VMWare IP Fix for Cluster Demos
Novell Cool Solutions: Tip
Digg This -
Posted: 3 May 2002
We've gotten (unofficial) word of a bug in VMware that causes traffic sent out the bridged network to be reflected back to the VM that issued it. This causes NetWare to think there are duplicate IP addresses on the wire when it ARPs at boot time.
The problem comes when your "master" cluster node fails -- its unload script sets the "allow" to "off". At next reboot, IP won't bind to the NIC because it thinks there's another machine with the same address.
Steps to resolution:
- Use ConsoleOne to disable the property page for cluster resource failover policies -- this allows you to edit them via the Other tab instead.
- Use the Other tab to edit the "NCS:CRM Failover Mode" attribute of the master IP address resource. The most significant bit of this SYN_integer will be set. It indicates master IP address. Clear that bit - i.e. save the attribute back to the object with all other bits the same but the top-most bit clear. The top-most bit means this:
- Use ConsoleOne to re-enable the property page for cluster resource failover policies.
// This bit is set to identify a resource as the cluster master // IP resource. It is not currently used by the server side // CRM code. The bit is private to the GUI which uses it to veto // certain operations. E.g. can't delete the master resource. NCS_RESOURCE_MASTER_IP_ADDRESS = 0x80000000;
With the top-most bit clear, you can now edit the master IP just like any resource.
Be careful -- if you're offline, the ConsoleOne GUI will not reconnect to the cluster unless you're online via command-line interface. With the master IP bit clear, you can also delete the resource (which would be bad too).
This isn't a documented or supported activity -- but running clusters on VMware is kinda out there too -- and loads of fun it seems.
Novell Cool Solutions (corporate web communities) are produced by WebWise Solutions. www.webwiseone.com