GaOUG Oracle Tech Day 2016

On March 3, 2016, I’ll be at AMA Executive Conference Centers for GaOUG’s Tech Day 2016. This Oracle user conference promises to be packed with the best content from the best the industry has to offer.  If you will be in the Atlanta area and use Oracle products, consider attending the event!

Specifically, I will be presenting:

Presentation Name:
Application High Availability and Upgrades Using Oracle GoldenGate

This presentation will discuss the techniques and methods used to deploy a High Availability Active / Active configuration using Oracle GoldenGate.  Specifically, discussion will surround how to deploy GoldenGate in a standard configuration, utilization of the built in Conflict Detection and Resolution (CDR) functionality and available methodologies to deploy application or database architecture changes within this configuration.  Focus will also be given to how the DBA resources must also involve other IT resources to achieve a successful deployment.

Click here for more information or to register for GaOUG’s Tech Day.

We appreciate your efforts and look forward to seeing you on March 3. If you have any questions, please feel free to contact me anytime.


Data Pump May Cause Exponential Growth in USER_HISTORY$ Records

Data Pump is a common method for moving data from one schema to another.  Oftentimes, the easiest way to do this is to do a full schema export, followed by a full schema import using the appropriate ‘remap’ options.  Usually during a job like this, the user will already exist and it was this situation that uncovered an unexpected behavior in Data Pump.  In our situation, the weekly data copy job ran for 37 weeks and all of a sudden the job started to noticeably take longer and longer, until the point to where it would run for days and then never finish.

Upon investigation, we found that the data pump job was initially hanging on the import Data Pump during this step:


Because we didn’t really know what was going on and we needed the job to finish, we excluded “PASSWORD_HISTORY” from the import DataPump.  Then just a few short weeks later, the job then was exhibiting the same behavior on the export.  Clearly something more was going on.  After taking some time to analyze ASH reports, it was clear that the process was getting hung during processing of the USER_HISTORY$ table.

What is the USER_HISTORY$ table?

As it turns out, this table stores the actual password history for each user in the database.

SQL> desc user_history$
Name Null? Type
----------------------- -------- ----------------

This table is never purged so each time the user receives a password change, a row is written to this table.  In our case a single password change for this user resulted in millions of rows for this user over the 44 weeks that the job had been occurring and Data Pump was exacerbating the issue.  With each export / import Data Pump, and because we were alternating schemas, this one row became two, two became three, three became five and so on, until week 37 where the table had almost 15 million rows.


To demonstrate the issue, we will use very simple parameter files and export / import the same schema:

USERID='/ as sysdba'


USERID='/ as sysdba'

We should initially check the user_history$ table to determine the baseline for a particular user:

SQL> select name, count(*)
2 from user_history$, user$
3 where user_history$.user# = user$.user#
4 and name like 'HR%'
5 group by name order by name;

----- ---------------
HR    2
HR2   6

We will then run an export and import using the parameter files above and then re-run the query:

SQL> select name,count(*)
 2 from user_history$ a, user$ b
 3 where a.user#=b.user#
 4 and name like 'HR%'
 5 group by name order by 1;

----- ---------------
HR    2
HR2   12

And sure enough, you see the rows for password history have doubled just from running the import 1 time!


Luckily we now have two solutions available to us with minimal impacts.  Oracle has made a patch available (16811897) for all versions and above, you can upgrade to or you can exclude ‘PASSWORD_HISTORY’ from the export / import process.

Of course if you are running into this issue, make sure you test thoroughly as results may vary!

Beware the Orphan Streams Capture Effect on Archivelogs

Recently, I ran into an issue where the Oracle Backup and Recovery Manager (RMAN) would not delete old database archivelogs.  I have had this happen before when there as a lag condition with either data guard or GoldenGate extract process.  Upon further investigation, I found that RMAN was issuing the following error.  It looked just like the other times I have encountered this issue:

RMAN-08137: WARNING: archived log not deleted, needed for standby or upstream capture process
archived log file name=+RECOC1/……

This error repeated many times, representing a ton of space being consumed in the DB_FILE_RECOVERY_DEST.  In order to begin diagnosis of the problem (which also happens to be a data guard environment), I went through the normal motions of ensuring proper application of logs to the physical standby environments:

/* On Primary */
SELECT * FROM v$dataguard_stats;
/* To determine which logs are not shipped and applied yet */</p>
<p style="padding-left: 30px;">SELECT *
FROM v$archived_log
ORDER BY completion_time;
/* Run on both source and standby to compare the last logs that were applied */</p>
<p style="padding-left: 30px;">SELECT THREAD#, MAX(SEQUENCE#) AS "LAST_APPLIED_LOG"
/* On Primary */
WHERE metric_name='Redo Generated Per Sec'
WHERE rownum<=10;
/* On Primary */
SELECT current_scn FROM v$database;
/* On Standby */
SELECT current_scn FROM v$database;
/* On Primary - To determine actual time difference */

