[High availability] VM not moved

[Mirage]

Hi,

I set up a cluster of 2 PVE 2.0 servers.

* I have an iSCSI shared storage between the 2 nodes.
* Fencing is configured and working.
* I can use fence_node on one node to reboot the other node.
* ACPI is disabled at kernel boot (acpi=off)
* There is no option to disable ACPI in bios, but nmi button is enabled (servers are Dell PE R710)


The problem occurs with running VMs not being moved to the other node when the host is fenced :
1. VM running on node 1
2. node 2 issue 'fence_node node1'
3. node 1 reset (looks like power outlet are removed, no services are properly stopped), VM is keept on node1
4. ASAP node 1 is back alive, VM is rebooted on node 2.

cluser.conf :

<?xml version="1.0"?>
<cluster config_version="16" name="PROXMOX">
  <cman keyfile="/var/lib/pve-cluster/corosync.authkey"/>
  <fencedevices>
    <fencedevice agent="fence_drac5" cmd_prompt="admin1->" ipaddr="172.20.70.123" login="fence" name="PROX01" passwd="password" secure="1"/>
    <fencedevice agent="fence_drac5" cmd_prompt="admin1->" ipaddr="172.20.70.124" login="fence" name="PROX02" passwd="password" secure="1"/>
  </fencedevices>
  <clusternodes>
    <clusternode name="PROX01" nodeid="1" votes="1">
      <fence>
        <method name="1">
          <device name="PROX01"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="PROX02" nodeid="2" votes="1">
      <fence>
        <method name="1">
          <device name="PROX02"/>
        </method>
      </fence>
    </clusternode>
  </clusternodes>
  <rm>
    <pvevm autostart="1" vmid="101"/>
  </rm>
</cluster>

What makes my VM not moving the the node still alive ?

 

[dietmar]

What makes my VM not moving the the node still alive ?

You only have 2 nodes, so you loose quorum as soon as one node fails.

 

[Mirage]

Thank you Dietmar,

Can you tell me what is the mecanism that choose on which node a VM is moved when its hosting node is failing ?
Is the new hosting node choose regarding its load? its number of active VMs? Can it be user-defined ?

 

[dietmar]

Can you tell me what is the mecanism that choose on which node a VM is moved when its hosting node is failing ?

You have only two node, so there is no choice. (I guess I do not really understand your question)

 

[Mirage]

I'm sorry I was not clear enough.

As I can't have HA the way I want with my 2-nodes setup, I've planned to add a 3rd node soon. That way all my VMs will be moved to another node if their hosting node is failing (Am I correct on this point ?)

My question was : In a more-than-2-nodes setup, when one is failing, how does PVE choose on which node a VM shoul go ?

 

[dietmar]

My question was : In a more-than-2-nodes setup, when one is failing, how does PVE choose on which node a VM shoul go ?

You can define failover domains, as described in 'man rgmanager'