This content has been marked as final. Show 5 replies
Same device names (and types) is a requirement. In Solaris you could get away with creating a new interface aggr1 with dladm on both nodes. These would have the same name on both servers regardless of the underlying device type.1 person found this helpful
So, is there a way to instruct oracle to use different interface names for nodes? Or will vipca ask for matching NIC names just like disk mount names have to match?
I want to avoid reinstallation of RAC because that will disrupt oracle database that has been running on shared oracle home all along.
No, unfortunately the NIC name must be the same across all nodes for example if public interface in node 1 is ipge0 then node 2 must have the same ipge0 however you can change the e1000g0 to ipge0.
Fzheng, I am very interested in your statement "however you can change the e1000g0 to ipge0".
You mean to say there is an OS script that will downgrade ipge0 interfaces to e1000g0?
I got some more information from SysAdmins yesterday. Apparently, when OS was reinstalled on node 2 after root partition corruption, they also patched it up to Solaris 10u9 (which by default created e1000g0-type interfaces). The first node still had 10u7 with ipge0 interfaces. We, the DBAs, were not aware of the upgrade. So, nodeapps cant start saying "no such interface"on node 2.
You are probably referring to script /usr/sbin/e1000g_transition.
Here is the description:
This script allows you to perform an optional driver upgrade
for the Intel PRO/1000 device present on this system from the \"ipge\"
network driver to the Sun standard \"e1000g\" network driver. The
e1000g driver include features such as link aggregation as well as
superior performance in most environments. Moving forward, e1000g
will benefit from new features as they become available, bug fixes,
and performance enhancements. The disadvantage is that some
applications configuration files may require modification to use
e1000g if they reference the ipge network interface.
This script must be run in single user mode.
This script will alter network interfaces and related system
configuration files. Configuration files for 3rd party applications
will NOT be changed. After running this script, applications present
on the system may require maintenance. The system may be in an
unsecured state. The effects of this script may be reversed using
the backout (-b) option."
So, if I understand you correctly, you are proposing to run this this script with the backout (-b) option on node 2, so interface names will match.
Is this correct?
Sorry If my explanation if not really clear, the one I mentioned might be a bit tricky and needs extra work however there are easier options to do this
- use dladm to rename datalink -> your solaris admin must know about it
- create symbolic alias in /dev to the interface
Hope this helps
Thank you so much for the reply.
Unfortunately, this option works in opensolaris only, not SPARC.
Regardless, I sincerely appreciate you both taking the time to answer.