Skip Headers

Oracle9i User-Managed Backup and Recovery Guide
Release 2 (9.2)

Part Number A96572-01
Go To Documentation Library
Home
Go To Product List
Book List
Go To Table Of Contents
Contents
Go To Index
Index

Master Index

Feedback

Go to previous page Go to next page

5
Troubleshooting User-Managed Media Recovery

This chapter describes how to troubleshoot user-managed media recovery, and includes the following topics:

About User-Managed Media Recovery Problems

Table 5-1 describes potential problems that can occur during media recovery.

Table 5-1 Media Recovery Problems  
Problem Description

Missing or misnamed archived log

Recovery stops because Oracle cannot find the archived log recorded in the control file.

When you attempt to open the database, error ORA-1113 indicates that a file needs media recovery

This error commonly occurs because:

  • You are performing incomplete recovery but failed to restore all needed datafile backups.
  • Incomplete recovery stopped before datafiles reached a consistent SCN.
  • You are recovering datafiles from an online backup, but not enough redo was applied to make the datafiles consistent.
  • You are performing recovery with a backup control file, and did not specify the location of a needed online log.
  • A datafile is undergoing media recovery when you attempt to open the database.
  • Datafiles needing recovery were not brought online before executing RECOVER DATABASE, and so were not recovered.

Redo record problems

Two possible cases are as follows:

  • Recovery stops because of failed consistency checks, a problem called stuck recovery. Stuck recovery can occur when an underlying operating system or storage system loses a write issued by Oracle during normal operation of the database.
  • Oracle signals an internal error when applying the redo. This problem can be caused by an Oracle bug. If checksums are not being used, it can also be caused by corruptions to the redo or data blocks.

Corrupted archived logs

Logs may be corrupted while they are stored on or copied between storage systems. If DB_BLOCK_CHECKSUM is enabled, then Oracle usually signals checksum errors. If checksumming is not on, then log corruption may appear as a problem with redo.

Archived logs with incompatible parallel redo format

If you enable the parallel redo feature in Oracle9i Release 2 (9.2), then Oracle generates redo logs in a new format. Prior releases of Oracle are unable to apply parallel redo logs. However, Oracle9i Release 1 (9.0.1) can detect the parallel redo format and indicate the inconsistency with the following error message: External error 00303, 00000, "cannot process Parallel Redo".

See Also: Oracle9i Database Performance Tuning Guide and Reference to learn about the parallel redo feature

Corrupted data blocks

A datafile backup may have contained a corrupted data block, or the data block may become corrupted either during recovery or when it was copied to the backup. If checksums are being used, then Oracle signals a checksum error. Otherwise, the problem may also appear as a redo corruption.

Random problems

Memory corruptions and other transient problems can occur during recovery.

The symptoms of media recovery problems are usually external or internal errors signaled during recovery. For example, an external error indicates that a redo block or a data block has failed checksum verification checks. Internal errors can be caused by either bugs in Oracle or errors arising from the underlying operating system and hardware.

If media recovery encounters a problem while recovering a database backup, whether it is a stuck recovery problem or a problem during redo application, Oracle always stops and leaves the datafiles undergoing recovery in a consistent state, that is, at an SCN preceding the failure. You can then do one of the following:

In general, opening the database read-only or opening with the RESETLOGS option require all online datafiles to be recovered to the same SCN. If this requirement is not met, then Oracle may signal ORA-1113 or other errors when you attempt to open. Some common causes of ORA-1113 are described in Table 5-1.

The basic methodology for responding to media recovery problems occurs in the following phases:

  1. Try to identify the cause of the problem. Run a trial recovery if needed.
  2. If the problem is related to missing logs or you suspect there is a log, memory, or data block corruption, then try to resolve it using the methods described in Table 5-2.
  3. If you cannot resolve the problem using the methods described in Table 5-2, then do one of the following:
    • Open the database with the RESETLOGS option if you are recovering a whole database backup. If you have performed serial media recovery, then the database contains all the changes up to but not including the changes at the SCN where the corruption occurred. No changes from this SCN onward are in the recovered part of the database. If you have restored online backups, opening RESETLOGS succeeds only if you have recovered through all the ALTER ... END BACKUP operations in the redo stream.
    • Proceed with recovery by allowing media recovery to corrupt data blocks. After media recovery completes, try performing block media recovery using RMAN.
    • Call Oracle Support Services as a last resort.

      See Also:

      Oracle9i Recovery Manager User's Guide to learn about block media recovery

Investigating the Media Recovery Problem: Phase 1

If media recovery encounters a problem, then obtain as much information as possible after recovery halts. You do not want to waste time fixing the wrong problem, which may in fact make matters worse.

