Skip Headers
Oracle® Database Application Developer's Guide - Fundamentals
10g Release 2 (10.2)

Part Number B14251-01
Go to Documentation Home
Home
Go to Book List
Book List
Go to Table of Contents
Contents
Go to Index
Index
Go to Master Index
Master Index
Go to Feedback page
Contact Us

Go to previous page
Previous
Go to next page
Next
View PDF

10 Developing Flashback Applications

This chapter discusses the following flashback topics:

Overview of Flashback Features

Oracle Database has a group of features, known collectively as flashback, that provide ways to view past states of database objects or to return database objects to a previous state without using point-in-time media recovery.

You can use flashback features of the database to do the following:

Flashback features use the Automatic Undo Management system to obtain metadata and historical data for transactions. They rely on undo data, which are records of the effects of individual transactions. For example, if a user executes an UPDATE statement to change a salary from 1000 to 1100, then Oracle would store the value 1000 in the undo data.

Undo data is persistent and survives a database shutdown. By using flashback features, you can employ undo data to query past data or recover from logical corruptions. Besides using it in flashback operations, Oracle Database uses undo data to perform the following actions:

Application Development Features

In application development, you can use flashback features to report on historical data or undo erroneous changes. Flashback features include the following:

  • Oracle Flashback Query

    You can use this feature to retrieve data for a time in the past that you specify using the AS OF clause of the SELECT statement.

  • Oracle Flashback Version Query

    You can use this feature to retrieve metadata and historical data for a specific time interval. For example, you can view all the rows of a table that ever existed during a given time interval. Metadata about the different versions of rows includes start and end time, type of change operation, and identity of the transaction that created the row version. You use the VERSIONS BETWEEN clause of the SELECT statement to create a Flashback Version Query.

  • Oracle Flashback Transaction Query

    You can use this feature to retrieve metadata and historical data for a given transaction or for all transactions within a given time interval. You can also obtain the SQL code to undo the changes to particular rows affected by a transaction. Typically, you use Flashback Transaction Query in conjunction with a Flashback Version Query that provides the transaction IDs for the rows of interest. To perform a Flashback Transaction Query, select from the FLASHBACK_TRANSACTION_QUERY view.

  • DBMS_FLASHBACK package

    You can use this feature to set the internal Oracle clock to a time in the past so that you can examine data current at that time.

Database Administration Features

You can use the following features for application development or interactively as a database user or administrator:

  • DBMS_FLASHBACK package

  • Flashback Query

  • Flashback Version Query

  • Flashback Transaction Query

Typically, you use the following flashback features only in database administration:

  • Oracle Flashback Table

    You can use this feature to recover a table to its state at a previous point in time. You can restore table data while the database is on line, undoing changes to only the specified table.

  • Oracle Flashback Drop

    You can use this feature to recover a dropped table. This reverses the effects of a DROP TABLE statement.

  • Oracle Flashback Database

    You can use this feature to quickly return the database to an earlier point in time, by undoing all of the changes that have taken place since then. This is fast, because you do not have to restore database backups.

Flashback Database, Flashback Table, and Flashback Drop are primarily data recovery mechanisms and are therefore documented elsewhere. The other flashback features, while valuable in data recovery scenarios, are useful for application development. They are the focus of this chapter.

See Also:

Database Administration Tasks Before Using Flashback Features

Before you can use flashback features in your application, you must perform the following administrative tasks to configure your database. Consult with your database administrator to perform these tasks:

Using Flashback Query (SELECT ... AS OF)

You perform a Flashback Query by using a SELECT statement with an AS OF clause. You use a Flashback Query to retrieve data as it existed at some time in the past. The query explicitly references a past time by means of a timestamp or SCN. It returns committed data that was current at that point in time.

Potential uses of Flashback Query include:

Examining Past Data: Example

This example uses a Flashback Query to examine the state of a table at a previous time. Suppose that a DBA discovers at 12:30 PM that the row for employee Chung had been deleted from the employees table. The DBA also knows that at 9:30AM the data for Chung was correctly stored in the database. The DBA can use a Flashback Query to examine the contents of the table at 9:30 to find out what data had been lost. If appropriate, the DBA can then re-insert the lost data.

