Monday, August 18, 2008

Cannot re-create cluster disk resource

A cluster in User Acceptance Testing was required to test Windows 2003 Service Pack 2 and driver updates. For one reason or another some of the disk resources had been deleted but were required.

When attempting to recreate the disk resources, at the end of the New Resource wizard, Disk parameters, the Disk drop down dialog was not populated. Therefore the resource could not be recreated.



This drop down dialog is populated from :
HKLM\System\CurrentControlSet\ClusDisk\Parameters\Signatures\AvailableDisks

In this case, the disk was still listed under the "signatures" key.

To confirm the disk signature :
1. Locate the disk number from Device Manager. Properties of the disk, volume tab, populate.





2. Use diskpart :
Diskpart
Select disk
Detail disk





The resource can then be manually added from the command line using cluster.exe :

e.g. disk signature 0xA00AB325 and assigned letter K:
cluster . res "Disk F:" /create /group:"Disk Group" /type:"physical disk"
cluster . res "Disk F:" /priv signature=0xA00AB325

The disk could then be brought on line successfully.


Sunday, August 10, 2008

The DHCP Bad Address Saga.

For some time now we've been working on an issue that resulted in "Bad Address" being reported in the DHCP console.

The client was getting an address successfully but this lease did not get registered in the database and hence not in the MMC. When the next client came along it was assigned an address already in use. Conflict detection was enabled and hence BAD_ADDRESS was recorded.

When the Bad Address is noted. The column that should contain the MAC address contains something different. This is the IP Address in hex and reversed. Example :

After lots of troubleshooting it was found that there two machines that had something wrong with their cable. Since removing these two cables from the sub net, the issue has gone away.

With hindsight, we could have got an idea that the issue was a rogue machine earlier. The network covered 3 class C subnets. If we'd split them, we'd have seen the issue was one just one of them. This was one of those suspicions that crossed our mind but didn't follow up.