关于DML Error Logging效率的问题,摘自网上一篇文章,作为单独一篇说明,原文如下:
DML Error Logging in Oracle 10g Database Release 2
In some situations the most obvious solution to a problem is a DML statement (INSERT ... SELECT,UPDATE,DELETE), but you may choose to avoid DML because of the way it reacts to exceptions. By default, when a DML statement fails the whole statement is rolled back, regardless of how many rows were processed successfully before the error was detected. In the past, the only way around this problem was to process each row individually, preferably with a bulk operation usingFORALLand theSAVE EXCEPTIONSclause. In Oracle 10g Database Release 2, the DML error logging feature has been introduced to solve this problem. Adding the appropriateLOG ERRORSclause on to most INSERT, UPDATE, MERGE and DELETE statements enables the operations to complete, regardless of errors. This article presents an overview of the DML error logging functionality, with examples of each type of DML statement.
Syntax
Restrictions
Sample Schema
Insert
Update
Merge
Delete
Performance
Syntax
The syntax for the error logging clause is the same for INSERT, UPDATE, MERGE and DELETE statements.
LOG ERRORS [INTO [schema.]table] [('simple_expression')] [REJECT LIMIT integer|UNLIMITED]
The optionalINTOclause allows you to specify the name of the error logging table. If you omit this clause, the the first 25 characters of the base table name are used along with the "ERR$_" prefix.
Thesimple_expressionis used to specify a tag that makes the errors easier to identify. This might be a string or any function whose result is converted to a string.
TheREJECT LIMITis used to specify the maximum number of errors before the statement fails. The default value is 0 and the maximum values is the keywordUNLIMITED. For parallel DML operations, the reject limit is applied to each parallel server.
Restrictions
The DML error logging functionality is not invoked when:
Deferred constraints are violated.
Direct-path INSERT or MERGE operations raise unique constraint or index violations.
UPDATE or MERGE operations raise a unique constraint or index violation.
In addition, the tracking of errors in LONG, LOB and object types is not supported, although a table containing these columns can be the target of error logging.
Sample Schema
This following code creates and populates the tables necessary to run the example code in this article.
-- Create and populate a source table.
CREATE TABLE source (
id NUMBER(10) NOT NULL,
code VARCHAR2(10),
description VARCHAR2(50),
CONSTRAINT source_pk PRIMARY KEY (id)
);
DECLARE
TYPE t_tab IS TABLE OF source%ROWTYPE;
l_tab t_tab := t_tab();
BEGIN
FOR i IN 1 .. 100000 LOOP
l_tab.extend;
l_tab(l_tab.last).id := i;
l_tab(l_tab.last).code := TO_CHAR(i);
l_tab(l_tab.last).description := 'Description for ' || TO_CHAR(i);
END LOOP;
-- For a possible error condition.
l_tab(1000).code := NULL;
l_tab(10000).code := NULL;
FORALL i IN l_tab.first .. l_tab.last
INSERT INTO source VALUES l_tab(i);
COMMIT;
END;
/
EXEC DBMS_STATS.gather_table_stats(USER, 'source', cascade => TRUE);
-- Create a destination table.
CREATE TABLE dest (
id NUMBER(10) NOT NULL,
code VARCHAR2(10) NOT NULL,
description VARCHAR2(50),
CONSTRAINT dest_pk PRIMARY KEY (id)
);
-- Create a dependant of the destination table.
CREATE TABLE dest_child (
id NUMBER,
dest_id NUMBER,
CONSTRAINT child_pk PRIMARY KEY (id),
CONSTRAINT dest_child_dest_fk FOREIGN KEY (dest_id) REFERENCES dest(id)
);
Notice that theCODEcolumn is optional in theSOURCEtable and mandatory in theDESTtable.
Once the basic tables are in place we can create a table to hold the DML error logs for theDEST. A log table must be created for every base table that requires the DML error logging functionality. This can be done manually or with theCREATE_ERROR_LOGprocedure in theDBMS_ERRLOGpackage, as shown below.
-- Create the error logging table.
BEGIN
DBMS_ERRLOG.create_error_log (dml_table_name => 'dest');
END;
/
PL/SQL procedure successfully completed.
SQL>
The owner, name and tablespace of the log table can be specified, but by default it is created in the current schema, in the default tablespace with a name that matches the first 25 characters of the base table with the "ERR$_" prefix.
SELECT owner, table_name, tablespace_name
FROM all_tables
WHERE owner = 'TEST';
OWNER TABLE_NAME TABLESPACE_NAME
------------------------------ ------------------------------ ------------------------------
TEST DEST USERS
TEST DEST_CHILD USERS
TEST ERR$_DEST USERS
TEST SOURCE USERS
4 rows selected.
