Creating a Windows Guest on Oracle Database Appliance(ODA) Virtualized Platform

When it comes to deploying a Windows guest as a virtual machine on Oracle's ODA, the documentation is lacking.  After some research and trial and error, I was able to put together a process that will successfully install Windows.

The ODA utilizes the oakcli command line, so OVM is not an option.  In this post, I will walk through the steps needed to prepare and deploy a Windows 7 virtual machine.

Here is a high level summary of the steps.

1. Create an unformatted virtual disk image
2. Create a configuration file for the VM template
3. Create a shared repository
4. Import the VM template
5. Clone a VM from the template
6. Boot from iso file and install Windows
7. Configure Windows network
8. Install Paravirtialized Drivers on Windows guest
9. Modify VM's network


NOTE: For the purposes of this example, all logins for Dom0 and ODA_BASE are on node 0.

First, we need to log into Domain 0 (Dom0) and create an empty unformatted virtual disk image.  The image file name must be System.img.  In the example below, the file size is fixed at 50G.

$ mkdir /OVS/staging/vm_temp/win7x64

$ dd if=/dev/zero of=/OVS/staging/vm_temp/win7x64/System.img oflag=direct bs=1M count=51200

In the same staging directory, create a vm.cfg file with the following content.  The "boot" parameter is set as "dc" which sets the CDROM first in the boot order.  The "disk" parameter names the System.img file created above as the device "hda" in "read/write" mode.  Before the Paravirtualized Drivers(PV) get installed, the "vif" parameter needs to be set with a type of "ioemu".  Later, when the VM is started, a vnc session for the console will be started.  The vnc port is set in the "vfb" parameter with a vncdisplay of 10.  The 10 sets the port to 5910.  A value of 1 would set it to 5901.

kernel = 'hvmloader'
builder = 'hvm'
vcpus = '4'
memory = '4096'
boot = 'dc'
disk = ['file:/OVS/staging/vm_temp/win7x64/System.img,hda,w']
name = 'win7x64'
vif = [ 'type=ioemu,bridge=net1' ]
on_poweroff = 'destroy'
on_reboot = 'destroy'
on_crash = 'destroy'
acpi = '1'
apic = '1'
usbdevice='tablet'
vfb = [ 'type=vnc,vnclisten=0.0.0.0,vncdisplay=10' ]

The next step is creating the tar file containing the two files created above to be imported.

$ cd /OVS/staging/vm_temp/win7x64

$ tar -Sczvf win7x64.tgz System.img vm.cfg

There are default repositories created as part of the ODA_BASE.  The repos "odarepo1" and "odarepo2" are local repositories and will not allow VMs running from these repos to fail over to the other node.  To avoid this, create a shared repository.

NOTE:  Once you create the repo, the size cannot be changed, so plan accordingly.

Log on to ODA_BASE and run the following command to create a 150GB shared repository in the DATA disk group.

$ oakcli create repo odashr -size 150G -dg DATA

Now the environment is set up to import the template.  Login to ODA_BASE as root and run the following command.

$ oakcli import vmtemplate tmpl_win7x64 -files "/OVS/staging/vm_temp/win7x64/win7x64.tgz" -repo odashr -node 0

Once the template has been imported, the VM can finally be cloned.  Using the vmtemplate, run the following command on ODA_BASE.

$ oakcli clone vm vm_win7x64 -vmtemplate tmpl_win7x64 -repo odashr -node 0

Copy the Windows installation iso file to a staging location on Dom0.  Then modify the file /OVS/Repositories/odashr/VirtualMachines/vm_win7x64/vm.cfg on Dom0 by adding the iso as a CDROM.  Below, the iso file /OVS/staging/X17-24281.iso has been defined as device "hdc" and made read only.

disk = [u'file:/OVS/Repositories/odashr/VirtualMachines/vm_win7x64/System.img,hda,w', 'file:/OVS/staging/X17-24281.iso,hdc:cdrom,r']

Now start the VM by logging into ODA_BASE and running the following command.

