Issue

• Cluster nodes intermittently and unexpectedly fenced at seemingly random intervals.

• Node was fenced and the following messages are seen in the logs:

  • Node 1:

    fenced[15406]: node2 not a cluster member after 0 sec post_fail_delay
    fenced[15406]: fencing node "node2"
    fenced[15406]: fence "node2" success
    clurgmgrd[15444]: <notice> Taking over service service1 from down member node2 
    

  • Node 2:

    qdiskd[12327]: <info> Heuristic: 'ping 10.0.0.1 -c1 -w1' DOWN (1/1)
    qdiskd[1232: <notice> Score insufficient for master operation (0/1; required=1); downgrading
    qdiskd[12327]: <info> Heuristic: 'ping 10.0.0.1 -c1 -w1' UP
    <node rebooted> 
    syslogd 1.4.1: restart.
    

• When the heuristic ping seems evaluating only once even the tko=9. I think it is using the default value of 21 seconds to leave the cluster. Are there other values that I need to change? Or cluster should not function as I observed.

  • Assuming that you set the interval=5 and tko=9, when heuristic all fail, it suppose cycle for about 45 seconds ( interval * tko) before leaving the cluster. But what really happening, it took only 21 seconds then the system will reboot, before the reboot I seen this info on the log file:

    qdiskd[28072]: <info> Heuristic: '/bin/ping 10.0.0.1 -c1 -w1' DOWN (1/1)  
    qdiskd[28072]: <notice> Score insufficient for master operation (0/1; required=1); downgrading </code>