【发布时间】：2016-10-31 16:07:38
【问题描述】：
当我们在分区表中插入数据时尝试使用 DML 错误日志记录时遇到了一些问题。
在我们的程序中，我们从一些目标表的不同表中加载数据。在我们的一个程序中，我们将数据插入到一个表中
- 是分区表
- 有一个唯一索引
- 一列或多列没有空约束
目前，在将数据插入最终表之前，我们不会删除违反 NULL/UNIQUE 约束的记录。当我们尝试插入违反 NULL 约束或 UNIQUE 约束的记录时，我们会在错误日志表中得到一些重复。
例如。如果我们有 2 条违反 NULL 约束的记录和 1 条违反 UNIQUE 约束的记录，那么错误日志表应该有 3 条记录。但在这里我们得到 4 或 5 条记录。
Oracle 版本详情-
Oracle 数据库 11g 企业版 11.2.0.3.0 - 64 位生产 PL/SQL 版本 11.2.0.3.0 - 生产 核心 11.2.0.3.0 生产
以下是创建表以复制此问题的示例查询
/* Sample table & index creation -- START */
drop table tmp_table purge;
create table tmp_table
(
CREATE_DT date
,REC_SEQ_NBR number( 12 )
,REC_TYP_CD varchar2( 3 byte )
,EMP_NBR number( 6 )
,DEPT_NBR number( 8 )
);
-- create final table with partitions to insert data into it
drop table tmp_final_table purge;
create table tmp_final_table
(
CREATE_DT date not null
,REC_SEQ_NBR number( 12 ) not null
,REC_TYP_CD varchar2( 3 byte ) not null
,EMP_NBR number( 6 )
,DEPT_NBR number( 8 )
)
nocompress
tablespace ALL_DATA_4M_01_D
result_cache (mode default)
pctused 0
pctfree 5
initrans 1
maxtrans 255
storage( buffer_pool default flash_cache default cell_flash_cache default )
partition by range
(create_dt)
(
partition
PM201603
values less than
(to_date( ' 2016-03-01 00:00:00'
,'SYYYY-MM-DD HH24:MI:SS'
,'NLS_CALENDAR=GREGORIAN'
))
logging
nocompress
tablespace ALL_DATA_4M_01_D
pctfree 5
initrans 1
maxtrans 255
storage( initial 4 m
next 4 m
maxsize unlimited
minextents 1
maxextents unlimited
pctincrease 0
buffer_pool default
flash_cache default
cell_flash_cache default
)
,partition
PM201606
values less than
(to_date( ' 2016-06-01 00:00:00'
,'SYYYY-MM-DD HH24:MI:SS'
,'NLS_CALENDAR=GREGORIAN'
))
logging
nocompress
tablespace ALL_DATA_4M_01_D
pctfree 5
initrans 1
maxtrans 255
storage( initial 4 m
next 4 m
maxsize unlimited
minextents 1
maxextents unlimited
pctincrease 0
buffer_pool default
flash_cache default
cell_flash_cache default
)
,partition
PM201609
values less than
(to_date( ' 2016-09-01 00:00:00'
,'SYYYY-MM-DD HH24:MI:SS'
,'NLS_CALENDAR=GREGORIAN'
))
logging
nocompress
tablespace ALL_DATA_4M_01_D
pctfree 5
initrans 1
maxtrans 255
storage( initial 4 m
next 4 m
maxsize unlimited
minextents 1
maxextents unlimited
pctincrease 0
buffer_pool default
flash_cache default
cell_flash_cache default
) )
nocache
noparallel
monitoring;
-- create a primary key
create unique index tmp_final_table_pk
on dkharwandikar.tmp_final_table( create_dt, rec_seq_nbr )
logging
tablespace ALL_DATA_4M_01_I
pctfree 10
initrans 2
maxtrans 255
storage( initial 4 m
next 4 m
maxsize unlimited
minextents 1
maxextents unlimited
pctincrease 0
buffer_pool default
flash_cache default
cell_flash_cache default
)
noparallel;
;
/* Sample table & index creation -- END */
/* PLSQL block to insert data in the final table - START */
begin
insert /*+ parallel(6) */
into tmp_final_table( create_dt
,REC_SEQ_NBR
,REC_TYP_CD
,EMP_NBR
,DEPT_NBR
)
select create_dt
,REC_SEQ_NBR
,REC_TYP_CD
,emp_nbr
,dept_nbr
from tmp_table
log errors into err$_tmp_final_table ( 'INSERT' )
reject limit unlimited;
commit;
end;
/* PLSQL block to insert data in the final table - END */
以下是不同的场景和结果（请在测试下面提到的每个场景之前截断/删除并创建表格）
-- Case 1- Few records violating UNIQUE constraint and few records violating NULL constraint
-- Result- For each record violating NULL constraint, we can see 2 records in the error log table.