$ oakcli start vm vm_win7x64

Once the VM starts, log onto Dom0 and attach to the console with a VNC client using the vnc port defined above.  In our example, the port is 5910.  From the VNC console, you need to install Windows.

After the installation is complete, log into Dom0 and modify the /OVS/Repositories/odashr/VirtualMachines/vm_win7x64/vm.cfg file.  Change the boot parameter from "dc" to "cd".  This will name the boot device to the virtual hard drive rather than the CDROM.

Log into ODA_BASE and restart the VM.

$ oakcli stop vm vm_win7x64

$ oakcli start vm vm_win7x64

Now attach to your vnc session again and configure the network settings for the VM guest as per your requirements.  Once the network is configured and functioning, you can download the Windows PV Drivers to the guest and install them by running the executable.  As of the date of this post, the PV drivers can be downloaded from edelivery.oracle.com.  Go to "Cloud Portal Oracle Linux/VM", and enter "Oracle VM" for the product pack and "x86 64 bit" for the platform.

After the PV drivers have been installed, modify the /OVS/Repositories/odashr/VirtualMachines/vm_win7x64/vm.cfg file by changing the "vif" parameter as follows.

From

vif = [ 'type=ioemu,bridge=net1']

To

vif = [ 'type=netfront,bridge=net1']

Now restart your VM, and the guest should be ready for use.  If you want to by-pass using the vnc session, configure remote desktop and connect to the console.

$ oakcli stop vm vm_win7x64

$ oakcli start vm vm_win7x64

SQL Tuning: Using Oracle's Global Hints With Multiple Query Blocks

There's nothing much more frustrating than knowing how to tune a query, but being unable to due to one or more third party views that cannot be modified.  A great way to work around this is by using Oracle's global hints in the main query block.  This is accomplished by referring to the view name and the table name with this format /*+ hint(view.table) */.

According to Oracle's documentation, however, the optimizer ignores global hints in this format that refer to multiple query blocks.  So if you wanted to use the "leading" hint with more than one view like this /*+ leading(view1.table1 view2.table2) */, Oracle will not utilize your input, and you will be left banging your head against the wall.

Luckily, there is another format for referencing these query blocks, but it takes a little digging to get it right.  In the following example, I will create a table with a view that references it along with another view on the ever famous "dual" table.  These two views will be joined in a query and we will see what options there are for manipulating the execution plan without changing the views.

Here are the table and view statements to set up the example.

CREATE TABLE orders AS
SELECT
        LEVEL order_id,
        SYSDATE + DBMS_RANDOM.VALUE(-1000, 1000) order_date,
        DBMS_RANDOM.STRING('A', 20) comments
FROM dual
CONNECT BY LEVEL <= 100000;


CREATE INDEX order_id_ix ON orders ( order_id );

CREATE INDEX order_dt_id_ix ON orders ( order_date, order_id );


CREATE VIEW vw_orders AS
SELECT * FROM orders;

CREATE VIEW vw_dual AS
SELECT * FROM dual;

Here is a basic query which joins the two views

SELECT /*+ gather_plan_statistics */
        order_id,
        order_date,
        comments
FROM vw_orders,
        vw_dual
WHERE TRUNC(order_date) = TO_DATE('12-JAN-13','DD-MON-YY')
AND order_id = 1;

After running the query, we can take a look at the execution plan chosen by the optimizer with the following script.

set pages 999
set lines 200

SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY_CURSOR
        (
        null,
        null,
        'allstats'
        ));

The plan shows that DUAL is the driving table and a nested loop is utilized to access the ORDERS table via an index.

