Oracle SQL对分区表的高效查询：“按大小写”

Question

我得到一个包含大约 1000 亿条记录的表，该表在键 date_insert 上分区。

MY_BIG_TABLE

• id NUMERIC 二级索引
• date_insert DATE 主索引
• date_fact DATE 二级索引
• 数据 BLOB

输入：

• :date_input1
• :date_input2
• :date_input3
• :id

注意：:date_input1 < :date_input2

我想要

1. IF date_fact < :date_input_1 按 date_fact DESC 排序的第一行
2. ELSE IF date_fact < :date_input_2 按 date_fact ASC 排序的第一行
3. ELSE最新的

查询原型应该是：

SELECT date_insert, date_fact, data (

        SELECT
            date_insert, 
            date_fact,
            CASE 
               WHEN (date_fact < :date_input_1 AND id= :id) : 2
               WHEN (date_fact < :date_input_2 AND id= :id) : 1
               ELSE : 0  
            check,

            CASE 
               WHEN (date_fact < :date_input_1 AND id= :id) : NULL
               WHEN (date_fact < :date_input_2 AND id= :id) : data
               ELSE : NULL  
            data

        FROM MY_BIG_TABLE
        WHERE date_insert > :date_input3   
        order by check, DECODE(check, 1, date_fact) ASC,
                        date_fact DESC 

) WHERE ROWNUM <2

我需要做这项艰苦的工作以避免过多地访问此表。 有什么建议可以完全重建查询以使其更简单、更容易吗？

Answer 1

您只寻找一排。所以，做三个不同的子查询，每个得到一行，然后选择你想要的：

select date_insert, date_fact, data
from (select date_insert, date_fact, data
      from ((select date_insert, date_fact, data, 2 as check
             from (select date_insert, date_fact,
                          (CASE WHEN (date_fact < :date_input_1 AND id= :id) then NULL
                                WHEN (date_fact < :date_input_2 AND id= :id) then data
                                ELSE NULL
                           end) as data
                   from MY_BIG_TABLE
                   where date_fact < :date_input_1 and date_insert > :date_input3
                   order by date_fact desc
                  ) t
             where rownum = 1
            ) union all
            (select date_insert, date_fact, data, 1 as check
             from (select date_insert, date_fact,
                          (CASE WHEN (date_fact < :date_input_1 AND id= :id) then NULL
                                WHEN (date_fact < :date_input_2 AND id= :id) then data
                                ELSE NULL
                           end) as data
                   from MY_BIG_TABLE
                   where date_fact < :date_input_2 and date_insert > :date_input3
                   order by date_fact asc
                  ) t
             where rownum = 1
            ) union all
            (select date_insert, date_fact, data, 0 as check
             from (select date_insert, date_fact,
                          (CASE WHEN (date_fact < :date_input_1 AND id= :id) then NULL
                                WHEN (date_fact < :date_input_2 AND id= :id) then data
                                ELSE NULL
                           end) as data
                   from MY_BIG_TABLE
                   where date_insert > :date_input3
                   order by date_fact desc
                  ) t
             where rownum = 1
            )
           ) t
      order by check desc
     ) t
where rownum = 1;

Oracle 应该足够聪明，可以将索引用于子查询，因此它实际上可能运行得非常快。