Sunday, February 25, 2018

Is corruption always true corruption ?


Well the short answer is that there are different types of corruption.  I am writing this blog because I recently ran across “nologging” corruption which is considered softl corruption, and the handling of this type of corruption has changed across versions (which I will cover at the end of this article).

First, how does this happen ?  It can be seen in a physical standby, but you can also get this type of corruption in a Primary database.  I will go through how to reproduce this type of corruption shortly. 

Next, what is nologging ?  Nologging processes are direct path updates to the database that do not contain detailed log information.  Notice I used the word “detailed” .  This is because some logging is captured as to what blocks are updated, but the individual updates are not captured.
Nologging is most common in datawarehouse load processes (ELT)  that are part of workflow that can be restarted.  Often tables are created during this processing that are only kept as part of the processing.  Nologging can also be used for performing large inserts into existing tables.  Because this type of processing tends to be “logging intensive”, and steps can be re-run, nologging is utilized for these objects.  Nologging can speed up processing by performing limited logging.  The downside of nologging is that for objects updated with nologging, there is no point in time recovery capability.  The object can only be restore/recovered to the point where a backup is taken (full or incremental).  More on that later.

I will start by showing a nologging workflow.

Below are the steps on how to reproduce a nologging test.


  1)      Ensure that force_logging=false   ---  If FORCE_LOGGING is turned on, any nologging processing is handled as logging



SQL> select force_logging from v$database;


FOR
---
NO


2)   Create a nologging table


SQL>  create table bgrenn.test nologging as select * from dba_objects where 0=1;

Table created.
 

3)      Ensure the new table is nologging


SQL> select owner,table_name,logging from dba_tables where owner='BGRENN';

OWNER                          TABLE_NAME                     LOG
------------------------------ ------------------------------ ---
BGRENN                         TEST                           NO


4)      Perform a full backup of the database


RMAN> backup incremental level 0 database;

Starting backup at 23-february -2018 12:45:45
using target database control file instead of recovery catalog

allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=1069 device type=DISK
allocated channel: ORA_DISK_2
channel ORA_DISK_1: starting incremental level 0 datafile backup set
channel ORA_DISK_1: specifying datafile(s) in backup set
input datafile file number=00005 name=/oradata/db1/soe.dbf
channel ORA_DISK_7: starting incremental level 0 datafile backup set
channel ORA_DISK_7: specifying datafile(s) in backup set
input datafile file number=00004 name=/oradata/db1/users01.dbf
channel ORA_DISK_7: starting piece 1 at 23-february -2018 12:46:22
channel ORA_DISK_8: starting incremental level 0 datafile backup set
channel ORA_DISK_8: specifying datafile(s) in backup set
including current SPFILE in backup set
channel ORA_DISK_8: starting piece 1 at 23-february -2018 12:46:28
channel ORA_DISK_2: finished piece 1 at 23-february -2018 12:46:30

….
Finished backup at 23-february -2018 12:50:01


5)      Update the table no logging using append hint

SQL> insert into /*+ append */ bgrenn.test select * from dba_objects nologging;

68947 rows created.

SQL> Commit;



6)      Switch the logfile to ensure the changes are written to archive logs.

SQL> alter system switch logfile;




OK. Now we have done a Full backup of the database, and performed a nologging change to my table “bgrenn.test”.  I did a log switch to ensure the change is written to the archive log.

The next step is to reproduce the nologging “soft corruption” through a restore.
At this point, blocks containing my table were inserted into, but the actual changes were not logged.  The block numbers were written to the log file, and on recovery these blocks will be marked as being changed.