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  dh3p87qwt5njy, child number 0
-------------------------------------
SELECT /*+ gather_plan_statistics */ order_id,  order_date,
comments FROM vw_orders,  vw_dual WHERE TRUNC(order_date) =
TO_DATE('12-JAN-13','DD-MON-YY') AND order_id = 1

Plan hash value: 1760841149

------------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name        | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |             |      1 |        |      1 |00:00:00.01 |       4 |
|   1 |  NESTED LOOPS                |             |      1 |      9 |      1 |00:00:00.01 |       4 |
|   2 |   FAST DUAL                  |             |      1 |      1 |      1 |00:00:00.01 |       0 |
|*  3 |   TABLE ACCESS BY INDEX ROWID| ORDERS      |      1 |      9 |      1 |00:00:00.01 |       4 |
|*  4 |    INDEX RANGE SCAN          | ORDER_ID_IX |      1 |      9 |      1 |00:00:00.01 |       3 |
------------------------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - filter(TRUNC(INTERNAL_FUNCTION("ORDER_DATE"))=TO_DATE('12-JAN-13','DD-MON-YY'))
   4 - access("ORDER_ID"=1)

Note
-----
   - dynamic sampling used for this statement (level=2)


34 rows selected.

Now, if we want to force the optimizer to use a full table scan on the ORDERS table, we can throw in a global hint that references the view and the table.  As you can see from the execution plan, the optimizer follows the directive and a full table scan is performed.

SELECT /*+ gather_plan_statistics full(vw_orders.orders) */
        order_id,
        order_date,
        comments
FROM vw_orders,
        vw_dual
WHERE TRUNC(order_date) = TO_DATE('12-JAN-13','DD-MON-YY')
AND order_id = 1;


PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  9xby7nxuz82tc, child number 0
-------------------------------------
SELECT /*+ gather_plan_statistics full(vw_orders.orders) */
order_id,  order_date,  comments FROM vw_orders,  vw_dual WHERE
TRUNC(order_date) = TO_DATE('12-JAN-13','DD-MON-YY') AND order_id = 1

Plan hash value: 2691672058

---------------------------------------------------------------------------------------
| Id  | Operation          | Name   | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
---------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |        |      1 |        |      1 |00:00:00.01 |     545 |
|   1 |  NESTED LOOPS      |        |      1 |      9 |      1 |00:00:00.01 |     545 |
|   2 |   FAST DUAL        |        |      1 |      1 |      1 |00:00:00.01 |       0 |
|*  3 |   TABLE ACCESS FULL| ORDERS |      1 |      9 |      1 |00:00:00.01 |     545 |
---------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - filter(("ORDER_ID"=1 AND TRUNC(INTERNAL_FUNCTION("ORDER_DATE"))=TO_DATE(
              '12-JAN-13','DD-MON-YY')))

Note
-----
   - dynamic sampling used for this statement (level=2)


32 rows selected.

If we try the same approach with the "leading" hint which references more than one query block, Oracle is not so cooperative.  After running this example, we see that the "leading" hint is ignored and DUAL is still the driving table.

SELECT /*+ gather_plan_statistics full(vw_orders) leading(vw_orders.orders vw_dual.dual) */
        order_id,
        order_date,
        comments
FROM vw_orders,
        vw_dual
WHERE TRUNC(order_date) = TO_DATE('12-JAN-13','DD-MON-YY')
AND order_id = 1;



PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  cbp5hxyg5mjkx, child number 0
-------------------------------------
SELECT /*+ gather_plan_statistics full(vw_orders)
leading(vw_orders.orders vw_dual.dual) */ order_id,  order_date,
comments FROM vw_orders,  vw_dual WHERE TRUNC(order_date) =
TO_DATE('12-JAN-13','DD-MON-YY') AND order_id = 1

Plan hash value: 2691672058

---------------------------------------------------------------------------------------
| Id  | Operation          | Name   | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
---------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT   |        |      1 |        |      1 |00:00:00.01 |     545 |
|   1 |  NESTED LOOPS      |        |      1 |      9 |      1 |00:00:00.01 |     545 |
|   2 |   FAST DUAL        |        |      1 |      1 |      1 |00:00:00.01 |       0 |
|*  3 |   TABLE ACCESS FULL| ORDERS |      1 |      9 |      1 |00:00:00.01 |     545 |
---------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   3 - filter(("ORDER_ID"=1 AND TRUNC(INTERNAL_FUNCTION("ORDER_DATE"))=TO_DATE(
              '12-JAN-13','DD-MON-YY')))

Note
-----
   - dynamic sampling used for this statement (level=2)


33 rows selected.

