【ClickHouse】row_number() over (partition by)的幾種實現方法

        hive中有row_number() over (partition by)函數，可以一句SQL實現想要的排序，在ClickHouse中有很多種實現方式，本篇就介紹一下幾種方法。

目錄

1.row_number排序

2.row_number排序後取出rank=1的結果

3.特殊場景

---

1.row_number排序

HIVE中寫法：

select number,
       row_number() over (partition by number order by time desc) as rank
  from table a
 GROUP BY number

ClickHouse寫法：

select number,
       groupArray(time) AS arr_val,
       arrayEnumerate(arr_val) as row_number
  from (select distinct orderid as number,
               toDate(operatetime) as time
          from table
         order by time desc
         ) a
 GROUP BY number

2.row_number排序後取出rank=1的結果

hive寫法：

select orderid
  from (select orderid,
               row_number() over(partition by orderid order by datachange_lasttime desc) as row_num
          from ods_htl_OrderDB.ord_orders
         where d = '${CurrentDate}'
         ) a
 where row_num = 1;

ClickHouse寫法：

方法1：利用groupArray

select orderid, 
       groupArray(1)(datachange_lasttime) as dates
  from (select orderid, 
               datachange_lasttime
          from olap_htlmaindb.tmp_ord_orders_status_s_pre
         ORDER BY orderid, datachange_lasttime desc
        ) a
 group by orderid

方法2：利用max函數實現倒序，如果正序使用min函數即可

select orderid,
       max(datachange_lasttime) as datachange_lasttime
  from olap_htlmaindb.tmp_ord_orders_status_s_pre
 group by orderid

方法3：利用rowNumberInAllBlocks函數

select orderid, status
  from (select orderid, status, rowNumberInAllBlocks() as rank
          from (select orderid, status, datachange_lasttime
                  from olap_htlmaindb.tmp_ord_orders_status_s_pre
                 order by orderid, datachange_lasttime desc
                 ) a
       ) b LIMIT 1 BY orderid

方法4：利用arrayEnumerate函數

select orderid
  from (select orderid,
               groupArray(datachange_lasttime) AS arr_val,
               arrayEnumerate(arr_val) as row_number
          from (select orderid, datachange_lasttime
                  from olap_htlmaindb.tmp_ord_orders_status_s_pre
                 order by datachange_lasttime desc
                 ) a
         GROUP BY number
         ) b
 where row_number = 1

3.特殊場景

要求：

對於以下場景，需要按照orderid分組，按照日期倒序，取最新一條，若日期一致，則隨機取一條作爲結果即可

hive寫法：

select orderid
  from (select orderid,
               status,
               row_number() over(partition by orderid order by datachange_lasttime desc) as row_num
          from from ods_htl_OrderDB.ord_orders
         where d = '${CurrentDate}'
         ) as b
 where row_num = 1

ClickHouse寫法：

 通過上面的案例，我們很容易想到，把上面的結果作爲一個子表，與原表進行關聯，只是這樣關聯，隨便舉一個關聯的寫法：

select a.orderid as orderid_a, a.status as status
  from olap_htlmaindb.tmp_ord_orders_status_s_pre a
 inner join (select orderid, groupArray(1)(datachange_lasttime) as dates
               from (select orderid, datachange_lasttime
                       from olap_htlmaindb.tmp_ord_orders_status_s_pre
                      ORDER BY orderid, datachange_lasttime desc
                     ) a
              group by orderid) b
    on a.orderid = b.orderid
   and cast(a.datachange_lasttime as String) = cast(b.dates [ 1 ] as String) 

這裏我們是先把符合要求的orderid和時間取出來，再回去關聯，取出需要的列，因爲這些函數都有一個缺點是只能有partition by的字段和排序字段，不能有其他字段，所以要返回關聯，所以上面四種方法，ininer join原表，都不能解決上面案例的問題。

這裏就想到了LIMIT 1 BY這個方法，這個方法其實是最有效的，如下：

select orderid, 
       status, 
       datachange_lasttime
  from olap_htlmaindb.tmp_ord_orders_status_s_pre
 order by orderid, datachange_lasttime desc 
 LIMIT 1 BY orderid