Part 13 RACing ahead with Oracle on VMware Series covers the first steps towards installing Oracle Clusterware on Redhat 4.2.
A Brief Pep talk: Choosing a perfect Oracle
RAC configuration
We have done the preparation and gone about installing the
OCFS2 and ASM on our RHEL4.2 servers. Oracle RAC is growing in popularity and I
liked this
page at Oracle myself. Here, you can get pre-tested, documented best
practices of installing Oracle on Linux based systems. In addition, some good
recommendations, from this
author on Infostor, will help companies decide what to go for. I like the
quote in which he says that you can also deploy RAC on systems besides Linux
(which I will be documenting here as an ongoing effort to demo RAC on Solaris,
HP-UX and AIX servers as well).
Getting the Oracle RAC software
We have gone through this several times, but just to keep you
in the loop, go to Part
6 to get all the necessary software you need.
Now go ahead and unzip all the files, while logged on as an
"Oracle" user, into /tmp/orasoft directory (you are free to choose the
directory, I name it orasoft to avoid any confusion when looking for the
files).
We are talking about these three files and will go ahead and
unzip them in the /tmp/orasoft folder
Extract the Clusterware package as follows:
# su - oracle
$ cd ~tmp/orasoft
$ unzip 10201_clusterware_linux32.zip
$ unzip 10201_database_linux32.zip
$ unzip 10201_companion_linux32.zip
OK now make sure that you are logged in as user "Oracle," and have unset a
couple of parameters such as ORACLE_HOME, etc. and allow for xserver
connections for access from any console.
After being logged in as root, do the following:
# xhost +
Then logon as Oracle and follow the steps as detailed out in the print
screens:
Welcome screen, click next.
Specify inventory directory and credentials. We choose
/u01/app/oracle/oraInventory and specify "dba" as the operating system username.
Specify home details.
Product specific pre-requisite checks.
Add a new node: The primary node where you are running the
installer will be picked up by the installer and you will have to specify the
additional node(s) by clicking the "Add" button:
We will leave the cluster name as "crs", you can however
choose a name that fits your installation template.
Your Ethernet interfaces are presented to you:
Specify the Oracle CRS location: Remember
we created three files, which we formatted with OCFS.
Specify Voting Disk Location:
You are presented with a summary. If all is satisfactory
click on install:
Installer runs:
When the installation is almost done, you will be prompted to run orainstRoot.sh
and root.sh scripts. You will have to open a new console (do not do it
in the same console as your installer!) beginning from the primary node where
you are carrying the installation and logged in as "root" run the script.
Now go the the /u01/app/oracle/oraInventory directory and run
orainstRoot.sh. You have to do this on ALL NODES!
As you can see, it changes the permissions of the oraInventory and assigns
DBA the group name of the same directory.
There are many discussions about miscounts, specifically to the VMware
environment (like VMware Server OR Workstation), where the overhead might just
not be supportive enough to allow for a full and robust functioning of our RAC.
This is also necessary when running DBCA in order to install our database. Not
doing so might result in an ora-03113
error.
What actually is the problem then? In VMware you might encounter slower disk
responsiveness (depending on how fast and what configuration you have for your
disks, I had success on an old server with RAID0 and problems with a RAID 5/SAN
disk) . However, this does not have to be the case for VMware RAC testbed, it
can also occur in a poorly selected hardware configuration. The CSS miscount
has a default value of 60 and happens to be a bit too small for the CSS to miss
the heartbeat and eventually ending up evicting the node and thus throwing the
ora-03113 error. Beginning with the Oracle 10gR1 series, the algorithm has
changed to allow a lengthier timeout. I have had success with 300, but you can
go as high as 600. So now, the question arises of how to do it. If you havent
done it, do it now by doing:
Query your nodes:
$ORA_CRS_HOME/bin/crsctl get css misscount
60
Query again:
$ORA_CRS_HOME/bin/crsctl get css misscount
360
Now we go about installing the root.sh script:
You do see that the clscfg complains that the miscount must be below 360 and
that is when I went looking for the trade off, trying to set it just below 300.
Moreover, after doing so you will need to run this root.sh on ALL NODES!
Despite the error of that ONS (Notification Services daemon), I was able to
successfully conclude the installation. The last node runs the VIPCA (VIP
Configuration Agent).
As you can see, the configuration starts:
And completes successfully.
Doing a quick check on the cluster nodes:
And also checking if all the daemons will start automatically:
Conclusion:
In this
article, we took our first steps towards installing the Oracle Clusterware and
as you saw yourself, the installation went fine. That was due to the
painstaking and yet rewarding preparatory efforts that we put in the beginning
of our series. In our next article, we will go ahead and install Oracle
software and the Companion CD. Then we will take on our DBCA (Database Creation
Assistant). In our Oracle RAC administration series, we will go ahead and
administer the clusterware components. Although I had started this series to do
a Windows and Linux RAC installation, I am tempted to look at HP-UX, AIX and
Solaris too, hopefully my VMware server will get out of beta and I can get on
with these configurations as well.
»
See All Articles by Columnist Tarry Singh