In order to give the optimizer the information it needs in this situation, we need to change our "DBMS_XPLAN" script so it will include the "alias" and the "outline" information.  In the "DBMS_XPLAN" script below, I have also eliminated the "predicate" and "note" data.

SELECT /*+ gather_plan_statistics */
        order_id,
        order_date,
        comments
FROM vw_orders,
        vw_dual
WHERE TRUNC(order_date) = TO_DATE('12-JAN-13','DD-MON-YY')
AND order_id = 1;


set pages 999
set lines 200

SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY_CURSOR
        (
        null,
        null,
        'allstats +alias +outline -note -predicate'
        ));


PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID  12b497umk9ytc, child number 0
-------------------------------------
SELECT /*+ gather_plan_statistics */ order_id,  order_date,
comments FROM vw_orders,  vw_dual WHERE TRUNC(order_date) =
TO_DATE('12-JAN-13','DD-MON-YY') AND order_id = 1

Plan hash value: 1760841149

------------------------------------------------------------------------------------------------------
| Id  | Operation                    | Name        | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT             |             |      1 |        |      1 |00:00:00.01 |       4 |
|   1 |  NESTED LOOPS                |             |      1 |      9 |      1 |00:00:00.01 |       4 |
|   2 |   FAST DUAL                  |             |      1 |      1 |      1 |00:00:00.01 |       0 |
|   3 |   TABLE ACCESS BY INDEX ROWID| ORDERS      |      1 |      9 |      1 |00:00:00.01 |       4 |
|   4 |    INDEX RANGE SCAN          | ORDER_ID_IX |      1 |      9 |      1 |00:00:00.01 |       3 |
------------------------------------------------------------------------------------------------------

Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------

   1 - SEL$5428C7F1
   2 - SEL$5428C7F1 / DUAL@SEL$3
   3 - SEL$5428C7F1 / ORDERS@SEL$2
   4 - SEL$5428C7F1 / ORDERS@SEL$2

Outline Data
-------------

  /*+
      BEGIN_OUTLINE_DATA
      IGNORE_OPTIM_EMBEDDED_HINTS
      OPTIMIZER_FEATURES_ENABLE('11.2.0.2')
      DB_VERSION('11.2.0.2')
      ALL_ROWS
      OUTLINE_LEAF(@"SEL$5428C7F1")
      MERGE(@"SEL$2")
      MERGE(@"SEL$3")
      OUTLINE(@"SEL$1")
      OUTLINE(@"SEL$2")
      OUTLINE(@"SEL$3")
      INDEX_RS_ASC(@"SEL$5428C7F1" "ORDERS"@"SEL$2" ("ORDERS"."ORDER_ID"))
      LEADING(@"SEL$5428C7F1" "DUAL"@"SEL$3" "ORDERS"@"SEL$2")
      USE_NL(@"SEL$5428C7F1" "ORDERS"@"SEL$2")
      END_OUTLINE_DATA
  */


53 rows selected.

In the output above, the operations are numbered in the execution plan.  These numbers correspond to the numbers identifying the query block and object alias.  These two query blocks and aliases are found in the outline section with a leading hint.  This leading hint can be modified by listing the ORDERS table first as shown below and using the ORDERS alias in the full table scan hint as well.  The optimizer now accepts the hints and executes the query as directed.