After looking at data guard performance, it was clear that data guard itself was not holding RMAN from removing the archivelogs.  In addition to data guard being present, this system is in the process of being migrated from an older system via GoldenGate.  I have seen GoldenGate cause this issue before when an extract was registered with “log retention”, but this is the target system so this isn’t possible, RIGHT?  Just to be sure, I ran a query to see if there were any items which caused the database to think there was a GoldenGate object registered with the system:

/* On Primary - Where archivelogs are being held */


02-NOV-15 AM


Hey, wait a minute, we do not have an GoldenGate extract or any other registered GoldenGate objects running on this database so why is this entry here?  I then ran the same query as above on my source server and there it is.  The same object on the source.

So how did this happen?

It all points back to the initial data pump that we used to instantiate the database before turning on GoldenGate.  We used a ‘FULL’ data pump export which was taken AFTER the extract was started on the source.  Because of this the export also contained the capture objects necessary to register the extract with ‘LOGRETENTION’.


In order to remedy this situation we need to completely remove this ‘orphaned’ capture object from the database.  To do this we need to use the dbms_streams_adm package.  Utilize all of the package defaults, so that you will raise an error should you try to delete the incorrect queue:


You should now re-execute the query against dba_capture and the queue just deleted, should no longer be there:

/* On Primary - Where archivelogs are being held */

no rows selected

From this point on, your archivelogs should not be required for any “standby or upstream capture process” and RMAN should now delete your backed up archivelogs providing free space in your DB_FILE_RECOVERY_DEST!

Understand Integrated Replicat Performance using the GGSCI STATS Command

With GoldenGate 12c, Integrated Replicat has now become the preferred option to apply transactions (Classic and Coordinated are the others). In order to get the most efficiency out of the Integrated Replicat, it is important to understand what types of operations cannot be applied by the integrated apply server. GoldenGate considers operations which cannot be applied in “integrated” mode to be applied in “direct” mode.

So what is the difference between Integrated and Direct?

“Integrated” applies transactions via a Logical Change Record (LCR) vs. “Direct” which applies transactions with a SQL Statement via OCI. “Direct” also requires that transactions be applied serially vs. “Integrated” where they can be applied in a parallel, coordinated fashion.

There are some limitations to applying changes via a LCR.  The following are items which can only be applied in “direct” mode:

  • DDL operations
  • Sequence operations
  • SQLEXEC parameter within a TABLE or MAP parameter
  • EVENTACTIONS processing
  • UDT Note, if the extract uses USENATIVEOBJSUPPORT to capture the UDT, then Integrated Replicat will apply it with the inbound server, otherwise it will be handled by Replicat directly.

Understanding the types of operations going on inside your database should be one of the first steps in setting up a GoldenGate environment and heavy use of “direct” transactions within an Integrated Replicat will likely reduce the performance of the replicat.

Integrated Replicat Statistics Explained:

  • Total transactions – Total transactions processed by replicat
  • Redirected – Number of transactions for which replicat is redirected to classic mode
  • DDL operations – Number of DDL operations processed in direct mode
  • Stored procedures – Number of Stored procedures in direct mode
  • Datatype functionality – Number of user defined data types processed in direct mode
  • Event actions – Number of event actions processed in direct mode
  • Direct transactions ratio – Percentage of transactions that the Integrated Replicat converts itself to direct mode to apply transactions.

Direct transactions ratio are computed by either:

Direct transactions ratio = (Redirected/Total transactions)*100
Direct transactions ratio = ((DDL operations + Stored procedures + Datatype functionality + Event actions) / Total transactions)*100

The lower the direct transactions ratio, the better the performance. This is one thing that should be considered when changing to Integrated Replicat.  If the ratio is high, then it could be better to use Classic Replicat to improve performance.

Example of Integrated Replicat Statistics:

GGSCI (orcl12c-rac1.localdomain) 4> stats R_TST_R1 totalsonly *.*

Sending STATS request to REPLICAT R_TST_R1 ...

Start of Statistics at 2015-09-14 19:34:26.
 Integrated Replicat Statistics:

Total transactions 11829288.00
 Redirected 0.00
 DDL operations 0.00
 Stored procedures 0.00
 Datatype functionality 295977.00
 Event actions 0.00
 Direct transactions ratio 2.50%

 GGSCI (orcl12c-rac1.localdomain) 2> stats R_TST_R2 totalsonly *.*

Sending STATS request to REPLICAT R_TST_R2 ...

