moving systems on KVM + HA

I would like to know what is the recommended setup for high availability with KVM + Cloudmin Pro.

In particular:

1- How to move systems?

This seems to be missing from the online documentation and also from the help system on Cloudmin ("Failed to read help file /usr/libexec/webmin/server-manager/help/smove.html")

2- What does "move" actually do?

3- How to ensure that in case one server fails, the VMs can be started on a different one? External storage for the VM images? Do you recomend iSCSI or NFS?

4- Does replication imply common storage?

http://www.virtualmin.com/documentation/cloudmin/repl

Status: 
Closed (fixed)

Comments

  1. Thanks for pointing this out - the link was to the wrong page.

  2. The move feature takes a VM from one host system and moves it to another, by copying across disk images and config files as necessary.

  3. I recommend external storage for this case - either NFS or iSCSI is supported by Cloudmin, in which case a move can be done without the need to copy any data.

  4. Yes. What you really want is not replication (which is for the Cloudmin master), but failover as documented on http://www.virtualmin.com/documentation/cloudmin/vm/failover

Thanks Jamie. What we want is in fact failover. Do we install Cloudmin on the failover KVM host as well or only on the main host?

You only need it on the master system. That said, failover won't work if the Cloudmin master is also a KVM host, and that system goes down.

So how do you configure a Cloudmin Pro cluster in a way that the failure of a single machine (any of the machines) is always properly handled?

In a failover setup, you would normally want to separate Cloudmin from the KVM systems.

Cloudmin would go onto a server not running VM's, and then you would have two or more separate servers for acting as KVM hosts.

Cloudmin would then monitor those KVM servers, and if either goes down, can bring the other online.

How to configure that is described here:

http://www.virtualmin.com/documentation/cloudmin/vm/failover

Now, although the Cloudmin server would not be hosting VM's, there's also a procedure for handling failover on the primary Cloudmin system. That is described here:

http://www.virtualmin.com/documentation/cloudmin/repl

I don't mean to confuse the matter, but to provide you with an additional option -- there is another supported way of handling failover of the KVM hosts, in case you don't have network storage, as needed by the above option. That alternative option, which uses DNS records, is here:

http://www.virtualmin.com/documentation/cloudmin/vm/roundrobin

Andrey,

If what we have in mind moves forward we will have shared storage on a unit that handles its own redundance.

So, you suggest:

  • a pair of Cloudmin Pro servers using replication
  • a set of KVM servers to run the machines with failover (CentOS 6)

Is this setup production-ready from your side and fail scenarios fully tested?

I mean, it is a big investment to "reserve" 2 servers just for running Cloudmin without VMs, so I want to make sure this is rock solid.

Any comments?

Yes, that would be the recommended setup for maximum reliability.

We are working on support for a layout in which the Cloudmin master can also be a KVM host, and failover both the master duties and VM hosting in the case of a machine failure.

what are the minimum requirements for the "pair of Cloudmin Pro servers using replication", in order to manage something like 10 physical KVM hosts? We might do this with small appliances.

I mean the minimum HARDWARE requirements, for the double Cloudmin Pro pair.

Cloudmin pro itself has very small resource requirements - maybe 1 CPU core and 1 GB of RAM, plus enough disk space to store all VM images.

The host systems need enough CPU, RAM and disk to cover the combined usage of all running VMs.

For now this information is enough. I'm closing this issue. Will reopen in the future if we need support for such a situation.