Example 10-1 retrieves the state of the record for Chung at 9:30AM, April 4, 2004:

Example 10-1 Retrieving a Row with Flashback Query

SELECT * FROM employees AS OF TIMESTAMP 
   TO_TIMESTAMP('2004-04-04 09:30:00', 'YYYY-MM-DD HH:MI:SS')
   WHERE last_name = 'Chung';

The update in Example 10-2 restores Chung's information to the employees table:

Example 10-2 Reinserting a Row After a Flashback Query

INSERT INTO employees 
    (SELECT * FROM employees AS OF TIMESTAMP 
     TO_TIMESTAMP('2004-04-04 09:30:00', 'YYYY-MM-DD HH:MI:SS')
     WHERE last_name = 'Chung');

Tips for Using Flashback Query

Keep the following in mind when using a Flashback Query (SELECT ... AS OF):

  • You can specify or omit the AS OF clause for each table and specify different times for different tables. Use an AS OF clause in a query to perform DDL operations (such as creating and truncating tables) or DML operations (such as inserting and deleting) in the same session as the query.

  • To use the results of a Flashback Query in a DDL or DML statement that affects the current state of the database, use an AS OF clause inside an INSERT or CREATE TABLE AS SELECT statement.

  • When choosing whether to use a timestamp or an SCN in Flashback Query, remember that Oracle Database uses SCNs internally and maps these to timestamps at a granularity of 3 seconds. If a possible 3-second error (maximum) is important to a Flashback Query in your application, then use an SCN instead of a timestamp. Refer to "Flashback Tips – General".

  • You can create a view that refers to past data by using the AS OF clause in the SELECT statement that defines the view. If you specify a relative time by subtracting from the current time on the database host, then the past time is recalculated for each query. For example:

    CREATE VIEW hour_ago AS
      SELECT * FROM employees AS OF
        TIMESTAMP (SYSTIMESTAMP - INTERVAL '60' MINUTE);
    -- SYSTIMESTAMP refers to the time zone of the database host environment
    
    
  • You can use the AS OF clause in self-joins, or in set operations such as INTERSECT and MINUS, to extract or compare data from two different times. You can store the results by preceding a Flashback Query with a CREATE TABLE AS SELECT or INSERT INTO TABLE SELECT statement. For example, the following query reinserts into table employees the rows that existed an hour ago:

    INSERT INTO employees 
      (SELECT * FROM employees AS OF 
         TIMESTAMP (SYSTIMESTAMP - INTERVAL '60' MINUTE))
    -- SYSTIMESTAMP refers to the time zone of the database host environment
      MINUS SELECT * FROM employees);
    

Using the DBMS_FLASHBACK Package

In general, the DBMS_FLASHBACK package provides the same functionality as Flashback Query, but Flashback Query is sometimes more convenient.

The DBMS_FLASHBACK package acts as a time machine: you can turn back the clock, carry out normal queries as if you were at that time in the past, and then return to the present. Because you can use the DBMS_FLASHBACK package to perform queries on past data without special clauses such as AS OF or VERSIONS BETWEEN, you can reuse existing PL/SQL code to query the database at times in the past.

You must have the EXECUTE privilege on the DBMS_FLASHBACK package.

To use the DBMS_FLASHBACK package in your PL/SQL code:

  1. Call DBMS_FLASHBACK.ENABLE_AT_TIME or DBMS_FLASHBACK.ENABLE_AT_SYSTEM_CHANGE_NUMBER to turn back the clock to a specified time in the past. Afterwards all queries retrieve data that was current at the specified time.

  2. Perform normal queries, that is, without any special flashback-feature syntax such as AS OF. The database is automatically queried at the specified past time. Perform only queries; do not try to perform DDL or DML operations.

  3. Call DBMS_FLASHBACK.DISABLE to return to the present. You must call DISABLE before calling ENABLE again for a different time. You cannot nest ENABLE /DISABLE pairs.

