Teradata 解析器问题

Question

我正在尝试执行以下查询，但出现错误

INSERT INTO TABLEA(
CUSTOMER_CT_KEY,
CUSTOMER_ST_KEY,
CUSTOMER_TEST_KEY,
JAN_AMT,
FEB_AMT, 
MAR_AMT)
SELECT
A.CUSTOMER_CT_KEY,
A.CUSTOMER_ST_KEY,
A.CUSTOMER_TEST_KEY,
SUM(CASE WHEN EXTRACT(MONTH FROM A.DATECOL) = '01'
            THEN A.AAA_AMT
            ELSE 0
           END),
SUM(CASE WHEN EXTRACT(MONTH FROM A.DATECOL) = '02'
            THEN A.BBB_AMT
            ELSE 0
           END),
SUM(CASE WHEN EXTRACT(MONTH FROM A.DATECOL) = '03'
            THEN A.CCC_AMT
            ELSE 0
           END)
FROM TABLEB B, TABLEC C
WHERE B.DATECOL<= C.DATECOL
AND B.CUSTOMER_CT_KEY NOT IN
(SELECT FS_CUSTOMER_CT_KEY FROM TABLED WHERE REF_NBR = 'VALUE')
GROUP BY 1,2,3;

Insert failed. 3899: Internal Error in Teradata SQL parser

输出：

CUSTOMER_CT_KEY CUSTOMER_ST_KEY CUSTOMER_TEST_KEY JAN_AMT FEB_AMT MAR_AMT
123456789       541245812       541245812         114.00  524.00   62.00
658412457       632514257       632514257           0.00   12.00  214.00

中的总行数

TABLEA - EMPTY
TABLEB - 420,098,323
TABLEC - 1
TABLED - 218,074
INNER SUBQUERY - 5

当我尝试对子查询的值进行硬编码时，它起作用了。插入的行数：105,615,541

请指导我如何进一步进行。谢谢

Answer 1

如果您遇到这样的错误，您应该向 Teradata 支持发起一个事件。这是来自 Messages 手册：

3899 Internal error in the Teradata SQL Parser.
Explanation: The Teradata SQL Parser erred.
Generated By: CON, LEX, PAR, SYN, RES and OPT modules.
For Whom: System Support Representative.
Notes: This is usually caused by a request that the Teradata SQL Parser could not correctly process, yet it did not detect
an error.
Remedy: Save all relevant information and notify your support representative.

Answer 2

基本上原始查询想要从 TABLEB 和 TABLEC 中获取一些数据，过滤条件是您不想要 TABLED 中的记录，当 REF_NBR='VALUE'

正如 Tim 所建议的那样，一种可能的解决方法（当您向 Teradata 提出申请时，正如 dnoeth 所建议的那样）可能是使用连接编写 IN 过滤条件。

我对此类查询的建议如下：

SELECT
 A.CUSTOMER_CT_KEY,
 A.CUSTOMER_ST_KEY,
 A.CUSTOMER_TEST_KEY,
 SUM(CASE WHEN EXTRACT(MONTH FROM A.DATECOL) = '01'
        THEN A.AAA_AMT
        ELSE 0
       END),
 SUM(CASE WHEN EXTRACT(MONTH FROM A.DATECOL) = '02'
        THEN A.BBB_AMT
        ELSE 0
       END),
 SUM(CASE WHEN EXTRACT(MONTH FROM A.DATECOL) = '03'
        THEN A.CCC_AMT
        ELSE 0
       END)
  FROM TABLEB B
       JOIN TABLEC C
         ON B.DATECOL<= C.DATECOL
       LEFT JOIN (SELECT FS_CUSTOMER_CT_KEY FROM TABLED
                   WHERE REF_NBR = 'VALUE' ) AS D
         ON B.CUSTOMER_CT_KEY=D.FS_CUSTOMER_CT_KEY
 WHERE D.FS_CUSTOMER_CT_KEY is null
 GROUP BY 1,2,3;

D子查询保存了过滤条件的所有记录（你要过滤掉的）。 "D.FS_CUSTOMER_CT_KEY is null" 条件确保只有 CUSTOMER_CT_KEY 没有出现在子查询 D 中的记录才会出现在结果集中。

