如何刪除UNDO表空間并處理ORA-01548問題

In this Document

APPLIES TO:

PURPOSE

This troubleshooting guide is for resolving ORA-1548 error reported while dropping UNDO tablespace.

The main purpose is to provide the Database Administrators an understanding of the issue and steps to resolve the same. It also provides you the diagnostic information to be collected before raising a Service Request with Oracle Support.

TROUBLESHOOTING STEPS

What is ORA-1548

ORA-1548 is the error reported when you try to drop a tablespace that contains active rollback segments ie, the segments which will be required for a rollback or when a dead transaction is present which requires a transaction recovery. The tablespace can be dropped only after clearing the active (or dead) transaction accessing the Undo Segment. The issue could also happen if the transaction cannot be rolled back due to various reasons like Undo datafile is offline or not accessible, Undo requires recovery and so on.

You can check for the Undo Segment in use, using the query:

select segment_id, segment_name,status,tablespace_name from dba_rollback_segs where status not in ('ONLINE','OFFLINE');

If the above query returns 'ONLINE' or 'PARTLY AVAILABLE', it means that there are still pending transaction entries in the rollback. You must wait until the corresponding transactions are either committed or rolled back. Repeat the above query again until the status becomes 'OFFLINE'. The status means that a dead transaction exists that still needs to be rolled back. There is no way we can simply forget about it, it would leave logical corruptions in the database.</span></span>

A 'NEEDS RECOVERY' status means that there are problems with the rollback. See Case 3 in this document

 Case 1 : An active or dead transaction present in the Undo Segment.

1). Check the status of the undo segments that are not Online or Offline:

Select segment_name,status,tablespace_name from dba_rollback_segs where status not in ('ONLINE','OFFLINE') ;

If there are segments with status 'Partly Available' , it means they still have active transactions pending and you can not drop the tablespace until the transaction is committed or rolled back. In case of a dead transaction you will have to wait till the recovery is done.

2. To check for dead transactions:

 select KTUXEUSN, KTUXESLT, KTUXESQN, KTUXESTA, KTUXECFL, KTUXESIZ
 from x$ktuxe
 where KTUXESTA='ACTIVE' and KTUXECFL='DEAD';

  You can monitor the recovery using the column KTUXESIZ . This will decrease as the recovery progresses.

Once the transaction is recovered, you can try dropping the Undo tablespace

Case 2: Undo segment held by a dead distributed transaction

In this case , the issue is caused by dead distributed transactions. We have to clear the dead distributed transactions before trying to drop the Undo tablespace.

To check for any active transactions on a rollback segment:
  

SELECT KTUXEUSN, KTUXESLT, KTUXESQN, /* Transaction ID */
KTUXESTA Status,
KTUXECFL Flags
FROM x$ktuxe
WHERE ktuxesta!='INACTIVE';

 If the status shows as 'Prepared', It implies this is a distributed transaction , which should be committed or rolled back.

b. You can use the following queries to check for in doubt transactions:
  

SELECT LOCAL_TRAN_ID,GLOBAL_TRAN_ID,STATE,MIXED,COMMIT# FROM DBA_2PC_PENDING;

SELECT LOCAL_TRAN_ID,IN_OUT,DATABASE,INTERFACE FROM dba_2pc_neighbors;

At this point you have to cleanup Distributed transaction without corresponding dba_2pc entries

Please Contact Oracle Support to clean the stranded dba_2pc transactions

 Once these transactions are cleared, you should be able to drop the undo tablespace without encountering the ORA-1548 error.

Case 3: When Undo Segment requires recovery

 Check the status of the undo segments

Select segment_name,status,tablespace_name from dba_rollback_segs where status not in ('ONLINE','OFFLINE') ;

If the Undo Segment status shows as 'Needs Recovery' Please refer the document to recover the Undo segment
Unable to Drop Undo tablespace Since Undo Segment is in Needs Recovery ( Doc ID 1295294.1 )

Case 4: If Event 10513 is set.

 Check the alert log file if the event 10513 is set.

event="10513 trace name context forever, level 2"

  This can be set dynamically as:

SQL> oradebug setorapid <SMON's Oracle PID>
SQL> oradebug event 10513 trace name context forever, level 2

 If event 10513 is set, transaction recovery from SMON will be disabled and hence the Undo Segment status wont change. So, if the event is set, restart the database after removing the event and monitor the recovery of the dead transactions.

 Remove the event from the init.ora and restart the instance. If it is set dynamically you can turn it off as :

SQL> oradebug setorapid <SMON's Oracle PID>
SQL> oradebug event 10513 trace name context off

 Once the event is reset, you can monitor the trasaction recovery progress using:

select KTUXEUSN, KTUXESLT, KTUXESQN, KTUXESTA, KTUXECFL, KTUXESIZ
 from x$ktuxe
 where KTUXESTA='ACTIVE' and KTUXECFL='DEAD';

 You can monitor the recovery using the column KTUXESIZ . This will decrease as the recovery progresses.

Case 5: When FlashBack Data Archiver (FBDA) is enabled

 Undo tablespace drop operation is blocked since some transaction is marked to have to be archived for flashback.

 1. Check current_scn from v$database after updating tracked table

2. Do not change undo_tablespace parameter

3. select barrierscn from sys_fba_barrierscn;

wait until this scn is greater than scn at step 1

4. After barrierscn become greater, wait one more minute so that FBDA can update transaction table extension

5. Then change undo_tablespace, and drop old one.

For step 3, FBDA will update sys_fba_barrierscn after a while. If you don't want to wait, fastest way is restarting instance.

Case 6: FlashBack Data Archiver (FBDA) was enabled in the past.

In this case also, the Undo marked to have to be archived for flashback. The extents used by transactions with the FBA bit "on" are considered "active" until the archiver has mined the undo and recorded the history for the flashback archive enabled tables as these extents cannot be recycled until the history is recorded.

If the instance parameter "_disable_flashback_archiver" is set, the FBDA processs wont be starting and hence wont process these undo extents.

To verify:

SQL> SELECT a.ksppinm "Parameter",
            b.ksppstvl "Session Value",
            c.ksppstvl "Instance Value"
     FROM sys.x$ksppi a, sys.x$ksppcv b, sys.x$ksppsv c
     WHERE a.indx = b.indx
     AND a.indx = c.indx
     AND a.ksppinm LIKE '%flashback%';

This must be reset and the instance must be restarted.  

SQL> ALTER SYSTEM SET "_disable_flashback_archiver"=0 SCOPE=SPFILE;

 
 and restart the instance. This will return the parameter to its default value.