SELECT /*+ gather_plan_statistics full(@"SEL$5428C7F1" "ORDERS"@"SEL$2") leading(@"SEL$5428C7F1" "ORDERS"@"SEL$2" "DUAL"@"SEL$3") */
        order_id,
        order_date,
        comments
FROM vw_orders,
        vw_dual
WHERE TRUNC(order_date) = TO_DATE('12-JAN-13','DD-MON-YY')
AND order_id = 1;

PLAN_TABLE_OUTPUT
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
SQL_ID bpy75y38s6phy, child number 0
-------------------------------------
SELECT /*+ gather_plan_statistics full(@"SEL$5428C7F1"
"ORDERS"@"SEL$2") leading(@"SEL$5428C7F1" "ORDERS"@"SEL$2"
"DUAL"@"SEL$3") */    order_id,      order_date,
comments FROM vw_orders,  vw_dual WHERE TRUNC(order_date) =
TO_DATE('12-JAN-13','DD-MON-YY') AND order_id = 1

Plan hash value: 795833099

-----------------------------------------------------------------------------------------
| Id  | Operation            | Name   | Starts | E-Rows | A-Rows |   A-Time   | Buffers |
-----------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT     |        |      1 |        |      0 |00:00:00.01 |     544 |
|   1 |  MERGE JOIN CARTESIAN|        |      1 |     10 |      0 |00:00:00.01 |     544 |
|   2 |   TABLE ACCESS FULL  | ORDERS |      1 |     10 |      0 |00:00:00.01 |     544 |
|   3 |   BUFFER SORT        |        |      0 |      1 |      0 |00:00:00.01 |       0 |
|   4 |    FAST DUAL         |        |      0 |      1 |      0 |00:00:00.01 |       0 |
-----------------------------------------------------------------------------------------

Query Block Name / Object Alias (identified by operation id):
-------------------------------------------------------------

   1 - SEL$5428C7F1
   2 - SEL$5428C7F1 / ORDERS@SEL$2
   4 - SEL$5428C7F1 / DUAL@SEL$3

Outline Data
-------------

  /*+
      BEGIN_OUTLINE_DATA
      IGNORE_OPTIM_EMBEDDED_HINTS
      OPTIMIZER_FEATURES_ENABLE('12.1.0.1')
      DB_VERSION('12.1.0.1')
      ALL_ROWS
      OUTLINE_LEAF(@"SEL$5428C7F1")
      MERGE(@"SEL$2")
      MERGE(@"SEL$3")
      OUTLINE(@"SEL$1")
      OUTLINE(@"SEL$2")
      OUTLINE(@"SEL$3")
      FULL(@"SEL$5428C7F1" "ORDERS"@"SEL$2")
      LEADING(@"SEL$5428C7F1" "ORDERS"@"SEL$2" "DUAL"@"SEL$3")
      USE_MERGE_CARTESIAN(@"SEL$5428C7F1" "DUAL"@"SEL$3")
      END_OUTLINE_DATA
  */


48 rows selected.

Using Linux's Logrotate to Manage Alert and Listener Log Files

If you are running Oracle on Linux, there is a handy little utility called logrotate that can be used to manage those unwieldy alert logs and listener logs.

I recently worked on a client's database, and their alert log had information going back four years. When trying to examine the log with vi, it would take several minutes to open the file due to it's enormity. To help them out, I used this simple little tool to keep the files in check.

As root, I simply created a new file in /etc/logrotate.d on each RAC node called ora_cleanup. The example below is from node 2. This will copy the alert log with a numerical extension and truncate the existing file. It performs this monthly and keeps 13 months worth of log information before deleting old files. The listener log grows much faster, so it is configured to rotate weekly and keep 53 weeks of files. The copytruncate parameter is important for the listener log because the listener process holds an open handle on the existing file.

