關於union和join區別和聯絡
union和join是需要聯合多張表時常見的關聯詞,具體概念我就不說了,想知道上網查就行,因為我也記不準確。
先說差別:union對兩張表的操作是合併資料條數,等於是縱向的,要求是兩張表字段必須是相同的(Schema of both sidesof union should match.)。也就是說如果A表中有三條資料,B表中有兩條資料,那麼A unionB就會有五條資料。說明一下union 和union all的差別,對於union如果存在相同的資料記錄會被合併,而unionall不會合並相同的資料記錄,該有多少條記錄就會有多少條記錄。例如在mysql下執行以下語句:
select * from tmp_libingxue_a;
name number
libingxue 1001
yuwen 1002
select * from tmp_libingxue_b;
name number
libingxue 1001
feiyao 1003
select * from tmp_libingxue_a union select * fromtmp_libingxue_b;
libingxue 1001
yuwen 1002
feiyao 1003
select * from tmp_libingxue_a union all select * fromtmp_libingxue_b;
libingxue 1001
yuwen 1002
libingxue 1001
feiyao 1003
但是這樣在hive裡面是不能執行的,執行select * from tmp_libingxue_a union all select* from tmp_libingxue_b;會failed,hive中union必須在子查詢中進行。如
select * from (select * from tmp_yuwen_a union all select * fromtmp_yuwen_b) t1;
注意,必須是unionall,單獨用union它會提示你缺少ALL,而且後面的t1必須寫,你可以寫成a或者b,但是一定要寫,不寫會出錯。
而join則是偏於橫向的聯合,僅僅是偏向於,等下詳細說明。join跟union比起來顯得更寬鬆,對兩個表的欄位不做要求,沒有限制條件的join等於兩個表的笛卡爾乘積,所有join需要有限制條件來約束,經過限制的join就是橫向的擴張了。對於滿足限制條件的join會被提取出來,不滿足的直接過濾掉。用法可以很靈活,下面有兩個簡單的例子:
select * from (select * from tmp_yuwen_a)t1 join (select * fromtmp_yuwen_b) t2;
select * from tmp_yuwen_a t1 join (select * from tmp_yuwen_b)t2;
left outer join和right outer join用法類似,區別就是left outerjoin會把左邊表的欄位全部選擇出來,右邊表的欄位把符合條件的也選擇出來,不滿足的全部置空,也就是說以左邊表為參照。rightouter join同理以右邊表為參照。這三個join之間的差別說過很多次,網上也有更詳細的解釋,不再贅述。
相同點:在某些特定的情況下,可以用join實現union all的功能,這種情況是有條件的,當出現這種情況的時候選擇unionall還是groupby就可以看情況或者看兩者的消耗而決定。sql雖然就在那麼幾個關鍵詞,但變化多端、功能強大,只要能實現想要的功能,怎麼用隨便你。需求情況sql簡單重現如下
drop table tmp_libingxue_resource;
create external table if not exists tmp_libingxue_resource(
user_id string,
shop_id string,
auction_id string,
search_time string
)partitioned by (pt string)
row format delimited fields terminated by '\t'
lines terminated by '\n'
stored as sequencefile;
drop table tmp_libingxue_result;
create external table if not exists tmp_libingxue_result(
user_id string,
shop_id string,
auction_id string,
search_time string
)partitioned by (pt string)
row format delimited fields terminated by '\t'
lines terminated by '\n'
stored as sequencefile;
insert overwrite table tmp_libingxue_result where(pt=20041104)select * from tmp_libingxue_resource;
sudo -u taobao hadoop dfs -rmr/group/tbads/warehouse/tmp_libingxue_result/pt=20041104
sudo -u taobao hadoop jar/home/taobao/dataqa/framework/DailyReport.jarcom.alimama.loganalyzer.tool.SeqFileLoadertmp_libingxue_resource.txthdfs://v039182.sqa.cm4:54310/group/tbads/warehouse/tmp_libingxue_result/pt=20041104/part-00000
hive< select * from tmp_libingxue_resource;
OK
2001 0 11 101 20041104
2002 0 11 102 20041104
hive< select * from tmp_libingxue_result;
OK
2001 0 12 103 20041104
2002 0 12 104 20041104
select user_id,shop_id,max(auction_id),max(search_time)
from
(select * from tmp_libingxue_resource
union all
select * from tmp_libingxue_result )t1
group by user_id,shop_id;
2001 0 12 103
2002 0 12 104
select t1.user_id,t1.shop_id,t2.auction_id,t2.search_time
from
(select * from tmp_libingxue_resource) t1
join
(select * from tmp_libingxue_result) t2
on t1.user_id=t2.user_id and t1.shop_id=t2.shop_id;
2001 0 12 103
2002 0 12 104
寫這麼個東西花的時間比我想象中的要長很多,但是執行別人的sql跟執行自己的sql感覺不一樣,是的,太不一樣了。