Start of Statistics at 2015-09-14 19:11:48.
 Integrated Replicat Statistics:

Total transactions 2781917.00
 Redirected 0.00
 DDL operations 0.00
 Stored procedures 0.00
 Datatype functionality 733578.00
 Event actions 0.00
 Direct transactions ratio 26.37%

As you can see in the second example, the “Direct transactions ratio” is much higher. Because I know this system, I know that this higher ratio is entirely attributable to the replication of sequence objects. Using this as an example, a few options to lower the Direct transaction ratio and improve performance might include the discontinuation of sequence replication or splitting sequence replication into their own ‘classic’ mode replicat. Of course if current performance is adequate, you could do nothing. I would probably consider changing to classic replicat when the “Direct transactions ratio” approached 50%.

No matter the situation, it is important to understand the statistics which GoldenGate is recording on your behalf. It may lend good insight into what is going on in your environment.

Oracle 11g Data Pump EXCLUDE Parameter Caveat When the FULL=Y Parameter is Used

During the course of my work, I migrate many databases from one host to another.  Occasionally, DBAs must create backup tables when diagnosing an issue or implementing a change and more frequently than not, these objects never get dropped.  Usually this is no problem except for when you need to move a database from one host to another and you do not want to move these old backup objects.  In my latest migration, I found over 300GB of backup tables which would certainly elongate the time needed to complete the data pump.  After obtaining approval that these tables did not need to be migrated, I began to determine how to exclude these tables from the data pump.  At first, I thought I had it all figured out, data pump everything and exclude only the objects I did not want to bring over.  High level the steps looked like this:

  1. Create a work table in the database which contained the list of excluded tables.
  2. Load the reference data which will be queried by data pump for the exclude list.
  3. Create the data pump parameter file
  4. Execute the data pump
Exclusion table DDL (this simplifies the parameter file):


Data pump parameter file:

USERID='/ as sysdba'

You will have all objects except for the few tables which you excluded.  Right?  Well, almost.   You have everything you asked for except, the data pump left out ALL of the following paths:


So after many hours of research, I found this as a bug documented in Oracle Doc 1491557.1 and effects nearly all versions of 11g.  As a work around, you can do one of the following:

  1. Do another export of metadata without exclusions to get these object paths and import them one by one for each schema
  2. Do a FULL=Y with no exclusions and add the EXCLUDE parameter on the impdp parameter file like this:
    USERID='/ as sysdba'
  3. Apply patch 14095143 and the EXCLUDE will work properly
  4. Upgrade to

Have fun!

To see a where exactly the data pump paths were left out, feel free to see a better representation here:
datapump export compare

Set Up and Use of ACFS for GoldenGate

Recently, I found myself in a situation where I needed a Linux mount point of sufficient space for GoldenGate binaries / trail files.  I’ve used the Oracle Database File System (DBFS) option in the past, although I never really was a big fan of it since its use creates additional database objects and in my opinion unnecessary additional database I/O as well as additional redo and rman activity.  Based on this, I decided to explore the use of Oracle ASM Clustered File System (ACFS) for this use case.  At first glance, it seemed to be much faster to set up and was available on all nodes by default, which would also allow GoldenGate to fail over to other nodes.  In addition,  ACFS does not require the database to be up so the filesystem can also be used for other purposes.  If you are using this mount solely for GoldenGate, make sure you follow the best practices document which is updated periodically (Oracle GoldenGate Best Practice: NFS Mount options for use with GoldenGate (Doc ID 1232303.1))

***  Refer to the following steps at your own risk and always test for your use case prior to using in a production setting.


  • Root user access
  • Sufficient ASM Space
  • Separate ASM Diskgroup (Optional)
  • Latest Oracle Grid Infrastructure and Database Patchset


Verify that ACFS/ADVM modules are present in memory (on each node):

 $ lsmod | grep oracle

If the modules are not present, the command will return something similar to:
oracleasm              53591  1

If the modules are present, the command will return something similar to:
oracleacfs 3308260 0
oracleadvm 508030 0
oracleoks 506741 2 oracleacfs,oracleadvm
oracleasm 53591 1

If the modules are not present or you would like to ensure that the latest version is loaded, run the following before proceeding (as the root user):

$ . oraenv


The Oracle base remains unchanged with value /u01/app/oracle

# $GRID_HOME/bin/acfsroot install

Reboot the node if the modules were already present and you are reloading them.

