2 from prject1.table1 t1 left outer join prject1.table1 t2 on t1.uuid=t2.uuid and t2.dt='20200730' where t1.dt='20200731' and(t2.uuid is null or coalesce(t1.other_columns,'')<>coalesce(t2.other_columns,'')) union all -- 今日删除 select t2.uuid ,t2.other_columns ,'D' as opcode2 from prject1.table1 t2 left outer join prject1.table1 t1 on t1.uuid=t2.uuid and t1.dt='20200731' where t2.dt='20200730' and t1.uuid is null)t3)t4 ; Summary: resource cost: cpu 13.37 Core * Min, memory 30.48 GB * Min inputs: prject1.table1/dt=20200730: 32530802 (946172216 bytes) prject1.table1/dt=20200731: 32533538 (947161664 bytes) outputs: prject1.tmp_opcode1: 4506 (271632 bytes) Job run time: 26.000 -- 方法二:集合 -- other_columns 代表很多列 create table tmp_opcode2 as select * from( select t3.* from( -- 今日新增+今日变化 select uuid,other_columns,'I' as opcode2 from( select uuid,other_columns from prject1.table1 where dt = '20200731' except all select uuid,other_columns from prject1.table1 where dt = '20200730')t union all -- 今日删除 select t2.uuid ,t2.other_columns ,'D' as opcode2 from prject1.table1 t2 left outer join prject1.table1 t1 on t1.uuid=t2.uuid and t1.dt='20200731' where t2.dt='20200730' and t1.uuid is null)t3)t4 ; Summary: resource cost: cpu 35.92 Core * Min, memory 74.26 GB * Min inputs: prject1.table1/rfq=20200730: 32530802 (946172216 bytes) prject1.table1/rfq=20200731: 32533538 (947161664 bytes) outputs: prject1.tmp_opcode2: 4506 (259416 bytes) Job run time: 66.000 性能集合的方法与JOIN的方法相比,在资源(1倍)使用和时间(1倍)上都有较多的劣势。建议实际使用JOIN方法。结果通过多种方法比对验证,两种方法的增量识别均正确,可以向下游提供增量数据。