# alert log
/u01/app/oracle/diag/rdbms/orcl/orcl2/trace/alert_orcl2.log {
monthly
rotate 13
notifempty
missingok
copytruncate
nocreate
}

# listener log
/u01/app/asm/diag/tnslsnr/db102/listener_db102/trace/listener_db102.log {
weekly
rotate 53
notifempty
missingok
copytruncate
nocreate
}

After a few months, this is what the managed files look like in their directories.

$ cd /u01/app/oracle/diag/rdbms/orcl/orcl2/trace
$ ls -arltp alert_orcl2*
-rw-rw---- 1 oracle oinstall 50051579 Oct 10 15:17 alert_orcl2.log.5
-rw-rw---- 1 oracle oinstall   993219 Nov  1 04:02 alert_orcl2.log.4
-rw-rw---- 1 oracle oinstall  1124244 Dec  1 04:02 alert_orcl2.log.3
-rw-rw---- 1 oracle oinstall  1088332 Jan  1 04:02 alert_orcl2.log.2
-rw-rw---- 1 oracle oinstall  2163268 Feb  1 04:02 alert_orcl2.log.1
-rw-rw---- 1 oracle oinstall   279484 Feb  7 09:34 alert_orcl2.log

$ cd /u01/app/asm/diag/tnslsnr/db102/listener_db102/trace
$ ls -arltp listener_db102*
-rw-rw---- 1 oracle oinstall 3857048716 Oct 10 15:28 listener_db102.log.18
-rw-rw---- 1 oracle oinstall  196362891 Oct 18 13:13 listener_db102.log.17
-rw-rw---- 1 oracle oinstall   40298571 Oct 20 04:02 listener_db102.log.16
-rw-rw---- 1 oracle oinstall  178408740 Oct 27 04:02 listener_db102.log.15
-rw-rw---- 1 oracle oinstall  188301497 Nov  3 04:02 listener_db102.log.14
-rw-rw---- 1 oracle oinstall  185073120 Nov 10 04:02 listener_db102.log.13
-rw-rw---- 1 oracle oinstall  178345071 Nov 17 04:02 listener_db102.log.12
-rw-rw---- 1 oracle oinstall  178945914 Nov 24 04:02 listener_db102.log.11
-rw-rw---- 1 oracle oinstall  165829858 Dec  1 04:02 listener_db102.log.10
-rw-rw---- 1 oracle oinstall  179395363 Dec  8 04:02 listener_db102.log.9
-rw-rw---- 1 oracle oinstall  175671704 Dec 15 04:02 listener_db102.log.8
-rw-rw---- 1 oracle oinstall  195136727 Dec 22 04:02 listener_db102.log.7
-rw-rw---- 1 oracle oinstall  195512012 Dec 29 04:02 listener_db102.log.6
-rw-rw---- 1 oracle oinstall  201759600 Jan  5 04:02 listener_db102.log.5
-rw-rw---- 1 oracle oinstall  204589312 Jan 12 04:02 listener_db102.log.4
-rw-rw---- 1 oracle oinstall  213276652 Jan 19 04:02 listener_db102.log.3
-rw-rw---- 1 oracle oinstall  212229296 Jan 26 04:02 listener_db102.log.2
-rw-rw---- 1 oracle oinstall  211244326 Feb  2 04:02 listener_db102.log.1
-rw-rw---- 1 oracle oinstall  159275570 Feb  7 09:35 listener_db102.log

GoldenGate: 12c Conflict Detection and Resolution (CDR)

One of my favorite 12c enhancement for GoldenGate is its new conflict detection resolution (CDR) feature for two way replication.  In previous version, the replicat parameter files had to contain SQLEXEC commands that would query the target table before applying any DML.  This added an additional call to the database and slowed performance.

CDR provides new parameters that simplify the detection and resolution of conflicting data.  Here are some of the optional keywords that can be combined with the new RESOLVECONFLICT parameter.