You can use a cursor to store the results of queries. To do this, open the cursor before calling DBMS_FLASHBACK.DISABLE. After storing the results and then calling DISABLE, you can do the following:

You can call DBMS_FLASHBACK.GET_SYSTEM_CHANGE_NUMBER at any time to obtain the current System Change Number (SCN). Note that the current SCN is always returned; this takes no account of previous calls to DBMS_FLASHBACK.ENABLE*.

See Also:

Using ORA_ROWSCN

ORA_ROWSCN is a pseudocolumn of any table that is not fixed or external. It represents the SCN of the most recent change to a given row, that is, the latest COMMIT operation for the row. For example:

SELECT ora_rowscn, last_name, salary 
FROM employees 
WHERE employee_id = 7788;

ORA_ROWSCN    NAME    SALARY
----------    ----    ------
    202553    Fudd      3000

The latest COMMIT operation for the row took place at approximately SCN 202553. You can use function SCN_TO_TIMESTAMP to convert an SCN, like ORA_ROWSCN, to the corresponding TIMESTAMP value.

ORA_SCN is in fact a conservative upper bound of the latest commit time: the actual commit SCN can be somewhat earlier. ORA_SCN is more precise (closer to the actual commit SCN) for a row-dependent table (created using CREATE TABLE with the ROWDEPENDENCIES clause).

Noteworthy uses of ORA_ROWSCN in application development include concurrency control and client cache invalidation. To see how you might use it in concurrency control, consider the following scenario.

Your application examines a row of data, and records the corresponding ORA_ROWSCN as 202553. Later, the application needs to update the row, but only if its record of the data is still accurate. That is, this particular update operation depends, logically, on the row not having been changed. The operation is therefore made conditional on the ORA_ROWSCN being still 202553. Here is an equivalent interactive command:

UPDATE employees 
  SET salary = salary + 100 
  WHERE employee_id = 7788 
  AND ora_rowscn = 202553;

0 rows updated.

The conditional update fails in this case, because the ORA_ROWSCN is no longer 202553. This means that some user or another application changed the row and performed a COMMIT more recently than the recorded ORA_ROWSCN.

Your application queries again to obtain the new row data and ORA_ROWSCN. Suppose that the ORA_ROWSCN is now 415639. The application tries the conditional update again, using the new ORA_ROWSCN. This time, the update succeeds, and it is committed. Here is an interactive equivalent:

SQL> UPDATE employees SET salary = salary + 100 
     WHERE empno = 7788 AND ora_rowscn = 415639;

1 row updated.

SQL> COMMIT;

Commit complete.

SQL> SELECT ora_rowscn, name, salary FROM employees WHERE empno = 7788;

ORA_ROWSCN    NAME    SALARY
----------    ----    ------
    465461    Fudd      3100

The SCN corresponding to the new COMMIT is 465461.

Besides using ORA_ROWSCN in an UPDATE statement WHERE clause, you can use it in a DELETE statement WHERE clause or the AS OF clause of a Flashback Query.

Using Flashback Version Query

You use a Flashback Version Query to retrieve the different versions of specific rows that existed during a given time interval. A new row version is created whenever a COMMIT statement is executed.

You specify a Flashback Version Query using the VERSIONS BETWEEN clause of the SELECT statement. Here is the syntax:

VERSIONS {BETWEEN {SCN | TIMESTAMP} start AND end}

where start and end are expressions representing the start and end of the time interval to be queried, respectively. The interval is closed at both ends: the upper and lower limits specified (start and end) are both included in the time interval.

The Flashback Version Query returns a table with a row for each version of the row that existed at any time during the time interval you specify. Each row in the table includes pseudocolumns of metadata about the row version, described in Table 10-1. This information can reveal when and how a particular change (perhaps erroneous) occurred to your database.

Table 10-1 Flashback Version Query Row Data Pseudocolumns

Pseudocolumn Name Description

VERSIONS_STARTSCN

VERSIONS_STARTTIME

Starting System Change Number (SCN) or TIMESTAMP when the row version was created. This identifies the time when the data first took on the values reflected in the row version. You can use this to identify the past target time for a Flashback Table or Flashback Query operation.