The goal of this initial investigation is to determine whether the problem is caused by incorrect setup, corrupted logs, corrupted data blocks, memory corruption, or other problems. If you see a checksum error on a data block, then the data block is corrupted. If you see a checksum error on a redo log block, then the redo log is corrupted.

Sometimes the cause of a recovery problem can be difficult to determine. Nevertheless, the methods in this chapter allow you to quickly recover a database sometimes even when you do not completely understand the cause of the problem.

To investigate media recovery problems:

  1. Examine the alert.log to see whether the error messages give general information about the nature of the problem. For example, does the alert_SID.log indicate any checksum failures? Does the alert_SID.log indicate that media recovery may have to corrupt data blocks in order to continue?
  2. Check the trace file generated by the Oracle process during recovery. It may contain additional error information.

Trying to Fix the Recovery Problem Without Corrupting Blocks: Phase 2

Depending on the type of media recovery problem you suspect, you have different solutions at your disposal. You can try one or a combination of the methods described in Table 5-2. Note that these methods are fairly safe: in almost all cases, they should not cause any damage to the database.

Table 5-2 Media Recovery Solutions  
If you suspect . . . Then . . .

Missing/misnamed archived logs

Determine whether you entered the correct filename. If you did, then check to see whether the log is missing from the operating system. If it is missing, and you have a backup, then restore the backup and apply the log. If you do not have a backup, then if possible perform incomplete recovery up to the point of the missing log.

ORA-1113 for ALTER DATABASE OPEN

Review the causes of this error in Table 5-1. Make sure that all read/write datafiles requiring recovery are online. If you use a backup control file for recovery, then the control file and datafiles must be at a consistent SCN for the database to be opened. If you do not have the necessary redo, then you must re-create the control file.

Corrupt archived logs

The log is corrupted if the checksum verification on the log redo block fails. If DB_BLOCK_CHECKSUM is not enabled either during the recovery session or when the database generated the redo, then recovery problems may be caused by corrupted logs. If the log is corrupt and an alternate copy of the corrupt log is available, then try to apply it and see whether this tactic fixes the problem.

The DB_BLOCK_CHECKSUM initialization parameter determines whether checksums are computed for redo log and data blocks.

Archived logs with incompatible parallel redo format

If you are running an Oracle release prior to Oracle9i Release 2, and if you are attempting to apply redo logs created with the parallel redo format, then you must do the following steps:

  1. Upgrade the database to Oracle9i Release 2.
  2. Perform media recovery.
  3. Shut down the database consistently and back up the database.
  4. Downgrade the database to the original release.

See Also: Oracle9i Database Performance Tuning Guide and Reference to learn about the parallel redo feature

Memory corruption or transient problems

Shut down the database and restart recovery. Sometimes this tactic fixes the problem. Oracle should leave the database in a consistent state if the second attempt also fails.

Corrupt data blocks

Restore and recover the datafile again with user-managed methods, or restore and recover individual data blocks with the RMAN BLOCKRECOVER command. This tactic may fix the problem.

Note that a data block is corrupted if the checksum verification on the data block fails. If DB_BLOCK_CHECKING is not enabled, a corrupted data block problem may appear as a redo problem. If you must proceed with recovery, then you may want to corrupt the block now and continue recovery, and use RMAN to perform block media recovery later.

If you cannot fix the problem with the methods described in Table 5-2, then there may be no easy way to fix the problem without losing data. You have these options:

Deciding Whether to Allow Recovery to Corrupt Blocks: Phase 3

When media recovery encounters a problem, the alert_SID.log may indicate that recovery can continue if it is allowed to corrupt the data block causing the problem. The alert_SID.log always contains information about the block: its block type, block address, the tablespace it belongs to, and so forth. For blocks containing user data, the log may also report the data object number.

In this case, Oracle can proceed with recovery if it is allowed to mark the problem block as corrupt. Nevertheless, this response is not always advisable. For example, if the block is an important block in the SYSTEM tablespace, marking the block as corrupt can eventually prevent you from opening the recovered database. Another consideration is whether the recovery problem is isolated. If this problem is followed immediately by many other problems in the redo stream, then you may want to open the database with the RESETLOGS option.

For a block containing user data, you can usually query the database to find out which object or table owns this block. If the database is not open, then you should be able to open the database read-only, even if you are recovering a whole database backup. The following example cancels recovery and opens read-only:

CANCEL
ALTER DATABASE OPEN READ ONLY;

Assume that the data object number reported in the alert_SID.log is 8031. You can determine the owner, object name, and object type by issuing this query:

SELECT OWNER, OBJECT_NAME, SUBOBJECT_NAME, OBJECT_TYPE
FROM DBA_OBJECTS
WHERE DATA_OBJECT_ID = 8031;

To determine whether a recovery problem is isolated, you can run a diagnostic trial recovery, which scans the redo stream for problems but does not actually make any changes to the recovered database. If a trial recovery discovers any recovery problems, it reports them in the alert_SID.log. You can use the RECOVER ... TEST statement to invoke trial recovery.