-- Use below data set to test this scenario
insert into tmp_table values ( to_date('04192016','MMDDYYYY'), 1, 'AA', 1, 10 );
insert into tmp_table values ( to_date('03292016','MMDDYYYY'), 2, 'AA', 2, 10 );
insert into tmp_table values ( to_date('12132015','MMDDYYYY'), 3, 'AB', 3, 10 );
insert into tmp_table values ( to_date('01102015','MMDDYYYY'), 4, 'AA', 4, 10 );
insert into tmp_table values ( to_date('06092014','MMDDYYYY'), 5, 'AC', 5, 10 );
insert into tmp_table values ( to_date('05152016','MMDDYYYY'), 6, 'AA', 6, 10 );
insert into tmp_table values ( to_date('02182015','MMDDYYYY'), 7, 'DH', 7, 10 );
insert into tmp_table values ( to_date('02122011','MMDDYYYY'), 8, 'XX', 8, 10 );
insert into tmp_table values ( to_date('03102016','MMDDYYYY'), 9, 'AB', 9, 10 );
insert into tmp_table values ( to_date('12232015','MMDDYYYY'), 10, 'AA', 10, 10 );
-- insert some records with REC_TYP_CD value as NULL. This is having NULL constraint on final table
insert into tmp_table values ( to_date('04192016','MMDDYYYY'), 11, null, 1, 10 );
insert into tmp_table values ( to_date('03292016','MMDDYYYY'), 2, null, 2, 10 );
insert into tmp_table values ( to_date('12132015','MMDDYYYY'), 12, null, 3, 10 );
-- insert some duplicates
insert into tmp_table values ( to_date('04192016','MMDDYYYY'), 1, 'AA', 1, 10 );
insert into tmp_table values ( to_date('03292016','MMDDYYYY'), 2, 'AA', 2, 10 );
COMMIT;
--
-- Case 2- No record violating NULL constraint; few records with UNIQUE constraint violation
-- Result- For first record causing duplicate, we can see 2 records in the error log table. So, in below example, we should get 3 records in error log but we get 4 records.
-- Use below data set to test this scenario
insert into tmp_table values ( to_date('04192016','MMDDYYYY'), 1, 'AA', 1, 10 );
insert into tmp_table values ( to_date('03292016','MMDDYYYY'), 2, 'AA', 2, 10 );
insert into tmp_table values ( to_date('12132015','MMDDYYYY'), 3, 'AB', 3, 10 );
insert into tmp_table values ( to_date('01102015','MMDDYYYY'), 4, 'AA', 4, 10 );
insert into tmp_table values ( to_date('06092014','MMDDYYYY'), 5, 'AC', 5, 10 );
insert into tmp_table values ( to_date('05152016','MMDDYYYY'), 6, 'AA', 6, 10 );
insert into tmp_table values ( to_date('02182015','MMDDYYYY'), 7, 'DH', 7, 10 );
insert into tmp_table values ( to_date('02122011','MMDDYYYY'), 8, 'XX', 8, 10 );
insert into tmp_table values ( to_date('03102016','MMDDYYYY'), 9, 'AB', 9, 10 );
insert into tmp_table values ( to_date('12232015','MMDDYYYY'), 10, 'AA', 10, 10 );
-- insert some duplicates
insert into tmp_table values ( to_date('04192016','MMDDYYYY'), 1, 'AA', 1, 10 );
insert into tmp_table values ( to_date('03292016','MMDDYYYY'), 2, 'AA', 2, 10 );
insert into tmp_table values ( to_date('12132015','MMDDYYYY'), 3, 'AB', 3, 10 );
COMMIT;
--
-- Case 3- Try to insert duplicate records by running insert query twice.
-- Result- For first run, it will insert all the data & in 2nd run it will insert all records in error log. In error log you can see one additional record is present.