测试解决方案

由于对提议的解决方案和实际的 OP 要求存在一些误解，我尝试准备一个基本的测试用例，以便我们可以使用它来更好地理解彼此的解决方案和角落案例。

create multiset volatile table TABLEB (
 CUSTOMER_CT_KEY BIGINT,
 CUSTOMER_ST_KEY BIGINT,
 CUSTOMER_TEST_KEY BIGINT,
 AAA_AMT decimal(38,18),
 DATECOL date
 )
no primary index
on commit preserve rows;

create multiset volatile table TABLEC (
 DATECOL date
 )
no primary index
on commit preserve rows;

create multiset volatile table TABLED (
 FS_CUSTOMER_CT_KEY bigint,
 REF_NBR  VARCHAR(5) 
 )
no primary index
on commit preserve rows;


insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (123456789, 541245812, 541245812, 111, '2016-01-15');
insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (123456789, 541245812, 541245812, 524, '2016-02-15');
insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (123456789, 541245812, 541245812, 63, '2016-03-15');          

insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (777777777, 111222333, 444555666, 42, '2016-03-15');          


insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (658412457, 632514257, 632514257, 0, '2016-01-15');
insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (658412457, 632514257, 632514257, 12, '2016-02-15');
insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (658412457, 632514257, 632514257, 214, '2016-03-15');                                                              

insert into tableb (CUSTOMER_CT_KEY, CUSTOMER_ST_KEY, CUSTOMER_TEST_KEY, AAA_AMT, DATECOL)
                values (777777777, 999888777, 666555444, 42, '2016-03-15');          

insert into tablec ( DATECOL)                
                values ( '2016-04-01');                                                              

insert into tabled ( FS_CUSTOMER_CT_KEY, REF_NBR)                
                values ( 777777777, 'VALUE');      
insert into tabled ( FS_CUSTOMER_CT_KEY, REF_NBR)                
                values ( 658412457, 'OK1');                                                                              
insert into tabled ( FS_CUSTOMER_CT_KEY, REF_NBR)                
                values ( 658412457, 'OK2');                                                                              
insert into tabled ( FS_CUSTOMER_CT_KEY, REF_NBR)                
                values ( 658412457, 'OK3');                                                                              

select 
 B.CUSTOMER_CT_KEY,
 B.CUSTOMER_ST_KEY,
 B.CUSTOMER_TEST_KEY,
 SUM(case when extract(MONTH from b.DATECOL)='01' then b.aaa_amt else 0 end) jan_amt,
SUM(case when extract(MONTH from b.DATECOL)='02' then b.aaa_amt else 0 end) feb_amt,
SUM(case when extract(MONTH from b.DATECOL)='03' then b.aaa_amt else 0 end) mar_amt
 from tableb b, tablec c
 where b.datecol<=c.datecol
 and b.CUSTOMER_CT_KEY not in (select FS_CUSTOMER_CT_KEY from TABLED WHERE REF_NBR='VALUE')
 group by 1,2,3;

 select 
 B.CUSTOMER_CT_KEY,
 B.CUSTOMER_ST_KEY,
 B.CUSTOMER_TEST_KEY,
 SUM(case when extract(MONTH from b.DATECOL)='01' then b.aaa_amt else 0 end) jan_amt,
SUM(case when extract(MONTH from b.DATECOL)='02' then b.aaa_amt else 0 end) feb_amt,
SUM(case when extract(MONTH from b.DATECOL)='03' then b.aaa_amt else 0 end) mar_amt
 from tableb b 
   inner join tablec c
   on  b.datecol<=c.datecol
   left join TABLED D
      ON B.CUSTOMER_CT_KEY=D.FS_CUSTOMER_CT_KEY
             AND D.REF_NBR<>'VALUE'
 group by 1,2,3;

 select 
 B.CUSTOMER_CT_KEY,
 B.CUSTOMER_ST_KEY,
 B.CUSTOMER_TEST_KEY,
 SUM(case when extract(MONTH from b.DATECOL)='01' then b.aaa_amt else 0 end) jan_amt,
SUM(case when extract(MONTH from b.DATECOL)='02' then b.aaa_amt else 0 end) feb_amt,
SUM(case when extract(MONTH from b.DATECOL)='03' then b.aaa_amt else 0 end) mar_amt
 from tableb b 
   inner join tablec c
   on  b.datecol<=c.datecol
   left join (select FS_CUSTOMER_CT_KEY from  TABLED where REF_NBR='VALUE' ) as D
      ON B.CUSTOMER_CT_KEY=D.FS_CUSTOMER_CT_KEY
WHERE D.FS_CUSTOMER_CT_KEY is null
 group by 1,2,3;

根据（有偏见且完全虚构的 :-) 数据，我得到了附加结果。

模拟原始查询

初始提议的模拟

模拟我目前的提议（我已经告诉过你数据是有偏见和虚构的，对吧？:-)）

OP 能否提供一些反馈或一些假数据以插入易变表中？

Answer 3

只是关于这个问题的更新。我已从 Teradata 支持部门获得澄清。在执行插入查询之前，必须执行以下语句。

diagnostic evlinterp on for session;

Teradata 解析器问题

Teradata parser issue

sql

teradata