If this is NULL, then the row version was created before the lower time bound of the query BETWEEN clause.

VERSIONS_ENDSCN

VERSIONS_ENDTIME

SCN or TIMESTAMP when the row version expired. This identifies the row expiration time.

If this is NULL, then either the row version was still current at the time of the query or the row corresponds to a DELETE operation.

VERSIONS_XID

Identifier of the transaction that created the row version.

VERSIONS_OPERATION

Operation performed by the transaction: I for insertion, D for deletion, or U for update. The version is that of the row that was inserted, deleted, or updated; that is, the row after an INSERT operation, the row before a DELETE operation, or the row affected by an UPDATE operation.

Note: For user updates of an index key, a Flashback Version Query may treat an UPDATE operation as two operations, DELETE plus INSERT, represented as two version rows with a D followed by an I VERSIONS_OPERATION.


A given row version is valid starting at its time VERSIONS_START* up to, but not including, its time VERSIONS_END*. That is, it is valid for any time t such that VERSIONS_START* <= t < VERSIONS_END*. For example, the following output indicates that the salary was 10243 from September 9, 2002, included, to November 25, 2003, not included.

VERSIONS_START_TIME     VERSIONS_END_TIME     SALARY
-------------------     -----------------     ------
09-SEP-2003             25-NOV-2003           10243

Here is a typical Flashback Version Query:

SELECT versions_startscn, versions_starttime, 
       versions_endscn, versions_endtime,
       versions_xid, versions_operation,
       name, salary  
  FROM employees 
  VERSIONS BETWEEN TIMESTAMP 
      TO_TIMESTAMP('2003-07-18 14:00:00', 'YYYY-MM-DD HH24:MI:SS')
  AND TO_TIMESTAMP('2003-07-18 17:00:00', 'YYYY-MM-DD HH24:MI:SS')
  WHERE name = 'JOE';

Pseudocolumn VERSIONS_XID provides a unique identifier for the transaction that put the data in that state. You can use this value in connection with a Flashback Transaction Query to locate metadata about this transaction in the FLASHBACK_TRANSACTION_QUERY view, including the SQL required to undo the row change and the user responsible for the change – see "Using Flashback Transaction Query".

See Also:

Oracle Database SQL Reference for information on the Flashback Version Query pseudocolumns and the syntax of the VERSIONS clause

Using Flashback Transaction Query

A Flashback Transaction Query is a query on the view FLASHBACK_TRANSACTION_QUERY. You use a Flashback Transaction Query to obtain transaction information, including SQL code that you can use to undo each of the changes made by the transaction.

See Also:

Oracle Database Backup and Recovery Advanced User's Guide. and Oracle Database Administrator's Guide for information on how a DBA can use the Flashback Table feature to restore an entire table, rather than individual rows

As an example, the following statement queries the FLASHBACK_TRANSACTION_QUERY view for transaction information, including the transaction ID, the operation, the operation start and end SCNs, the user responsible for the operation, and the SQL code to undo the operation:

SELECT xid, operation, start_scn,commit_scn, logon_user, undo_sql
     FROM flashback_transaction_query
     WHERE xid = HEXTORAW('000200030000002D');

As another example, the following query uses a Flashback Version Query as a subquery to associate each row version with the LOGON_USER responsible for the row data change.

SELECT xid, logon_user FROM flashback_transaction_query
     WHERE xid IN (SELECT versions_xid FROM employees VERSIONS BETWEEN TIMESTAMP 
      TO_TIMESTAMP('2003-07-18 14:00:00', 'YYYY-MM-DD HH24:MI:SS') AND
      TO_TIMESTAMP('2003-07-18 17:00:00', 'YYYY-MM-DD HH24:MI:SS'));

Flashback Transaction Query and Flashback Version Query: Example

This example demonstrates the use of a Flashback Transaction Query in conjunction with a Flashback Version Query. The example assumes simple variations of the employees and departments tables in the sample hr schema.

