關於重復行問題:
在SQL Server中則可以自動排出重復行,不需要處理。在Oracle中經常遇到upsert語法,來排出沖突行。在PostgreSQL中,也需要手動排出重復行,否則會爆出錯誤,upsert用法如下:
WITH upsert AS ( UPDATE TEST AS a SET updatetime=b.updatetime, msec=b.msec FROM TEMP_WRITEEVTUPDATETMSTAMP AS b WHERE a.deviceid=b.deviceid AND a.eventtype=b.eventtype RETURNING a.deviceid,a.eventtype ) INSERT INTO TEST SELECT b.deviceid, b.eventtype, b.updatetime,b.msec FROM TEMP_WRITEEVTUPDATETMSTAMP b WHERE NOT EXISTS (SELECT 1 FROM upsert as a WHERE a.deviceid=b.deviceid AND a.eventtype=b.eventtype);
經過測試,發現這種方法效率很低,於是將upsert過程拆分為兩個過程,先insert不重復的數據,然后再執行update:
INSERT INTO TEST Select b.* from TEMP_WRITEEVTUPDATETMSTAMP b, (select deviceid, eventtype, max(ctid) max_ctid from TEMP_WRITEEVTUPDATETMSTAMP group by deviceid, eventtype) c where (b.deviceid = c.deviceid and b.eventtype = c.eventtype and b.ctid = c.max_ctid) and NOT EXISTS (SELECT 1 FROM TEST as a WHERE a.deviceid=b.deviceid AND a.eventtype=b.even UPDATE TEST AS a SET updatetime=b.updatetime, msec=b.msec FROM TEMP_WRITEEVTUPDATETMSTAMP AS b WHERE a.deviceid=b.deviceid AND a.eventtype=b.eventtype;
當時在存儲過程中大量使用該方法來去除重復數據的插入,但是最后發現還有更好的方法來處理,upsert:
INSERT INTO TEST select * from TEMP_WRITEEVTUPDATETMSTAMP on conflict pk_test_deviceid_eventtype do update set updatetime = excluded.updatetime, msec = excluded.msec;
