How to change IP on Oracle VM Server?

Question

Problem statement:  Need to move OVMS1 (OVM Server 1) to an IP the 10.40.40.X subnet;
~~~
environment: 2 OVS servers 2.1.2 with identical hardware (18gRAM/4cores) with shared SAN storage via OCFS2 currently validated in High Availability Config;
~~~
OVMS1 is on a 192.168.0.X IP & is setup as Server Pool Master, Util Server, & VM Server.
OVMS2 is on a 10.40.40.X IP & is setup as VM Server.
Note: currently there is 2 way bridging between these subnets and we have created many machines and had no issues...
Note: DNS is setup for the hosts and the OCFS cluster configs and such were made using fully qualified domain names (believe there are no hard-coded IPs).
Note: OVMS1 was the original server where the OCFS2 was configured and is currently the pool master & util server...
~~~
Proposed solution1: 'Change IP': Not sure yet but with no other input proposed steps:
- 1.a: make sure no VMs are running on OVMS1 & physically disconnect it from the network,
- 1.b: change the IP on the server ({color:#ff0000}*ifconfig on peth0?*{color}) via ILO on the console and shut it down.
- 1.c: change the DNS entries on the DNS server and change the cabling to be on the 10.40.40.X subnet;
- 1.d: restart OVMS1.
~~~
Proposed solution2: 'Re-install OVMS': Not sure yet but with no other input proposed steps:
- 2.a: make sure no VMs are running on OVMS1, shut down OVMS1, & physically disconnect it from the netork,
- 2.b: +{color:#ff0000}Question: Do I need to unregister OVMS1 if I am going to do an identical install with just a new IP?{color}+
- 2.b: physically change the cable to the 10.40.40.X subnet & reinstall OVS 2.1.2 with the new IP and resetup OCFS2
- 2.c: start OVMS1.
~~~
In theory Proposed solution 1 is cleanest, assuming DNS & no hard coding of IPs, as OVMS1 is already registered and it's agents are configured; Solution2 gives me some concerns about how the server/host are registred and whether the OVM Manager holds the entires or if they are local to the server... Not sure
~~~
Any ideas or input to this would be appreciated as it is critical that we change the IP of OVMS1...
Edited by: user8309620 on Nov 24, 2008 10:20 AM
Edited by: user8309620 on Nov 24, 2008 10:20 AM

Answer 1

First, Avi, my hat is off to you for always trying to help everyone.
~~~
This was a little more trouble than planned (likely due to being HA). After a bunch of tinkering, I am now back up with both servers now on same subnet in HA config... After IP change, OCFS stopped working, then VM Manager would not manage the VMs properly & kept losing connection with the VM Servers, then trying to re-enable HA kept overwriting the cluster files stopping the servers from coming back up...
~
Something that may be relevant is that my OVMS1 is also my initial node in my OCFS2 cluster (for HA) which I configured as OCFS2 over SAN via the documentation: "Oracle® VM Server User’s Guide Release 2.1 E10898-03:9.2.2 Creating an OCFS2 Cluster for HA"
~
I will try to summarize what worked.
~~~
Note: In retrospect, I would un-configure the HA before doing anything then re-enable when done. It looks like enabling HA overwrites the cluster.conf file so make backups of it as you go through the process... You can alternatively just leave HA enabled and 'test' when done but I preferred to re-configure it after the changes...
~~
1) As Avi said, change /etc/sysconfig/network-scripts/ifcfg-eth0:
2) If your gateway changes, then update /etc/sysconfig/network (my machine didn't work without this).
3) modify your /etc/ocfs2/cluster.conf to reflect the IPs. I recommend you reboot each server multiple times to make sure the OCFS works and auto mounts on both servers after reboots. Note that I had to reboot many times and restart/reload/reconfigure OCFS many times... More details at the bottom of this post but I hit many issues/errors including:
- connect attempt to node failed with errno -117
- connect attempt to node failed with errno -22
- Unable to stop cluster as heartbeat region still active
- Transport endpoint is not connected
4) Test your OV Manager... I was getting weird errors like I couldn't delete a VM server from the Pool & VMs wouldn't come up right on both servers (using preferred servers). I thought that the old IP may be cached somewhere so I restarted the VM Manager:
/sbin/service oc4j stop
/sbin/service oc4j start
5) OV Manager was still messed up and not able to load machines properly so I changed the Master from SV1 to SV2 using Avi's advice in the thread:
Re: Changing the Server Pool Master
6) things were working fine until I edited the pool and re-Enabled High Availability. Upon reboot of the servers the OCFS2 /OVS shared cluster didn't come up. That is where I found the /etc/ocfs2/cluster.conf was being overwritten. Replacing the over-written cluster.conf with the one that got me working in step 3 above...
7) Now I had a working infrastructure but feared that my OVMS1 was still primary for OCFS2 and now OVMS2 was now the Pool Master... I decided that I wanted OVMS1 to be primary for both so I repeated steps 5) & 6) above to re-make OVMS1 the Primary in the pool...
That is all I am up for now. I have tested that VMs come up on both OVMS's with both manual and auto preferred server settings (same VMs restarted on both servers).
Note that I have NOT gone in and tested auto-fail-over of the HA infrastructure to make sure that VMs auto-migrate or auto-restart on the other OVMS. I will take that up again later.
~~~~~~~~~~~~~~~~
Some references that helped me that may help you:
Network settings: Oracle VM Network Setting
when ocfs not working: http://oss.oracle.com/pipermail/ocfs2-users/2007-April/001616.html
more ocfs: http://oss.oracle.com/pipermail/ocfs2-users/2007-April/001591.html
more ocfs 'Transport endpoint is not connected': Re: Error Message:  mount.ocfs2:  Transport endpoint is not connected while
more ocfs: Add servers to a cluster
more ocfs http://www.mail-archive.com/[email protected]/msg02522.html
more ocfs: http://oss.oracle.com/projects/ocfs2/dist/documentation/ocfs2_faq.html