[virt-tools-list] question about clustering

[Josip Deanovic]

On Sunday 2011-02-20, Kenneth Armstrong wrote:
> Ok, so rebooting from within the guest OS itself might not be bad,
> since the PID remains the same?  So RHCS is basically checking for the
> PID on a vm's instance then?

AFAIK, RHCS is checking the state of a VM using virsh.

So if you poweroff/power on your vm there is a chance for rgmanager to get 
the status while your VM is "shut off", although it's a very short time.

About the PID of a VM...
Generally rgmanager is using a good logic. In case you add your VM to 
cluster configuration, start VM manually using virsh and then you try to 
start it through clusvcadm, rgamnager will check all the nodes and if it 
find out VM is already running, rgmanager will not start another instance of 
your VM. It will just print out warning and inform you that your VM is now 
started on some specific node.

Because RHCS is using virsh to check the state of VMs, there is also a 
danger of restarting libvirtd through nodes in a cluster.
Documentation of rgmanager states that if you want to restart (or stop) 
libvirtd daemon, you must freeze VMs on that node.
Freezing of VMs doesn't mean VMs will be blocked, it means cluster will stop 
checking the state of these VMs (domains) until your unfreeze them.

Check these links for details about rgmanager:
http://sourceware.org/cluster/wiki/VirtualMachineBehaviors
http://sourceware.org/cluster/wiki/RGManager


P.S.
This is now a bit out of the scope of thid mailinglist.


-- 
Josip Deanovic