1)      Check for block corruption before restoring

SQL> select * from v$database_block_corruption;

no rows selected


2)       Retart the database mount and restore the database

RMAN> connect target /

connected to target database: DB16 (DBID=3618808394)

RMAN> shutdown immediate;

using target database control file instead of recovery catalog
startup mount;
database closed
database dismounted
Oracle instance shut down
using target database control file instead of recovery catalog
startup mount;
database closed
database dismounted
Oracle instance shut down


RMAN> startup mount;

connected to target database (not started)
Oracle instance started
database mounted

Total System Global Area   14564409344 bytes

Fixed Size                     2149720 bytes
Variable Size               6308233896 bytes
Database Buffers            8187281408 bytes
Redo Buffers                  66744320 bytes

RMAN> restore database;

Starting restore at 23-february -2018 12:55:54
allocated channel: ORA_SBT_TAPE_1
channel ORA_SBT_TAPE_1: SID=1110 device type=SBT_TAPE
channel ORA_SBT_TAPE_1: Oracle Secure Backup
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=1109 device type=DISK
channel ORA_DISK_5: starting datafile backup set restore
channel ORA_DISK_5: specifying datafile(s) to restore from backup set
channel ORA_DISK_5: restoring datafile 00004 to /oradata/db1116/users01.dbf
channel ORA_DISK_5: reading from backup piece /u01/app/oracle/flash_recovery_area/DB1116/backupset/2018_02_23/o1_mf_nnnd0_TAG20180223T12454
7_f90vwp3j_.bkp
Finished restore at 23-february -2018 12:57:41






3)      Recover database


RMAN> recover database;

Starting recover at 23-february -2018 12:58:22
using channel ORA_SBT_TAPE_1
using channel ORA_DISK_1
using channel ORA_DISK_2
using channel ORA_DISK_3
using channel ORA_DISK_4
using channel ORA_DISK_5
using channel ORA_DISK_6
using channel ORA_DISK_7
using channel ORA_DISK_8

starting media recovery
media recovery complete, elapsed time: 00:00:01

Finished recover at 23-february -2018 12:58:23

RMAN> alter database open;

database opened

RMAN>




4)      Check for corruption after restoring the database


SQL> select * from v$database_block_corruption;

no rows selected


5)      Select from the table in which we ran our nologging process



SQL> select * from bgrenn.test;
select * from bgrenn.test
                     *
ERROR at line 1:
ORA-01578: ORACLE data block corrupted (file # 4, block # 60)
ORA-01110: data file 4: '/oradata/db1/users01.dbf'
ORA-26040: Data block was loaded using the NOLOGGING option




6)      Check corruption again


SQL> select * from v$database_block_corruption;

no rows selected



7)       Validate datafile



RMAN> validate datafile 4;

Starting validate at 23-february -2018 14:15:36
using target database control file instead of recovery catalog
allocated channel: ORA_DISK_1
channel ORA_DISK_1: SID=1107 device type=DISK
channel ORA_DISK_8: SID=1073 device type=DISK
channel ORA_DISK_1: starting validation of datafile
channel ORA_DISK_1: specifying datafile(s) for validation
input datafile file number=00004 name=/oradata/db1/users01.dbf
channel ORA_DISK_1: validation complete, elapsed time: 00:00:01
List of Datafiles
=================



File Status Marked Corrupt Empty Blocks Blocks Examined High SCN
---- ------ -------------- ------------ --------------- ----------
4    OK     1016           307          1440            1760824163
  File Name: /oradata/db1/users01.dbf
  Block Type Blocks Failing Blocks Processed
  ---------- -------------- ----------------
  Data       0              63             
  Index      0              2              
  Other      0              1068           

Finished validate at 23-february -2018 14:15:42


Now this is where it gets interesting between version of Oracle.  Oracle 10g/11.1 reports this soft corruption differently from Oracle 11.2, and  both of these report it differently from 12.1+

Oracle 11.1   -
1)      Check for corruption in v$DATABASE_BLOCK_CORRUPTION  -- NOTE the corrupt_type only reports "CORRUPTION"



SQL> select * from v$database_block_corruption;

     FILE#     BLOCK#     BLOCKS CORRUPTION_CHANGE# CORRUPTIO
---------- ---------- ---------- ------------------ ---------
         4         60         13         1760822701 CORRUPT
         4         74         15         1760822716 CORRUPT
         4         90         15         1760822716 CORRUPT
         4        106         15         1760822730 CORRUPT
         4        122         15         1760822730 CORRUPT
         4        138         15         1760822745 CORRUPT
         4        154         15         1760822745 CORRUPT
         4        170         15         1760822759 CORRUPT
         4        267        126         1760822759 CORRUPT
         4        395        126         1760822763 CORRUPT
         4        523        126         1760822767 CORRUPT
         4        651        126         1760822771 CORRUPT
         4        779        126         1760822775 CORRUPT
         4        907        126         1760822779 CORRUPT
         4       1035        126         1760822784 CORRUPT
         4       1163         16         1760822788 CORRUPT


