Cluster Question

[drwho]

I have set up two nodes with a third PC providing shared iSCSI using OpenFiler.

Generally I am very impressed with Proxmox, having looked at Virtualbox, Xen, Xenserver, Oracle VM, "raw" KVM. It very nearly provides exactly what I want in a very neat package.

I would like HA, but understand that is on its way.

Just a couple of questions:

I'm finding that migrating a VM to the other node is slow. I didn't do any timings, but it didn't seem much faster with shared storage than with VM hosted on local node.

Also, I tried pulling the network on a node and then migrating a VM from that node to the remaining "live" node; my thinking being that if the VM is hosted on shared storage then that should work. But it failed because the original host node was unavailable.

I would be interested to know if these observations are as expected because I assumed the cluster would provide a benifit in each case...


Many thanks,

 

[udo]

I have set up two nodes with a third PC providing shared iSCSI using OpenFiler.

Generally I am very impressed with Proxmox, having looked at Virtualbox, Xen, Xenserver, Oracle VM, "raw" KVM. It very nearly provides exactly what I want in a very neat package.

I would like HA, but understand that is on its way.

Just a couple of questions:

I'm finding that migrating a VM to the other node is slow. I didn't do any timings, but it didn't seem much faster with shared storage than with VM hosted on local node.

Also, I tried pulling the network on a node and then migrating a VM from that node to the remaining "live" node; my thinking being that if the VM is hosted on shared storage then that should work. But it failed because the original host node was unavailable.

I would be interested to know if these observations are as expected because I assumed the cluster would provide a benifit in each case...


Many thanks,

Hi,
at this time the cluster is not ha - it stands more for spread the vms over diffrent nodes.
If you want to start the nodes also on survived nodes, you murt recreate the config-file (for kvm they are below /etc/qemu-server). Just rsync them to a backup-place on the other node and you are get your vm in a short time back.
For my case, the migrating between two nodes are not slow. It's depends on your externel storage (e.g. fast FC-Raids...) and how busy your VM are. With very busy VM it takes some times longer to sync the memory.

Udo

 

[drwho]

I am curious - when I make a cluster it spends some time syncing - in which case what is it that gets synchronised?

Can I do anything with DRBD to improve resiliance?

What I am trying to achieve is the ability to manually restart a VM on the remaining server with the most recent state in the event of a hardware failure.

Is anything known on the timeline for HA? - maybe I should wait...

 

[udo]

I am curious - when I make a cluster it spends some time syncing - in which case what is it that gets synchronised?

The most data of syncing are the templates (openvz) and iso-images.

Can I do anything with DRBD to improve resiliance?

Of course, your data are more safe, like on a single-controller external raid. But you must (at this time) manualy sync your config.

What I am trying to achieve is the ability to manually restart a VM on the remaining server with the most recent state in the event of a hardware failure.

Is anything known on the timeline for HA? - maybe I should wait...

Yes there is a timeline respectivily a roadmap http://pve.proxmox.com/wiki/Roadmap. The developer wrote about a first beta in this summer (if i remember right).

Udo