BUG: Enabilin HA on Started Machine

[robynhub]

I've found that in HA environment, if you enable the ha on a started machine this will be started again on another node.
I've almost broken two machine with this bug.
https://www.google.com/search?hl=it...DwpiTCQ&ved=0CCcQvwUoAQ&q=environment&spell=1
WORKAROUND:
Turn off the VM and THEN add it to HA.

Please confirm this bug.

 

[tom]

as soon as you mark VM/CT as HA its managed by the HA stack. so this is not really a bug, but we should add this to the documentation.

 

[robynhub]

Well, I consider it a bug because it don't check if the vm is running or not. At least a forced stop to the vm will be good. Start a VM twice is equals to damage permanently his filesystem.

 

[tom]

I do not see how this can happen. A VM can only run once.

Pls describe step by step how to reproduce this.

 

[robynhub]

I've created a fresh new install of a 4-node cluster.
I've configured the cluster with fencing with this conf:

<?xml version="1.0"?>
<cluster config_version="37" name="ClusterFO">
  <cman keyfile="/var/lib/pve-cluster/corosync.authkey"/>
  <fencedevices>
    <fencedevice agent="fence_ipmilan" ipaddr="192.168.X.X" login="ipmiusername" name="fence1" passwd="ipmipassword"/>
    <fencedevice agent="fence_ipmilan" ipaddr="192.168.X.X" login="ipmiusername" name="fence2" passwd="ipmipassword"/>
    <fencedevice agent="fence_ipmilan" ipaddr="192.168.X.X" login="ipmiusername" name="fence3" passwd="ipmipassword"/>
    <fencedevice agent="fence_ipmilan" ipaddr="192.168.X.X" login="ipmiusername" name="fence4" passwd="ipmipassword"/>
  </fencedevices>
  <clusternodes>
    <clusternode name="VM1" nodeid="3" votes="1">
      <fence>
        <method name="1">
          <device name="fence1"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="VM2" nodeid="2" votes="1">
      <fence>
        <method name="1">
          <device name="fence2"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="VM3" nodeid="1" votes="1">
      <fence>
        <method name="1">
          <device name="fence3"/>
        </method>
      </fence>
    </clusternode>
    <clusternode name="VM4" nodeid="4" votes="1">
      <fence>
        <method name="1">
          <device name="fence4"/>
        </method>
      </fence>
    </clusternode>
  </clusternodes>
</cluster>

I've joined every node in the fencing domain and I've manually started rgmanager service on all nodes.

After that i've enabled HA to four VM and then i've activated the new config via Web UI.
The system have started the new vms on the other nodes (random choosing?) and the VM which have started previously disappear from the web UI on the nodes where the VM where running.
BUT
on those nodes i've done a "ps ax" to show which VM are really running and i've found that the old vm where never stopped and started again on other nodes.
In other words the kvm process where still running and was not reported by the web ui.

 

[tom]

pls file a bug at https://bugzilla.proxmox.com

to prevent this issue (for now), make sure the VM´s are not running when you add them to HA.