See Also:

"Performing Trial Recovery"

After you have done these investigations, you can follow the guidelines in Table 5-3 to decide whether to allow recovery to corrupt blocks.

Table 5-3 Guidelines for Allowing Recovery to Permit Corruption  
If the problem is . . . and the block is . . . Then . . .

not isolated

n/a

You should probably open the database with the RESETLOGS option. This response is important for stuck recovery problems, because stuck recovery can be caused by the operating system or a storage system losing writes. If an operating system or storage system suddenly fails, it can cause stuck recovery problems on several blocks.

isolated

in the SYSTEM tablespace

Do not corrupt the block, because it may eventually prevent you from opening the database. However, sometimes data in the SYSTEM tablespace is unimportant. If you must corrupt a SYSTEM block and recover all changes, contact Oracle Support.

isolated

index data

Consider corrupting index blocks because the index can be rebuilt later after the database has been recovered.

isolated

user data

Decide based on the importance of the data. If you continue with datafile recovery and corrupt a block, you lose data in the block. However, you can use RMAN to perform block media recovery later after datafile recovery completes. If you open RESETLOGS, then the database is consistent but loses any changes made after the point where recovery was stopped.

isolated

rollback or undo data

Consider corrupting the rollback or undo block because it does not harm the database if the transactions that generated the undo are never rolled back. However, if those transactions are rolled back, then corrupting the undo block can cause problems. If you are unsure, then call Oracle Support.

See Also:

"Performing Trial Recovery" to learn how to perform trial recovery, and "Allowing Recovery to Corrupt Blocks: Phase 4" if you decide to corrupt blocks

Allowing Recovery to Corrupt Blocks: Phase 4

If you decide to allow recovery to proceed in spite of block corruptions, then run the RECOVER command with the ALLOW n CORRUPTION clause, where n is the number of allowable corrupt blocks.

To allow recovery to corrupt blocks:

  1. Ensure that all normal recovery preconditions are met. For example, if the database is open, then take tablespaces offline before attempting recovery.
  2. Run the RECOVER command, allowing a single corruption, repeating as necessary for each corruption to be made. The following statements shows a valid example:
    RECOVER DATABASE ALLOW 1 CORRUPTION
    

Performing Trial Recovery

This section contains these topics:

About Trial Recovery

When problems such as stuck recovery occur, you have a difficult choice. If the block is relatively unimportant, and if the problem is isolated, then it is better to corrupt the block. But if the problem is not isolated, then it may be better to open the database with the RESETLOGS option.

Because of this situation, Oracle supports trial recovery. A trial recovery applies redo in a way similar to normal media recovery, but it never writes its changes to disk and it always rolls back its changes. Trial recovery occurs only in memory.

See Also:

"Allowing Recovery to Corrupt Blocks: Phase 4"

How Trial Recovery Works

By default, if a trial recovery encounters a stuck recovery or similar problem, then it always marks the data block as corrupt in memory when this action can allow recovery to proceed. Oracle writes errors generated during trial recovery to alert files. Oracle clearly marks these errors as test run errors.

Like normal media recovery, trial recovery can prompt you for archived log filenames and ask you to apply them. Trial recovery ends when:

When trial recovery ends, Oracles removes all effects of the test run from the system--except the possible error messages in the alert files. If the instance fails during trial recovery, then Oracle removes all effects of trial recovery from the system because trial recovery never writes changes to disk.

Trial recovery lets you foresee what problems might occur if you were to continue with normal recovery. For problems caused by ongoing memory corruption, trial recovery and normal recovery can encounter different errors.

Executing the RECOVER ... TEST Statement

You can use the TEST option for any RECOVER command. For example, you can start SQL*Plus and then issue any of the following commands:

RECOVER DATABASE TEST
RECOVER DATABASE USING BACKUP CONTROLFILE UNTIL CANCEL TEST
RECOVER TABLESPACE users TEST
RECOVER DATABASE UNTIL CANCEL TEST

By default, trial recovery always attempts to corrupt blocks in memory if this action allows trial recovery to proceed. In other words, trial recovery by default can corrupt an unlimited number of data blocks. You can specify the ALLOW n CORRUPTION clause on the RECOVER ... TEST statement to limit the number of data blocks trial recovery can corrupt in memory.

Note that a trial recovery command is usable in any scenario in which a normal recovery command is usable. Nevertheless, you should only need to run trial recovery when recovery runs into problems.


Go to previous page Go to next page
Oracle
Copyright © 2002 Oracle Corporation.

All Rights Reserved.
Go To Documentation Library
Home
Go To Product List
Book List
Go To Table Of Contents
Contents
Go To Index
Index

Master Index

Feedback