Software virtualization cluster

We can set up a cluster with a virtual layer for operating systems and live migration, why not the same for programs?

Add a virtual layer between OS and programs and it would be possible to live migrate a running program from one cluster node to another, probably with less time and overhead than moving an entire OS, right?

 

Proxmox: upgrade PVE 1.9 to 2.0

Even though I haven’ t tried any beta version I decided to go ahead and upgrade one of my cluster nodes to version 2.0 of Proxmox virtual environment.

The tricky part would most likely be the customized PVE installation I build; two disks running partially software RAID 1 with the system split in several partitions. That might break the upgrade script provided by Proxmox.
Actually, thing went pretty well so far. The biggest error was that I had previously left open a couple of remove assignments through aptitude which removed the pve kernel, so the script halted on that when it couldn’t find it.
Nothing a reversal to lenny in the sources.list and the install of the missing kernel couldn’t fix.

One thing that didn’t match up was not being able to start cman which complained about missing the nodename in cluster.conf. It was present in /etc/pve/cluster.conf but unfortunately the init script pointed to /etc/cluster/cluster.conf.
Adding the name to the latter conf was enough to start cman and enable me to log into the web interface. That one kept looping and timed out while the cluster deamon wasn’t running.
One thing te remember is that the web interface was listening to the default port again so I had to modify the Listen entry in pve.conf under the /etc/apache/sites directories.

Next is copying the backups from the old master node to the new and import them to see if they’ll run. When they do, the old master will be upgraded after confirming that the new master works as planned.

 

*update* 20120524

Just started the upgrade of the second node, so far I only had to do an apt-get install console-tools after an error mentioning that package during the apt-get upgrade round, and an apt-get install vzctl during the second attempt.

 

Proxmox Virtual Environment 1.9 cluster (en)

NL versie

Because it was time to refresh the network at home I started looking at avirtual cluster solution.

Previously I used FreeBSD with jails to seperate the servers and I wanted to keep that functionality. What I wanted op top of that was failover, to be able to move the servers in case of maintenance or problems with the virtual host.

My first experiences with Proxmox VE 1.9 were good, except that I wanted a different partitioning of the host as the standard bare-metal installation uses up the whole disk. Luckily that’s no problem as Proxmox can be installed on top of an existing Debian Lenny 64 bit server.

During the installation of Debian I set up a partial RAID layout with the software RAID option in Linux. Only the virtual environment needs to run on RAID 1, the important part of my data will be secured with a backup. All other data can be downloaded or ripped again (software, my own DVD and CD collection for my media center, etc.) so there’s no real need to use RAID here.
This way I still have redundancy but I won’t lose all my disk space on RAID.

I use the following layout:
/dev/md0, 512 MB RAID 1, ext2, /boot
/dev/sda2 + sdb2, 1GB, swap
/dev/md1, 8GB RAID 1, LVM, /
/dev/md2, 100GB RAID 1, LVM, /var/lib/vz
/dev/sda5 + sdb5, *, ext3, /data/local/01 + 02

Sda5 and sdb5 contain regular data, backup of the virtual machines from the other cluster and the backup of important data from the other cluster. This will be available through NFS shares.
Swap is optional as the servers have plenty of memory installed.

Installation of Debian is done with as few options as possible, after this the sources.list will be modified to include the Proxmox repository and the installation will be finished according to the instructions on the Proxmox Wiki.

What’s left are tweaks to the vzmigrate script to speed up migration and extra network bridges to seperate host traffic from virtual server traffic.

There’s no real high availability yet for VZ containers on shared storage but I’m looking at keeping downtime as short as possible using a combination of Heartbeat, DRBD and rsync.