DRBD primary/primary issues

[bago]

Hi all,

I have a proxmox 1.5 cluster over DRBD primary/primary but I only have 1 eth on each server, so DRBD communicate on the same channel the clients use.

I use 1 DRBD device for each VM:
LVM (pve) => n DRBD devices (r<n> / drbd<n>) => n LVM groups (vg<n>) => n Proxmox Storage (storage<n>) => n Proxmox KVM Virtual Machines. (vm<n>)

Sometimes it happen the eth link between the servers to go down for few ms and the result is split-brain and it not automatically recovered.

To recover from splitbrain I always have to go on the "secondary"and call

vgchange -an vg<n>
drbdadm secondary r<n>
drbdadm -- --discard-my-data connect r<n>

then, on the "primary":

drbdadm connect r<n>

So, I found that I have a much better behaviour if I keep the DRBD device in secondary mode on the secondary node (for each vm). When the link between a secondary and primary is temporarily lost DRBD automatically handle the reconnection/resync resulting in no split-brain.

Then, whenever I want to live migrate a VM I change its drbd volume to be primary, live migrate and make secondary the "old" device.

I know you're working on HA for 2.x so I'd like to know if 2.x will still use primary/primary as default or not.

 

[dietmar]

I know you're working on HA for 2.x so I'd like to know if 2.x will still use primary/primary as default or not.

I don't know all feature we will support in future, sorry.

The solution i your case seems to use a separate/reliable network for DRBD.

 

[bago]

I didn't thought about that. You're a genius

Joking aside, I take this as you don't have an answer for me! Maybe proxmox is not the right tool for me right now and I should simply manually script drbd+kvm as I was doing before.