SQL>
The structure of the log table includes maximum length and datatype independent versions of all available columns from the base table, as seen below.
SQL> DESC err$_dest
Name Null? Type
--------------------------------- -------- --------------
ORA_ERR_NUMBER$ NUMBER
ORA_ERR_MESG$ VARCHAR2(2000)
ORA_ERR_ROWID$ ROWID
ORA_ERR_OPTYP$ VARCHAR2(2)
ORA_ERR_TAG$ VARCHAR2(2000)
ID VARCHAR2(4000)
CODE VARCHAR2(4000)
DESCRIPTION VARCHAR2(4000)
SQL>
Insert
When we built the sample schema we noted that theCODEcolumn is optional in theSOURCEtable, but mandatory in thDESTtable. When we populated theSOURCEtable we set the code to NULL for two of the rows. If we try to copy the data from theSOURCEtable to theDESTtable we get the following result.
INSERT INTO dest
SELECT *
FROM source;
SELECT *
*
ERROR at line 2:
ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
SQL>
The failure causes the whole insert to roll back, regardless of how many rows were inserted successfully. Adding the DML error logging clause allows us to complete the insert of the valid rows.
INSERT INTO dest
SELECT *
FROM source
LOG ERRORS INTO err$_dest ('INSERT') REJECT LIMIT UNLIMITED;
99998 rows created.
SQL>
The rows that failed during the insert are stored in theERR$_DESTtable, along with the reason for the failure.
COLUMN ora_err_mesg$ FORMAT A70
SELECT ora_err_number$, ora_err_mesg$
FROM err$_dest
WHERE ora_err_tag$ = 'INSERT';
ORA_ERR_NUMBER$ ORA_ERR_MESG$
--------------- ---------------------------------------------------------
1400 ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
1400 ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
2 rows selected.
SQL>
Update
The following code attempts to update theCODEcolumn for 10 rows, setting it to itself for 8 rows and to the value NULL for 2 rows.
UPDATE dest
SET code = DECODE(id, 9, NULL, 10, NULL, code)
WHERE id BETWEEN 1 AND 10;
*
ERROR at line 2:
ORA-01407: cannot update ("TEST"."DEST"."CODE") to NULL
SQL>
As expected, the statement fails because theCODEcolumn is mandatory. Adding the DML error logging clause allows us to complete the update of the valid rows.
UPDATE dest
SET code = DECODE(id, 9, NULL, 10, NULL, code)
WHERE id BETWEEN 1 AND 10
LOG ERRORS INTO err$_dest ('UPDATE') REJECT LIMIT UNLIMITED;
8 rows updated.
SQL>
The rows that failed during the update are stored in theERR$_DESTtable, along with the reason for the failure.
COLUMN ora_err_mesg$ FORMAT A70
SELECT ora_err_number$, ora_err_mesg$
FROM err$_dest
WHERE ora_err_tag$ = 'UPDATE';
ORA_ERR_NUMBER$ ORA_ERR_MESG$
--------------- ---------------------------------------------------------
1400 ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
1400 ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
2 rows selected.
SQL>
Merge
The following code deletes some of the rows from theDESTtable, then attempts to merge the data from theSOURCEtable into theDESTtable.
DELETE FROM dest
WHERE id > 50000;
MERGE INTO dest a
USING source b
ON (a.id = b.id)
WHEN MATCHED THEN
UPDATE SET a.code = b.code,
a.description = b.description
WHEN NOT MATCHED THEN
INSERT (id, code, description)
VALUES (b.id, b.code, b.description);
*
ERROR at line 9:
ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
SQL>
As expected, the merge operation fails and rolls back. Adding the DML error logging clause allows the merge operation to complete.
MERGE INTO dest a
USING source b
ON (a.id = b.id)
WHEN MATCHED THEN
UPDATE SET a.code = b.code,
a.description = b.description
WHEN NOT MATCHED THEN
INSERT (id, code, description)
VALUES (b.id, b.code, b.description)
LOG ERRORS INTO err$_dest ('MERGE') REJECT LIMIT UNLIMITED;
99998 rows merged.
SQL>
The rows that failed during the update are stored in theERR$_DESTtable, along with the reason for the failure.
COLUMN ora_err_mesg$ FORMAT A70
SELECT ora_err_number$, ora_err_mesg$
FROM err$_dest
WHERE ora_err_tag$ = 'MERGE';
ORA_ERR_NUMBER$ ORA_ERR_MESG$
--------------- ---------------------------------------------------------
1400 ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
1400 ORA-01400: cannot insert NULL into ("TEST"."DEST"."CODE")
2 rows selected.
