12.5.3.2. Procedure – Configure additional Heartbeat parameters

1. Set timers for Heartbeat in the Timers section.

   

   Figure 12.20. Configuring timing parameters

   The following timers can be defined. Give all the values in seconds.

   keepalive

   The interval in seconds between subsequent Heartbeat signals.

   warning time

   The interval in seconds after which the slave node gives warning of the dead master node.

   dead time

   The interval in seconds after which the node is assumed to be dead.

   initial dead time

   First deadtime after system reboot in seconds.

   Tip
   It may take a while in some machines to have a fully functional network after a reboot. It is recommended to set the dead time twice as long as the warning time, and the initial deadtime twice as long as the dead time. For example, if the warning time is 20 seconds, the optimal dead time is 40 seconds and the optimal initial dead time is 80 seconds.

2. Set other Heartbeat options in the Misc subwindow.

   

   Figure 12.21. Configuring other Heartbeat parameters

   Select the log target, watchdog device and decide about service reacquiry in case of master node recovery.

   log facility

   This option determines the target device for logging.

   watchdog

   This option determines which watchdog device is used, if any. For example, /dev/watchdog.

   Watchdogs can monitor the heartbeat messages of a node and reboot the node if it fails.

   nice fallback

   This option prevents the master node from reacquiring cluster resources after a failover in case the master node gets functional again. Enabling nice_failback can cause problems in certain situations. Suppose there is a two-node cluster with a master (Node_A) and a slave (Node_B) node. If Node_A fails, Node_B acquires its resources and uses a STONITH device to power off the Node_A. When Node_A recovers, the resources remain on Node_B if nice_failback is enabled. However, if now Node_B seems to fail, Node_A can power off Node_B only if two STONITH devices are intalled to the system (one to power off Node_A, one to power off Node_B).

   Tip
   It is recommended to enable nice_failback, but after a takeover restore the original master-slave node hierarchy at the earliest possible time (for example, during the next maintenance break).