In this example, a DBA carries out the following series of actions in SQL*Plus:

connect hr/hr
CREATE TABLE emp 
   (empno   NUMBER PRIMARY KEY, 
    empname VARCHAR2(16), 
    salary  NUMBER);
INSERT INTO emp VALUES (111, 'Mike', 555);
COMMIT;

CREATE TABLE dept 
   (deptno   NUMBER, 
    deptname VARCHAR2(32));
INSERT INTO dept VALUES (10, 'Accounting');
COMMIT;

At this point, emp and dept have one row each. In terms of row versions, each table has one version of one row. Next, suppose that an erroneous transaction deletes employee id 111 from table emp:

UPDATE emp SET salary = salary + 100 WHERE empno = 111;
INSERT INTO dept VALUES (20, 'Finance'); 
DELETE FROM emp WHERE empno = 111;
COMMIT;

Subsequently, a new transaction reinserts employee id 111 with a new employee name into the emp table.

INSERT INTO emp VALUES (111, 'Tom', 777);
UPDATE emp SET salary = salary + 100 WHERE empno = 111;
UPDATE emp SET salary = salary + 50 WHERE empno = 111;
COMMIT;

At this point, the DBA detects the application error and needs to diagnose the problem. The DBA issues the following query to retrieve versions of the rows in the emp table that correspond to empno 111. The query uses Flashback Version Query pseudocolumns.

connect dba_name/password
SELECT versions_xid XID, versions_startscn START_SCN,
  versions_endscn END_SCN, versions_operation OPERATION,
  empname, salary FROM hr.emp
  VERSIONS BETWEEN SCN MINVALUE AND MAXVALUE
  where empno = 111;

XID              START_SCN  END_SCN   OPERATION  EMPNAME    SALARY
---------------- ---------- --------- ---------- ---------- ----------
0004000700000058 113855               I          Tom        927
000200030000002D 113564               D          Mike       555
000200030000002E 112670     113564    I          Mike       555
3 rows selected

The results table reads chronologically, from bottom to top. The third row corresponds to the version of the row in emp that was originally inserted in the table when the table was created. The second row corresponds to the row in emp that was deleted by the erroneous transaction. The first row corresponds to the version of the row in emp that was reinserted with a new employee name.

The DBA identifies transaction 000200030000002D as the erroneous transaction and issues the following Flashback Transaction Query to audit all changes made by this transaction:

SELECT  xid, start_scn START, commit_scn COMMIT, 
        operation OP, logon_user USER, 
        undo_sql FROM flashback_transaction_query
        WHERE xid = HEXTORAW('000200030000002D');

XID               START   COMMIT  OP       USER   UNDO_SQL
----------------  -----   ------  --       ----   ---------------------------
000200030000002D  195243  195244  DELETE   HR     insert into "HR"."EMP" 
("EMPNO","EMPNAME","SALARY") values ('111','Mike','655');

000200030000002D  195243  195244  INSERT   HR     delete from "HR"."DEPT" 
where ROWID = 'AAAKD4AABAAAJ3BAAB';

000200030000002D  195243  195244  UPDATE   HR     update "HR"."EMP" 
set "SALARY" = '555' where ROWID = 'AAAKD2AABAAAJ29AAA';

000200030000002D  195243  113565  BEGIN  HR

4 rows selected

The rightmost column (undo_sql) contains the SQL code that will undo the corresponding change operation. The DBA can execute this code to undo the changes made by that transaction. The USER column (logon_user) shows the user responsible for the transaction.

A DBA might also be interested in knowing all changes made in a certain time window. In our scenario, the DBA performs the following query to view the details of all transactions that executed since the erroneous transaction identified earlier (including the erroneous transaction itself):

SELECT xid, start_scn, commit_scn, operation, table_name, table_owner
  FROM flashback_transaction_query
  WHERE table_owner = 'HR' AND
        start_timestamp >=
          TO_TIMESTAMP ('2002-04-16 11:00:00','YYYY-MM-DD HH:MI:SS');

XID               START_SCN  COMMIT_SCN  OPERATION  TABLE_NAME  TABLE_OWNER
----------------  ---------  ----------  ---------  ----------  -----------
0004000700000058  195245     195246      UPDATE     EMP         HR
0004000700000058  195245     195246      UPDATE     EMP         HR
0004000700000058  195245     195246      INSERT     EMP         HR
000200030000002D  195243     195244      DELETE     EMP         HR
000200030000002D  195243     195244      INSERT     DEPT        HR
000200030000002D  195243     195244      UPDATE     EMP         HR

6 rows selected

Flashback Tips

The following tips and restrictions apply to using flashback features.