SQL>
Delete
TheDEST_CHILDtable has a foreign key to theDESTtable, so if we add some data to it would would expect an error if we tried to delete the parent rows from theDESTtable.
INSERT INTO dest_child (id, dest_id) VALUES (1, 100);
INSERT INTO dest_child (id, dest_id) VALUES (2, 101);
With the child data in place we ca attempt to delete th data from theDESTtable.
DELETE FROM dest;
*
ERROR at line 1:
ORA-02292: integrity constraint (TEST.DEST_CHILD_DEST_FK) violated - child record found
SQL>
As expected, the delete operation fails. Adding the DML error logging clause allows the delete operation to complete.
DELETE FROM dest
LOG ERRORS INTO err$_dest ('DELETE') REJECT LIMIT UNLIMITED;
99996 rows deleted.
SQL>
The rows that failed during the delete operation are stored in theERR$_DESTtable, along with the reason for the failure.
COLUMN ora_err_mesg$ FORMAT A69
SELECT ora_err_number$, ora_err_mesg$
FROM err$_dest
WHERE ora_err_tag$ = 'DELETE';
ORA_ERR_NUMBER$ ORA_ERR_MESG$
--------------- ---------------------------------------------------------------------
2292 ORA-02292: integrity constraint (TEST.DEST_CHILD_DEST_FK) violated -
child record found
2292 ORA-02292: integrity constraint (TEST.DEST_CHILD_DEST_FK) violated -
child record found
2 rows selected.
SQL>
Performance
The performance of DML error logging depends on the way it is being used and what version of the database you use it against. Prior to Oracle 12c, you will probably only use DML error logging during direct path loads, since conventional path loads become very slow when using it. The following example displays this, but before we start we will need to remove the extra dependency table.
DROP TABLE dest_child PURGE;
Truncate the destination table and run a conventional path load using DML error logging, using SQL*Plus timing to measure the elapsed time.
SET TIMING ON
TRUNCATE TABLE dest;
INSERT INTO dest
SELECT *
FROM source
LOG ERRORS INTO err$_dest ('INSERT NO-APPEND') REJECT LIMIT UNLIMITED;
99998 rows created.
Elapsed: 00:00:08.61
SQL>
Next, repeat the test using a direct path load this time.
TRUNCATE TABLE dest;
INSERT /*+ APPEND */ INTO dest
SELECT *
FROM source
LOG ERRORS INTO err$_dest ('INSERT APPEND') REJECT LIMIT UNLIMITED;
99998 rows created.
Elapsed: 00:00:00.38
SQL>
Finally, perform the same load usingFORALL ... SAVE EXCEPTIONSmethod.
TRUNCATE TABLE dest;
DECLARE
TYPE t_tab IS TABLE OF dest%ROWTYPE;
l_tab t_tab;
l_start PLS_INTEGER;
CURSOR c_source IS
SELECT * FROM source;
ex_dml_errors EXCEPTION;
PRAGMA EXCEPTION_INIT(ex_dml_errors, -24381);
BEGIN
OPEN c_source;
LOOP
FETCH c_source
BULK COLLECT INTO l_tab LIMIT 1000;
EXIT WHEN l_tab.count = 0;
BEGIN
FORALL i IN l_tab.first .. l_tab.last SAVE EXCEPTIONS
INSERT INTO dest VALUES l_tab(i);
EXCEPTION
WHEN ex_dml_errors THEN
NULL;
END;
END LOOP;
CLOSE c_source;
END;
/
PL/SQL procedure successfully completed.
Elapsed: 00:00:01.01
SQL>
From this we can see that DML error logging is very fast for direct path loads, but does not perform well for conventional path loads. In fact, it performs significantly worse than theFORALL ... SAVE EXCEPTIONSmethod.
The relative performance of these methods depends on the database version. The following table shows the results of the previous tests against a number of database versions. They are run on different servers, so don't compare version-to-version. Look at the comparison between the methods within a version.
10.2.0.4 11.2.0.3 11.2.0.4 12.1.0.1
======== ======== ======== ========
DML Error Logging : 07.62 08.61 04.82 00.94
DML Error Logging (APPEND) : 00.86 00.38 00.85 01.07
FORALL ... SAVE EXCEPTIONS : 01.15 01.01 00.94 01.37
同样的方式不同数据库版本性能可能不同,同版本数据库下,使用直接路径加