2)      Definition  of  CORRUPTION_TYPE

·         ALL ZERO - Block header on disk contained only zeros. The block may be valid if it was never filled and if it is in an Oracle7 file. The buffer will be reformatted to the Oracle8 standard for an empty block.
·         FRACTURED - Block header looks reasonable, but the front and back of the block are different versions.
·         CHECKSUM - optional check value shows that the block is not self-consistent. It is impossible to determine exactly why the check value fails, but it probably fails because sectors in the middle of the block are from different versions.
·         CORRUPT - Block is wrongly identified or is not a data block (for example, the data block address is missing)
·         LOGICAL - Specifies the range is for logically corrupt blocks. CORRUPTION_CHANGE# will have a nonzero values

3)      OEM Schedule Backup screen shows corruption

Oracle 11.2   -
1)      Check for corruption in v$DATABASE_BLOCK_CORRUPTION  -- NOTE 11.2 reports the soft corrupt in the view as "NOLOGGING" corruption.



SQL> select * from v$database_block_corruption;

     FILE#     BLOCK#     BLOCKS CORRUPTION_CHANGE# CORRUPTIO
---------- ---------- ---------- ------------------ ---------
         4         60         13         1760822701 NOLOGGING
         4         74         15         1760822716 NOLOGGING
         4         90         15         1760822716 NOLOGGING
         4        106         15         1760822730 NOLOGGING
         4        122         15         1760822730 NOLOGGING
         4        138         15         1760822745 NOLOGGING
         4        154         15         1760822745 NOLOGGING
         4        170         15         1760822759 NOLOGGING
         4        267        126         1760822759 NOLOGGING
         4        395        126         1760822763 NOLOGGING
         4        523        126         1760822767 NOLOGGING
         4        651        126         1760822771 NOLOGGING
         4        779        126         1760822775 NOLOGGING
         4        907        126         1760822779 NOLOGGING
         4       1035        126         1760822784 NOLOGGING
         4       1163         16         1760822788 NOLOGGING



2)      Definition  of  CORRUPTION_TYPE
·         ALL ZERO - Block header on disk contained only zeros. The block may be valid if it was never filled and if it is in an Oracle7 file. The buffer will be reformatted to the Oracle8 standard for an empty block.
·         FRACTURED - Block header looks reasonable, but the front and back of the block are different versions.
·         CHECKSUM - optional check value shows that the block is not self-consistent. It is impossible to determine exactly why the check value fails, but it probably fails because sectors in the middle of the block are from different versions.
·         CORRUPT - Block is wrongly identified or is not a data block (for example, the data block address is missing)
·         LOGICAL - Block is logically corrupt
·         NOLOGGING - Block does not have redo log entries (for example, NOLOGGING operations on primary database can introduce this type of corruption on a physical standby)

3)      OEM Schedule Backup screen shows corruption



Oracle 12.1   -
1)      Check for corruption in v$DATABASE_BLOCK_CORRUPTION  -- NOTE - 12.1 does not report it as corruption, but is reported in a new view  V$NONLOGGED_BLOCK


SQL> select * from v$database_block_corruption;

no rows selected



SQL> select file#,block#,blocks from v$nonlogged_block

     FILE#     BLOCK#     BLOCKS
---------- ---------- ----------
         6       6786        126
         6       6914        126
         6       7042        126
         6       7170        126
         6       7298        126
         6       7426        126
         6       7554        126
         6       7682        126
         6       7810        126
         6       7938        126
         6       8066        126

     FILE#     BLOCK#     BLOCKS
---------- ---------- ----------
         6       8194        126
         6       8322        126
         6       8512         64

47 rows selected.