UPDATEROWEXISTS
UPDATEROWMISSING
INSERTROWEXISTS
INSERTROWMISSING
DELETEROWEXISTS
DELETEROWMISSING

These can be configured to regard one data source the master and always overwrite the other.  They can utilize a timestamp column that determines the "winner" based on the most recent data.  For data such as inventory, the "delta" can be used to make changes to both sites and adjust the value rather than simply replicating it.

GoldenGate requires the "before" images to be captured for all relevant columns in the source database.  Here is how trandata can be configured for all columns in the SCOTT schema in the PDBORCL pluggable database.

GGSCI (oel1.localdomain) 103> ADD SCHEMATRANDATA pdborcl.scott ALLCOLS

2014-01-20 15:21:04  INFO    OGG-01788  SCHEMATRANDATA has been added on schema scott.

2014-01-20 15:21:05  INFO    OGG-01976  SCHEMATRANDATA for scheduling columns has been added on schema scott.

2014-01-20 15:21:05  INFO    OGG-01977  SCHEMATRANDATA for all columns has been added on schema scott.

In addition, the extract parameter files need the GETBEFORECOLS option as a part of the TABLE parameter.  In the example below, the extract will capture all of the columns in the SCOTT.ITEMS table for each update and delete.  The before image of each of these records will be loaded into the trail file.

EXTRACT e1aa
USERIDALIAS ggsadm domain d1
LOGALLSUPCOLS
EXTTRAIL ./dirdat/aa

TABLE pdborcl.scott.emp;
TABLE pdborcl.scott.dept;

SOURCECATALOG pdborcl
TABLE scott.bonus;
TABLE scott.salgrade;
TABLE scott.items,
  GETBEFORECOLS
    (
    ON UPDATE ALL,
    ON DELETE ALL
    );

The final step is to configure the replicat parameter files with the RESOLVECONFLICT paramter.  In the example below, CDR is configured for the SCOTT.ITEMS table.  If the record exists on the target for an update, the DML_TIMESTAMP column will be compared, and the record with the most recent time will "win" for all of the columns except QTY_ON_HAND.  This column will be updated with the difference between the old and new record value on the source using the UPDATEROWEXISTS and USEDELTA keywords.

For example:  If QTY_ON_HAND is updated on both databases in a bi-directional replication setup, the change needs to be reflected appropriately on each system.  Let's say that the value on both system starts out at 22.  The value on DB1 is decremented to 18, while the value on DB2 is incremented to 30.  When the two updates pass each other and get applied on their respective targets, the records cannot simply be replicated as is.  Simple arithmetic must be applied to both sides in order to account for the change in inventory on both sides.  The "delta" for DB1 is -4.  The "delta" for DB2 is +8.  Therefore, the value on DB1 (18) will be incremented by 8 for a new total of 26. The value on DB2 (30) will be decremented by 4 for a new total of 26.  Now, rather than having a "winner", both sources of data will be equal based on the "delta".

The replicat parameter file below is set up for such a situation.

REPLICAT r2aa
USERIDALIAS ggsadm domain d2
ASSUMETARGETDEFS

MAP pdborcl.scott.emp, TARGET scott.emp;
MAP pdborcl.scott.dept, TARGET scott.dept;

SOURCECATALOG pdborcl
MAP scott.bonus, TARGET scott.bonus;
MAP scott.salgrade, TARGET scott.salgrade;
MAP scott.items, TARGET scott.items,
  COMPARECOLS
    (
    ON UPDATE KEYINCLUDING (item_name, qty_on_hand, sales_price, dml_timestamp),
    ON DELETE KEYINCLUDING (item_name, qty_on_hand, sales_price, dml_timestamp)
    ),
  RESOLVECONFLICT (UPDATEROWEXISTS,
                     (delta_combine, USEDELTA, COLS (qty_on_hand)),
                     (DEFAULT, USEMAX (dml_timestamp))),
  RESOLVECONFLICT (INSERTROWEXISTS, (DEFAULT, USEMAX (dml_timestamp))),
  RESOLVECONFLICT (DELETEROWEXISTS, (DEFAULT, OVERWRITE)),
  RESOLVECONFLICT (UPDATEROWMISSING, (DEFAULT, OVERWRITE)),
  RESOLVECONFLICT (DELETEROWMISSING, (DEFAULT, DISCARD))
  ;

GoldenGate: 12c Credential Store Secure Login

The Credential Store is a new 12c security feature in GoldenGate that has been implemented as an autologin wallet in Oracle’s Credential Store Framework.  User IDs and passwords are encrypted in the store and, as a result, an encryption key in the connection string is no longer needed.