-- Use below data set to test this scenario
-- In below example, in 2nd run, we should get 10 records in error log. But we get 11 records in error log.
insert into tmp_table values ( to_date('04192016','MMDDYYYY'), 1, 'AA', 1, 10 );
insert into tmp_table values ( to_date('03292016','MMDDYYYY'), 2, 'AA', 2, 10 );
insert into tmp_table values ( to_date('12132015','MMDDYYYY'), 3, 'AB', 3, 10 );
insert into tmp_table values ( to_date('01102015','MMDDYYYY'), 4, 'AA', 4, 10 );
insert into tmp_table values ( to_date('06092014','MMDDYYYY'), 5, 'AC', 5, 10 );
insert into tmp_table values ( to_date('05152016','MMDDYYYY'), 6, 'AA', 6, 10 );
insert into tmp_table values ( to_date('02182015','MMDDYYYY'), 7, 'DH', 7, 10 );
insert into tmp_table values ( to_date('02122011','MMDDYYYY'), 8, 'XX', 8, 10 );
insert into tmp_table values ( to_date('03102016','MMDDYYYY'), 9, 'AB', 9, 10 );
insert into tmp_table values ( to_date('12232015','MMDDYYYY'), 10, 'AA', 10, 10 );
COMMIT;
--
-- Case 4- Few records violating NULL constraint; no records with UNIQUE constraint violation
-- Result- For each record violating NULL constraint we can see 1 records in the error log table.
-- No Issue
--
-- Case 4- Use any of the scenario (case 1, 2 or 3) data setup with CURSOR FORALL-BULK COLLECT- INSERT
-- Result- For each record violating constraint we can see 1 records in the error log table.
-- No Issue
-- code to try this scenario
declare
type t_create_dt is table of tmp_final_table.create_dt%type
index by binary_integer;
v_create_dt t_create_dt;
type t_rec_seq_nbr is table of tmp_final_table.rec_seq_nbr%type
index by binary_integer;
v_rec_seq_nbr t_rec_seq_nbr;
type t_rec_typ_cd is table of tmp_final_table.rec_typ_cd%type
index by binary_integer;
v_rec_typ_cd t_rec_typ_cd;
type t_emp_nbr is table of tmp_final_table.emp_nbr%type
index by binary_integer;
v_emp_nbr t_emp_nbr;
type t_dept_nbr is table of tmp_final_table.dept_nbr%type
index by binary_integer;
v_dept_nbr t_dept_nbr;
type t_cmd_cur is ref cursor;
cmd_cur t_cmd_cur;
v_sql_stmt varchar2 ( 2500 ) := null;
n_commit_cnt number := 2;
n_tot_rec_load_cnt number := 0;
begin
v_sql_stmt := ' select create_dt
,rec_seq_nbr
,rec_typ_cd
,emp_nbr
,dept_nbr
from tmp_table ';
open cmd_cur for v_sql_stmt;
loop
fetch cmd_cur
bulk collect into v_create_dt
,v_rec_seq_nbr
,v_rec_typ_cd
,v_emp_nbr
,v_dept_nbr
limit n_commit_cnt;
if cmd_cur%rowcount > n_tot_rec_load_cnt
then
forall j in v_create_dt.first .. v_create_dt.last
insert
into tmp_final_table ( create_dt
,rec_seq_nbr
,rec_typ_cd
,emp_nbr
,dept_nbr
)
values (
v_create_dt ( j )
,v_rec_seq_nbr ( j )
,v_rec_typ_cd ( j )
,v_emp_nbr ( j )
,v_dept_nbr ( j ) )
log errors into err$_tmp_final_table ( 'INSERT' )
reject limit unlimited;
n_tot_rec_load_cnt := n_tot_rec_load_cnt + sql%rowcount;
end if;
commit;
exit when cmd_cur%notfound;
end loop;
close cmd_cur;
commit;
end;
我知道我们可以处理 tmp_table 本身的这些问题并删除导致问题的记录。但我想知道这里出了什么问题？仅当我们在分区表上使用它时才会发生这种情况。如果我们从 final table 中删除分区，问题就会得到解决。
我们在许多程序中都遵循类似的逻辑和表/索引设置，我想避免修改这些程序。这是预期的行为吗？
在其中一个 oracle 论坛中，我看到错误表中的唯一约束对于直接路径插入的工作方式不同。我相信我没有在这里进行直接路径插入。
如果您需要这方面的更多信息，请告诉我。
【问题讨论】：
标签： database oracle oracle11g partitioning error-logging