SQL：大案例条件GROUPBY子句慢

我需要根据号码范围提取基于LC呼叫号码的报告。呼叫号码格式如下，我需要提取标点符号前的第二个字段进行分组：

CALL_NO_ID1
--------------
a!3243 .m43 12

a#435 234 1999

cs"345 1973.

...

下面是我的sql

select count("CALL_NO_ID1") "No_of_Items", 
case

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KG %') THEN 'KG 0-999  - Federal law Common and collective state law Individual states US - Latin AmericaGeneral'

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KH %') THEN 'KH 0-999 - Federal law Common and collective state law Individual states US -  South AmericaGeneral '

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 1 AND 100)AND ("CALL_NO_DESC1" LIKE 'DE %') THEN 'DE 1-100  - HistoryGeneral - The Mediterranean Region The Greco-Roman World'

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 1 AND 1050)AND ("CALL_NO_DESC1" LIKE 'TR %') THEN 'TR 1-1050  - Photography'

...  

... (around 450 case conditions)

...

else "CALL_NO_ID1"

end "Primary Call"

from DWH_FACT_ITEMS

group by 

case

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KG %') THEN 'KG 0-999  - Federal law Common and collective state law Individual states US - Latin AmericaGeneral'

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KH %') THEN 'KH 0-999 - Federal law Common and collective state law Individual states US -  South AmericaGeneral '

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 1 AND 100)AND ("CALL_NO_DESC1" LIKE 'DE %') THEN 'DE 1-100  - HistoryGeneral - The Mediterranean Region The Greco-Roman World'

WHEN (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 1 AND 1050)AND ("CALL_NO_DESC1" LIKE 'TR %') THEN 'TR 1-1050  - Photography'
...  
... (around 450 case conditions)
...

然而，这将需要很长的时间才能得到结果（2~3hr），我想知道有什么建议可以改进我的sql吗

谢谢

---

莫里斯

我会添加一个额外的列

CALL\u NO\u CLEARED

来保留这个号码。将表达式应用于所有值以填充人造列

您可以在插入/更新时添加触发器

，以在添加或更改列时动态填充该列

然后，您可以在选择的引入索引中使用
CALL\u NO\u CLEARED
，使其更快

更新：

我可以建议另一种方法。似乎最耗时的过程是通话

LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0')

因此，对于每一行，我们计算它450次（对于提到的
中的每一行）

尝试将计算放在子查询中，然后稍后应用组，例如

select *
FROM (
    select 
    CALL_NO_ID1,
    LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') as sub_num
    from DWH_FACT_ITEMS) sub
group by 
case
WHEN (sub.sub_num BETWEEN 0 AND 999)AND (sub.CALL_NO_DESC1 LIKE 'KG %') THEN 'KG 0-999  - Federal law Common and collective state law Individual states US - Latin AmericaGeneral'
...

您有几种可能改进查询。我的测试表明
通过（使用子查询）消除group by中的case可以提高查询的可维护性和大小，但性能保持不变

通过对案例陈述进行排序，观察到了特别的改善，以便将最常见的情况放在顶部

这个想法很简单，如果在案例的早期完成匹配，则跳过其余条件

通过对WHEN语句中的谓词重新排序，实现了更好的改进。如果call\u NO\u DESC子字符串不匹配，则不会调用regexp处理

WHEN ("CALL_NO_DESC1" LIKE 'TR %') and 
(LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') BETWEEN 1 AND 1050)   
THEN 'TR 1-4050  - Photography'