2)      Definition  of  CORRUPTION_TYPE
·         ALL ZERO - Block header on disk contained only zeros. The block may be valid if it was never filled and if it is in an Oracle7 file. The buffer will be reformatted to the Oracle8 standard for an empty block.
·         FRACTURED - Block header looks reasonable, but the front and back of the block are different versions.
·         CHECKSUM - optional check value shows that the block is not self-consistent. It is impossible to determine exactly why the check value fails, but it probably fails because sectors in the middle of the block are from different versions.
·         CORRUPT - Block is wrongly identified or is not a data block (for example, the data block address is missing)
·         LOGICAL - Block is logically corrupt

3)      OEM Schedule Backup screen

Nothing appears


Now we have seen how to recreate “soft corruption” caused by nologging.  I have also shown how this is displayed in different versions of Oracle.

There are a few items to note that I have learned from this testing.

·        This is considered “soft corruption” so it is not reported when restoring a database.  This makes it very hard detect.

·        The ZDLRA does validation during backups, but since this is “soft corruption”, the database is backed up without no alerting.

·        OEM reports this corruption differently between versions.  With version 12.1 it is no longer reported in V$DATABASE_BLOCK_CORRUPTION, so OEM does not alert on this.

How to avoid Nologging corruption.

Ensure that you schedule backups when there isn’t any nologging operations occurring.This is a situation where the ZDLRA shines.  You can take an incremental backup before and after your nologging process then you have the capability to perform a full restore from either of these checkpoints.




Monday, October 2, 2017

Creating Popup windows on your Apex Page

I have been playing with Apex for an internal application  Application Express is a great tool, and Oracle has an internal Apex environment that groups can use for their own internal applications.

In creating the application I learned how to do a Javascript window that pops up within a page to help enter data. This can be very useful to add a function to your application without adding more pages.


This is how its down..

First I created a table to contain breweries..  Since this is running on 12.1, I was able to use the new feature to automagically use a sequence as a default value (it's about time right) ?

Here is my table creation script..


CREATE SEQUENCE brewery_seq
 START WITH     1000
 INCREMENT BY   1
 NOCACHE
 NOCYCLE;

Create table breweries
(brewery_id number default brewery_seq.nextval primary key,
         brewery_name varchar2(255),
         Brewery_rating number);

So first I did was create a new region on the a new page.



ZDLRA and the FRA


ZDLRA and the FRA


I often get questions around the FRA (Flash Recovery Area), and how it should be used when moving backups to the ZDLRA.

First though, the recommendation is to ALWAYS set your db_recovery_file_dest_size to be 10% less than the amount of space available, and don't put other files in this same location (that are not managed as part of the FRA).  
Having a 10% buffer ensures that you can increase the available storage if necessary.  For those experienced on-call DBA's I'm sure there have been times where increasing the db_recovery_file_dest_size by that last 10% was used to keep the database running while space was cleaned up.. And of course this is often at 3:00 AM when the dreaded "archive log destination full" alert comes across.



First let's go through what's in the FRA and how it's being used.   


There is a lot of information in MOS and I will include pertinent MOS notes at the end of this post.

What's in the FRA  (V$FLASH_RECOVERY_AREA_USAGE shows us )?

Here is a sample output 


SQL> Select file_type, percent_space_used as used,
percent_space_reclaimable as reclaimable, 
    number_of_files as "number" from v$flash_recovery_area_usage; 
     
    FILE_TYPE          USED RECLAIMABLE     number 
    ------------ ---------- ----------- ---------- 
    CONTROLFILE           0           0          0 
    ONLINELOG             0           0          0 
    ARCHIVELOG         4.77           0          2 
    BACKUPPIECE       56.80           0         10 
    IMAGECOPY             0           0          0 
    FLASHBACKLOG      11.68       11.49         63 



From this you can see the following items are in the FRA.


CONTROLFILE  -- This comes from setting the location of the CONTROLFILE backup.


configure controlfile autobackup on;

If you configure controlfile backups using the 'FORMAT" option, it will not be managed by the FRA.


ONLINELOG


A copy of the online redo logs go to the FRA  when the DB_RECOVERY_FILE_DEST is set and the DB_CREATE_ONLINE_LOG_DEST_n is not set.





ARCHIVELOG -- 

Archive logs are managed by the FRA when the archive LOG_ARCHIVE_DEST_n parameter contains the clause 'LOCATION=USE_DB_RECOVERY_FILE_DEST'

alter system set LOG_ARCHIVE_DEST_1='LOCATION=USE_DB_RECOVERY_FILE_DEST'';


BACKUPPIECE or IMAGECOPY

RMAN backups are managed by the FRA when you configure RMAN to (or explicitly) send backups to disk AND the FORMAT option is not specified.