Flashback Tips – Performance

  • For better performance, generate statistics on all tables involved in a Flashback Query by using the DBMS_STATS package, and keep the statistics current. Flashback Query always uses the cost-based optimizer, which relies on these statistics.

  • The performance of a query into the past depends on how much undo data must be accessed. For better performance, use queries to select small sets of past data using indexes, not to scan entire tables. If you must do a full table scan, consider adding a parallel hint to the query.

  • The performance cost in I/O is the cost of paging in data and undo blocks that are not already in the buffer cache. The performance cost in CPU use is the cost of applying undo information to affected data blocks. When operating on changes in the recent past, flashback features essentially CPU bound.

  • Use index structures for Flashback Version Query: the database keeps undo data for index changes as well as data changes. Performance of index lookup-based Flashback Version Query is an order of magnitude faster than the full table scans that are otherwise needed.

  • In a Flashback Transaction Query, the type of the xid column is RAW(8). To take advantage of the index built on the xid column, use the HEXTORAW conversion function: HEXTORAW(xid).

  • Flashback Query against a materialized view does not take advantage of query rewrite optimizations.

Flashback Tips – General

  • Use the DBMS_FLASHBACK package or other flashback features? Use ENABLE/DISABLE calls to the DBMS_FLASHBACK package around SQL code that you do not control, or when you want to use the same past time for several consecutive queries. Use Flashback Query, Flashback Version Query, or Flashback Transaction Query for SQL that you write, for convenience. A Flashback Query, for example, is flexible enough to do comparisons and store results in a single query.

  • To obtain an SCN to use later with a flashback feature, use DBMS_FLASHBACK.GET_SYSTEM_CHANGE_NUMBER.

  • You can compute or retrieve a past time to use in a query by using a function return value as a timestamp or SCN argument. For example, you can perform date and time calculations by adding or subtracting an INTERVAL value to the value of the SYSTIMESTAMP function.

  • You can query locally or remotely (Flashback Query, Flashback Version Query, or Flashback Transaction Query). for example here is a remote Flashback Query:

    (SELECT * FROM employees@some_remote_host AS OF 
            TIMESTAMP (SYSTIMESTAMP - INTERVAL '60' MINUTE);
    
    
  • To ensure database consistency, always perform a COMMIT or ROLLBACK operation before querying past data.

  • Remember that all flashback processing is done using the current session settings, such as national language and character set, not the settings that were in effect at the time being queried.

  • Some DDLs that alter the structure of a table, such as drop/modify column, move table, drop partition, and truncate table/partition, invalidate any existing undo data for the table. It is not possible to retrieve data from a point earlier than the time such DDLs were executed. Trying such a query results in error ORA-1466. This restriction does not apply to DDL operations that alter the storage attributes of a table, such as PCTFREE, INITRANS, and MAXTRANS.

  • Use an SCN to query past data at a precise time. If you use a timestamp, the actual time queried might be up to 3 seconds earlier than the time you specify. Internally, Oracle Database uses SCNs; these are mapped to timestamps at a granularity of every 3 seconds.

    For example, assume that the SCN values 1000 and 1005 are mapped to the times 8:41 and 8:46 AM respectively. A query for a time between 8:41:00 and 8:45:59 AM is mapped to SCN 1000; a Flashback Query for 8:46 AM is mapped to SCN 1005.

    Due to this time-to-SCN mapping, if you specify a time that is slightly after a DDL operation (such as a table creation) the database might actually use an SCN that is just before the DDL operation. This can result in error ORA-1466.

  • You cannot retrieve past data from a V$ view in the data dictionary. Performing a query on such a view always returns the current data. You can, however, perform queries on past data in other views of the data dictionary, such as USER_TABLES.