The default location of the store is in the ./dircrd directory of the GoldenGate software home.  If you want to change the location, you can edit the ./GLOBALS file with the following CREDENTIALSTORELOCATION parameter.

GGSCI> edit params ./GLOBALS


CREDENTIALSTORELOCATION /home/oracle/ggs/dircrd

You must exit and restart ggsci before proceeding or the file will be created in the default location.

GGSCI> exit

$ ./ggsci

GGSCI> add credentialstore

Credential store created in /home/oracle/ggs/dircrd/.

GGSCI> exit

$ ls /home/oracle/ggs/dircrd
cwallet.sso

Once the credential store has been created, users and password can be added to it.  One of the key features of the store is the use of domains which can be used to logically group login aliases.  The same alias can be defined in different domains with different credentials.  This can be handy when developing and testing in different database environments from the same GoldenGate installation.  The default domain is “Oracle GoldenGate”.

In this example, the user c##ggsadmin is added to the store in the “test” domain.  If the “password” keyword is omitted, GoldenGate will prompt for the password and hide it from the output.

GGSCI> alter credentialstore add user c##ggsadmin, alias ggsadm, domain test
Password:

Credential store in /home/oracle/ggs/dircrd/ altered.

If you want to see the information maintained in the store, you can use the INFO CREDENTIONSTORE command.  If you don’t specify the domain, it will default to “Oracle GoldenGate”.  As you can see below, the default domain is still empty.

GGSCI> info credentialstore

Reading from /home/oracle/ggs/dircrd/:

No information found in credential store.

GGSCI> info credentialstore domain test

Reading from /home/oracle/ggs/dircrd/:

Domain: test
  Alias: ggsadm
  Userid: c##ggsadmin

In older version of GoldenGate, you had to supply the username and password in plain text or encrypted for your login credentials.


Using DBLOGIN at the command line.

GGSCI> DBLOGIN USERID c##ggsadmin@orcl, PASSWORD AACAAAAAAAAAAAJAUEUGODSCVGJEEIUGKJDJTFNDKEJFFFTC AES128, ENCRYPTKEY securekey1

Successfully logged into database CDB$ROOT.

Using credentials in the parameter files.

GGSCI> edit params e1aa


EXTRACT e1aa
USERID c##ggsadmin@orcl, PASSWORD AACAAAAAAAAAAAJAUEUGODSCVGJEEIUGKJDJTFNDKEJFFFTC AES128, ENCRYPTKEY securekey1
LOGALLSUPCOLS
EXTTRAIL ./dirdat/aa

TABLE pdborcl.scott.emp;
TABLE pdborcl.scott.dept;

SOURCECATALOG pdborcl
TABLE scott.bonus;
TABLE scott.salgrade;

In 12c, the credential store secures the information and makes the connection much easier through the use of the alias that was created.


Using DBLOGIN at the command line.

GGSCI> dblogin useridalias ggsadm domain test

Successfully logged into database CDB$ROOT.

Using credentials in the parameter files.

GGSCI> edit params e1aa


EXTRACT e1aa
USERIDALIAS ggsadm domain test
LOGALLSUPCOLS
EXTTRAIL ./dirdat/aa

TABLE pdborcl.scott.emp;
TABLE pdborcl.scott.dept;

SOURCECATALOG pdborcl
TABLE scott.bonus;
TABLE scott.salgrade;