FLASHBACKLOG 


If flashback database is turned on for the database, flashback logs will be kept in the FRA automatically.  The database parameter db_flashback_retention_target is set to determine the amount of flashback logs that are kept for the database.


Now let's take a look at how space is managed for each piece 


ONLINELOG  -- Since online redo logs are necessary for the database, this is not affected with space pressure.

FLASHBACKLOG   -- Flashback logs are automatically removed to keep the window specified in the DB_FLASHBACK_RETENTION_TARGET setting.  If there is space pressure, the flashback management will automatically release space until it hits a window of 1 hour.  This is the default and comes from the _MINIMUM_DB_FLASHBACK_RETENTION parameter.

BACKUPPIECE or IMAGECOPY and/or CONTROLFILE and ARCHIVELOG --  these are managed by the setting in RMAN for the retention policy. 


The recommendation for the ZDLRA (when using real-time redo apply) is to set this parameter to


CONFIGURE ARCHIVELOG RETENTION POLICY SHIPPED to all STANDBY;


NOTE - This policy marks archive logs as "reclaimable" as soon as they are considered shipped to ALL standby databases.  If the ZDLRA is the only external destination, then as soon as it's received on the ZDLRA, it is "reclaimable".  If you have Dataguard along with ZDLRA, then archive logs are considered "reclaimable" as soon as they are received by both.

OR

CONFIGURE ARCHIVELOG RETENTION POLICY APPLIED on all STANDBY;

NOTE - This policy marks archive logs as "reclaimable" as soon as they are considered APPLIED on ALL standby databases.  If the ZDLRA is the only external destination, then as soon as it's received on the ZDLRA AND has been cataloged, it is "reclaimable".   This takes less than a minute, and occurs independent of the protected database.  If you have Dataguard along with ZDLRA, then archive logs are considered "reclaimable" as soon as they are applied by both.

***  Note that "APPLIED" is the most desirable because this ensures that the log is not removed until it "processed" by all destinations, HOWEVER, if you have a standby location that has a gap in apply then you need to consider "SHIPPED".

Both Policies will work, but it depends on your configuration.


The recommendation for the ZDLRA (when using log sweeps) is to set this parameter to


CONFIGURE ARCHIVELOG RETENTION POLICY xx backed up 1 times device type SBT;

Where XXX is

1) not used if there is no Dataguard or other destinations beyond local disk for archive logs.
2) "SHIPPED to all STANDBY" if Dataguard or Golden Gate is used.
3) "APPLIED" on all STANDBY if Dataguard or Golden Gate is used.


Now let's take a look at how space is managed (315098.1)
There is a MOS note on this, but it there are still some misconceptions on what you see happening.



Looking at the output of view V$RECOVERY_FILE_DEST you will see 3 pertinent columns.

SPACE_LIMIT                              ==> this is the amount of space that is allocated to the FRA.
SPACE_USED                               ==> This is the amount of space currently used
SPACE_RECLAIMABLE             ==> This is the amount space that the FRA considers reclaimable.
                                                                ** This NOT the total amount space reclaimable, just the amount
                                                                that the FRA knows about (I'll get to what this means soon).


The note says that 


If the free space becomes less then 15% in the Flash Recovery Area then all  the archivelogs in the Flash Recovery Area which are not needed for recovery by the current backups in the FRA will become obsolete and the space occupied will be shown in the SPACE RECLAIMABLE column of V$RECOVERY_FILE_DEST.

From this I made some assumptions that were incorrect.

I assumed 

1) Since I was immediately sending all backups and archive logs to the ZDLRA (Real Time Apply), the space for the ARCHIVELOG would all show up in the SPACE_RECLAIMABLE column.
2) When the FRA reaches 85%, it would automatically clean up ARCHIVELOGS to bring it down to 85%.

Both of these assumptions were wrong.  By testing I found that this is what actually happens, and this still falls within the verbiage in that note.


This is what happens as the space fills up.

For my example I have a 1 TB FRA.


