Oracle RAC One Node revisited – 11.2.0.2

Since we published the RAC book, Oracle has released patchset 11.2.0.2. Amongst other things, this improved the RAC One Node option, exactly the way we expected.

How it was – the base release

A quick recap on the product as it was in 11.2.0.1: RAC One Node is part of Oracle Enterprise Edition, any other software editions are explicitly not allowed. Another restriction exists for 3^rd party Clusterware: it’s not allowed to use one. RAC One Node is a hybrid between full blown RAC and the active/passive cluster. The option uses Grid Infrastructure for cluster management and storage provisioning via ASM. The RAC One instance starts its life as a RAC database, limited to only one cluster node. It only ever runs on one node, but that node can change. It is strongly recommended to create a service for that RAC database. Utilities such as raconeinit provide a text based command line interface to transform that database to a “RAC One Node”-instance. In the process, the administrator can elect which nodes should be allowed to run the instance. The “omotion” utilities allowed the DBA to move the RAC One Node instance from the current node to another one. Optionally a time threshold could be set after which all ongoing transactions were to move to the new node. This feature required TAF or FAN to be set up correctly. The raconestatus utility allowed you to view the status of your RAC One Node instances. Conversion to full RAC was made possible by the racone2rac utility.

If you were after a Data Guard setup you’d be disappointed: that wasn’t supported. (The good news is: from 11.2.0.2 onwards, Data Guard can be used)

So all in all, that seemed a little premature. A patch to be downloaded and applied, no Data Guard and a new set of utilities are not really user friendly. Plus, initially this patch was available for Linux only. But at least a MOS note (which I didn’t find until after having finished writing this!) exists, RAC One — Changes in 11.2.0.2 [ID 1232802.1]

Changes

Instead of having to apply patch 9004119 to your environment, RAC One Node is available “out of the box” with 11.2.0.2. Sadly, the Oracle RAC One Node manual has not been updated, and searches on Metalink reveal no new information. One interesting piece of information: the patch for RAC One Node is listed as “undocumented Oracle Server” section.

The creation of a RAC One Node instance has been greatly simplified-dbca has added support for it, both from the command line for silent installations as well as the interactive GUI. Consider these options for dbca.

$ dbca -help
dbca  [-silent | -progressOnly | -customCreate] {<command> <options> }  |
 { [<command> [options] ] -responseFile  <response file > }
 [-continueOnNonFatalErrors <true | false>]
Please refer to the manual for details.
You can enter one of the following command:

Create a database by specifying the following parameters:
-createDatabase
 -templateName <name of an existing  template>
 [-cloneTemplate]
 -gdbName <global database name>
 [-RACOneNode
 -RACOneNodeServiceName  <Service name for the service to be
 created for RAC One Node database.>]
 [-policyManaged | -adminManaged <Policy managed or Admin managed
 Database, default is Admin managed
 database>]
 [-createServerPool <To create ServerPool which will be used by the
 database to be created>]
 [-force <To create serverpool by force when adequate free servers
 are not available. This may affect already running database>]
 -serverPoolName <One serverPool Name in case of create server pool and
 comma separated list of serverPool name in case of
 use serverpool>
 -[cardinality <Specify cardinality for new serverPool to be created,
 default is the number of qualified nodes>]
 [-sid <database system identifier prefix>]
...

With RAC One Node you will most likely end up with a policy managed database in the end, I can’t see how an admin managed database made sense.

The srvctl command line tool has been improved to deal with the RAC One node. The most important operations are to add, remove, config and status. The nice thing about dbca is that it actually registers the database in the OCR. Immediately after the installation, you see this status information:

$ srvctl status database -d rontest
Instance rontest_1 is running on node node2
Online relocation: INACTIVE

$ srvctl config database -d rontest
Database unique name: rontest
Database name:
Oracle home: /data/oracle/product/11.2.0.2
Oracle user: oracle
Spfile: + DATA/rontest/spfilerontest.ora
Domain:
Start options: open
Stop options: immediate
Database role: PRIMARY
Management policy: AUTOMATIC
Server pools: rontest
Database instances:
Disk Groups: DATA
Mount point paths:
Services: rontestsrv
Type: RACOneNode
Online relocation timeout: 30
Instance name prefix: rontest
Candidate servers: node2,node3
Database is administrator managed

Note that the instance_name, although the instance is administrator managed, changed to $ORACLE_SID_1. Relocating works now with the srvctl relocate database command as in this example:

$ srvctl relocate database -d rontest -n node2

You’ll get feedback about this in the output of the “status” command:

$ srvctl status database -d rontest
Instance rontest_1 is running on node node2
Online relocation: ACTIVE
Source instance: rontest_1 on node2
Destination instance: rontest_2 on node3

After the command completed, check the status again:

srvctl status database -d rontest
Instance rontest_2 is running on node node2
Online relocation: INACTIVE

The important difference between an admin managed database and a policy managed database is that you are responsible for undo tablespaces. If you don’t create and configure undo tablespaces, the relocate command will fail:

$ srvctl relocate database –d rontest -n node3                       <
PRCD-1222 : Online relocation of database rontest failed but database was restored to its original state
PRCD-1129 : Failed to start instance rontest_2 for database rontest
PRCR-1064 : Failed to start resource ora.rontest.db on node node3
CRS-5017: The resource action "ora.rontest.db start" encountered the following error:
ORA-01092: ORACLE instance terminated. Disconnection forced
ORA-30013: undo tablespace 'UNDOTBS1' is currently in use
Process ID: 1587
Session ID: 35 Serial number: 1

CRS-2674: Start of 'ora.rontest.db' on 'node3' failed

In this case, the database runs on the same node. Check the ORACLE_SID (rontest_2 in my case) and modify the initialisation parameter.

SQL> select tablespace_name from dba_data_files where tablespace_name like ‘%UNDO%’;

TABLESPACE_NAME
------------------------------
UNDOTBS1
UNDOTBS2

So the tablespace was there, but the initialisation parameter was wrong! Let’s correct this:

SQL> alter system set undo_tablespace='UNDOTBS1' sid='rontest_1';

System altered.

SQL> alter system set undo_tablespace='UNDOTBS2' sid='rontest_2';

System altered.

Now the relocate will succeed.

To wrap this article up, the srvctl convert database command will convert between single instance, RAC One Node and RAC databases.

Advertisement