最后一步是在子查询中只调用一次REGEXP处理

总而言之，我以这个查询结束，它大大减少了运行时间（使用我的测试数据）

使用
WITH
子句和
/*+MATERIALIZE*/
提示，让Oracle只执行一次昂贵的操作

这在400000行上的性能应该比2-3小时好得多：

WITH parsed_call_numbers as ( SELECT /*+ MATERIALIZE */
SELECT CALL_NO_ID1,
       (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') call_Number_part,
       CALL_NO_DESC1
from DWH_FACT_ITEMS ) ,
primary_calls AS ( SELECT /*+ MATERIALIZE */ 
CALL_NO_ID1,
case
WHEN call_number_part BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KG %') THEN 'KG 0-999  - Federal law Common and collective state law Individual states US - Latin AmericaGeneral'
WHEN call_number_part BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KH %') THEN 'KH 0-999 - Federal law Common and collective state law Individual states US -  South AmericaGeneral '
WHEN call_number_part BETWEEN 1 AND 100)AND ("CALL_NO_DESC1" LIKE 'DE %') THEN 'DE 1-100  - HistoryGeneral - The Mediterranean Region The Greco-Roman World'
WHEN call_number_part BETWEEN 1 AND 1050)AND ("CALL_NO_DESC1" LIKE 'TR %') THEN 'TR 1-1050  - Photography'
--...  
--... (around 450 case conditions)
--...
else "CALL_NO_ID1"
end "Primary Call"
from parsed_call_numbers )
select count("CALL_NO_ID1") "No_of_Items", "Primary Call"
FROM primary_calls
group by "Primary Call"

您能告诉我们您的
CASE
语句中使用的逻辑和分组过程吗？当分组基于LC分类，并且第一个或“第一和第二个”必须是英文字符时，您需要一个算法来剪切这些CASE。谢谢回复。这是信用证分类，第一个或“第一和第二个”必须是英文字符。然后后跟一个带有字段分隔符“空白”、！、#的数字或“a！3243.m43 12（以分贝记录）-->a（第一字段）-->3243（第二字段）-->.m43（第三字段）a#435 234 1999（以分贝记录）-->a（第一字段）-->234（第二字段）-->1999（第三字段）cs”345 1973。（在db中记录）-->cs（第一个字段）-->345（第二个字段）-->1973。（第三个字段）因为要求根据给定的分组第一和第二个字段提取范围内的记录总数，例如，字段1（以a开头）和字段2范围（1-400）是组字段1（以a开头），字段2范围（500-4000）是B组字段1（以B开头），字段2范围（1-400）是C组。。。。我认为以上情况是不规则的。谢谢你的建议，如果你能应用你的建议，应该会得到很好的结果。但db由供应商维护。修改该表可能违反合同，并可能导致未来升级出现问题。感谢您的建议，我尝试按照您的建议修改sql。执行时间没有差别。嗨，StanislavL，即使我改变了顺序或案例，我也能感觉到显著的改进。非常感谢，莫里斯
WITH parsed_call_numbers as ( SELECT /*+ MATERIALIZE */
SELECT CALL_NO_ID1,
       (LPAD(CAST(regexp_replace(REGEXP_SUBSTR(REGEXP_REPLACE("CALL_NO_ID1",'["]|[#]|[!]', ' '),'[^ ]+|["]|[#]',1,2), '[^0-9]+', '') as number),7,'0') call_Number_part,
       CALL_NO_DESC1
from DWH_FACT_ITEMS ) ,
primary_calls AS ( SELECT /*+ MATERIALIZE */ 
CALL_NO_ID1,
case
WHEN call_number_part BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KG %') THEN 'KG 0-999  - Federal law Common and collective state law Individual states US - Latin AmericaGeneral'
WHEN call_number_part BETWEEN 0 AND 999)AND ("CALL_NO_DESC1" LIKE 'KH %') THEN 'KH 0-999 - Federal law Common and collective state law Individual states US -  South AmericaGeneral '
WHEN call_number_part BETWEEN 1 AND 100)AND ("CALL_NO_DESC1" LIKE 'DE %') THEN 'DE 1-100  - HistoryGeneral - The Mediterranean Region The Greco-Roman World'
WHEN call_number_part BETWEEN 1 AND 1050)AND ("CALL_NO_DESC1" LIKE 'TR %') THEN 'TR 1-1050  - Photography'
--...  
--... (around 450 case conditions)
--...
else "CALL_NO_ID1"
end "Primary Call"
from parsed_call_numbers )
select count("CALL_NO_ID1") "No_of_Items", "Primary Call"
FROM primary_calls
group by "Primary Call"