849 GB used in the FRA.  The reclaimable space is NOT calculated yet because we have not hit the 85% full mark. 

Space_limit             1000GB
Space_used              849GB
Space_reclaimable       0GB


At 850GB  I reached 85% full. This is the point where the database  calculates SPACE_RECLAIMABLE.  Note that it does not fully calculate what's reclaimable, it only finds a portion of the reclaimable space.
 

Space_limit             1000GB
Space_used              850GB
Space_reclaimable       300GB
 

At 999GB it's not quite full.  The reclaimable space shows there is space available, 
but it still only finds a portion of the available space.
 
Space_limit             1000GB
Space_used              999GB
Space_reclaimable       300GB
 

At 1000GB it is completely full. The reclaimable space shows space available to be reclaimed.
At this point I can see in the alert log that archive logs are being removed. 
Only enough logs are removed to make space for new logs.  it remains at 99% used.

Space_limit             1000GB
Space_used              999GB
Space_reclaimable       160GB
 

As I added more logs, it remained at 99% used.

This makes it very difficult to know how much space you have available in your FRA. 
The warnings start occurring at the 85% full mark.
Since the FRA recalculates the SPACE_RECLAIMABLE at 85% full, but only adjusts it to keep  from warning,
it is impossible to tell how much TOTAL space is reclaimable.

Using the formula 

SPACE_LIMIT - SPACE_RECLAIMABLE 

Does not give you amount of space that is actually reclaimable.  
It is only useful to tell you when the amount of unreclaimable space > 85%.



Here are the useful MOS notes.



NOTE:305817.1 - FAQ - Flash Recovery Area feature



How is the space pressure managed in the Flash Recovery Area - An Example. (Doc ID 315098.1)

Correctly configuring the Flash Recovery Area to allow the release of reclaimable space (Doc ID 316074.1)



Monday, January 18, 2016

ZDLRA

I wanted to write a post on one of Oracle's newest products (Well not that new).. The ZDLRA.
The ZDLRA is often referred to as "Zelda".  I know the name ZDLRA does not roll off the tongue well.  Zelda is a much better (and easier to say name).
The other name you will hear the ZDLRA referred to as is RA or Recover Appliance.

Recover Appliance is probably the best description of the product.  One of the things that makes this product unique is the emphasis on RECOVERY.. Notice there is not a mention of backup in the name.


Here is a great starting point for information


It does a lot of it's magic by using incremental forevers   --

    I know what you are thinking... The incremental forever strategy has been around for a long time (since 10.2 I think).  The idea is simple.  You take a full backup (database copy NOT backup set), and then take incrementals from then on.  You use RMAN to apply the incremental to the full, and create a new full (destroying the old full in the process).  I've seen many customers (and me also) use the rolling incrementals in the online recovery area to keep a full backup online from the previous day.
This is used with a second backup strategy for longer term storage to a backup device.

  The way the RA handles this differently, is that it creates "virtual fulls" for each incremental backup you take.  You also tell it how far back to store virtual fulls.  Using this methodology, if you do an incremental backup nightly, you can keep "virtual fulls" from each night as far back as you want.
There is no need to keep an online backup, and one in backup appliance.

Why is the RA different from most backup strategies ?

1) The use of only needing to do incremental forevers uses less I/O to read database blocks, and less backup network I/O -  Using this method, the RA ONLY needs the incremental backups to keep a restore point.  This saves the I/O of doing a full, and it saves the bytes going across the network.

2) RMAN keeps track of all the backups.  RMAN is the backbone of the RA, and the RA contains a recovery catalog.  RMAN verifies that backups are good, and you will always know if you have a good backup to restore.

3) Real time apply.  You can think of the RA receiving redo log information in the same vein as a Dataguard database.  Without the RA, you would backup archive logs as they are written from the redo logs.  This leaves you open to data loss from your backup.. The RA reads from the current Redo log stream in the database (like dataguard) to ensure there there is almost no data loss. Nothing else does this.

4) Performance.  The performance the RA is phenomenal.  You can find a whitepaper here on the performance with multiple databases backing up to a single RA appliance.

This is a fantastic product to backup multiple databases and most importantly be able to recover your databases with next-to-no data loss.