[Linux-cluster] Halt nodes in cluster with cable disconnect
linux at alteeve.com
Tue Jan 24 21:09:55 UTC 2012
On 01/24/2012 03:57 PM, Miguel Angel Guerrero wrote:
> Hi i'm trying to setup a centos cluster with two nodes with cman, drbd,
> gfs2 and i'm using ipmi for fencing. DRBD is set up between the nodes
> using a dedicated interface. So, when I unplug the drbd network cable,
> both nodes power off immediatly (i tried using crossover cable and both
> nodes connected to a switch, but both scenarios fail), and the logs
> doesn't seem to show something useful. In a previous thread on this
> list, it is recommended to deactivate ACPID daemon, even at BIOS level,
> but I'm still having troubles.
> If I simulate a physical disconnection with ifdown command in some node,
> this node reboots with no hassle, but unpluging the cable kills both
> nodes. I think the first scenario is correct, but the second one is not
> what I expect.
> Thanks for your help the next are my cluster.conf
This is likely caused by both nodes getting their fence calls off before
one of them dies.
How do you have DRBD configured? Specifically, what fence handler are
you using? If you're interested in testing, I have rewritten lon's
obliterate-peer.sh and added explicit delays to help resolve this exact
Alternatively, add a 'sleep 10' or similar to one of your existing fence
handlers and you should find that the node with the delay consistently
loses while the other node remains up.
E-Mail: digimer at alteeve.com
Papers and Projects: https://alteeve.com
More information about the Linux-cluster