Start and enable the ACFS modules on each node:

On each node and as the root user:

# $GRID_HOME/bin/acfsload start
ACFS-9391: Checking for existing ADVM/ACFS installation.
ACFS-9392: Validating ADVM/ACFS installation files for operating system.
ACFS-9393: Verifying ASM Administrator setup.
ACFS-9308: Loading installed ADVM/ACFS drivers.
ACFS-9327: Verifying ADVM/ACFS devices.
ACFS-9156: Detecting control device '/dev/asm/.asm_ctl_spec'.
ACFS-9156: Detecting control device '/dev/ofsctl'.
ACFS-9322: completed

If running Grid Infrastructure, enable the driver modules in clusterware (only on one node as the root user):

# $GRID_HOME/bin/acfsroot enable
ACFS-9376: Adding ADVM/ACFS drivers resource succeeded.
CRS-2672: Attempting to start 'ora.drivers.acfs' on 'orcl-rac1'
CRS-2676: Start of 'ora.drivers.acfs' on 'orcl-rac1' succeeded
ACFS-9380: Starting ADVM/ACFS drivers resource succeeded.
ACFS-9368: Adding ACFS registry resource succeeded.
CRS-2672: Attempting to start 'ora.registry.acfs' on 'orcl-rac2'
CRS-2672: Attempting to start 'ora.registry.acfs' on 'orcl-rac1'
CRS-2676: Start of 'ora.registry.acfs' on 'orcl-rac2' succeeded
CRS-2676: Start of 'ora.registry.acfs' on 'orcl-rac1' succeeded
ACFS-9372: Starting ACFS registry resource succeeded.

Once installation is complete, and the mount is registered with clusterware, these modules will be loaded automatically.

If you like you can double check the driverstate by using the following executable:
usage: acfsdriverstate [-orahome ] [-s]

As oracle user, create an ASM volume for ACFS (run only on one node):

Source in the grid environment.

$ . oraenv
The Oracle base remains unchanged with value /u01/app/oracle

Create the volume using the volcreate command.
You can use an existing disk group or create a separate one to house ACFS.

$ asmcmd
ASMCMD> volcreate -G DATA -s 10G ACFSVOL1
ASMCMD> volinfo --all
Diskgroup Name: DATA

Volume Name: ACFSVOL1
Volume Device: /dev/asm/acfsvol1-370
Size (MB): 1024
Resize Unit (MB): 64
Redundancy: UNPROT
Stripe Columns: 8
Stripe Width (K): 1024

As oracle user, create the filesystem on the volume which was just created:

$ /sbin/mkfs -t acfs /dev/asm/acfsvol1-370

mkfs.acfs: version =
mkfs.acfs: on-disk version = 39.0
mkfs.acfs: volume = /dev/asm/acfsvol1-370
mkfs.acfs: volume size = 1073741824 ( 1.00 GB )
mkfs.acfs: Format complete.

As root, create an empty directory which will house the file system:

# mkdir -p /acfsmounts/acfsvol1
# chown root:oinstall /acfsmounts
# chmod 770 /acfsmounts
# chown -R oracle:oinstall /acfsmounts/acfsvol1
# chmod 775 /acfsmounts/acfsvol1

As root, setup the file system to be auto mounted by clusterware:

In a RAC 11g environment, you use acfsutil (srvctl may be supported – was not tested and the “-u option” will allow the oracle user to administer the mount):
# . /usr/local/bin/oraenv
The Oracle base remains unchanged with value /u01/app/oracle
# /sbin/acfsutil registry -a /dev/asm/acfsvol1-370 /acfsmounts/acfsvol1 -t "ACFS General Purpose Mount" -u oracle
In a RAC 12c GI environment, register it with clusterware using the following commands (the “-u option” will allow the oracle user to administer the mount):
# . /usr/local/bin/oraenv
The Oracle base remains unchanged with value /u01/app/oracle
# srvctl add volume -volume ACFSVOL1 -diskgroup DATA -device /dev/asm/acfsvol1-370
# srvctl add filesystem -device /dev/asm/acfsvol1-370 -path /acfsmounts/acfsvol1 -diskgroup DATA -user oracle -fstype ACFS -description "ACFS General Purpose Mount"

At this point the mount should be ready for read/write and will be automatically mounted by clusterware.

Administration of the ACFS mount:

If you need to resize the mount once created (since you granted control to the oracle user, this command can also be executed by the oracle user:

$ acfsutil size 25G /acfsmounts/acfsvol1
$ srvctl start filesystem -device /dev/asm/acfsvol1-370
$ srvctl stop filesystem -device /dev/asm/acfsvol1-370