Tuning the Failover Interval
This section describes how to tune configurable parameters associated with the cluqourumd daemon. These parameters control the amount of time that a healthy cluster member will grant a failed cluster member prior to concluding that a failure has occurred. After this time interval has elapsed, the healthy cluster member will power cycle the failed cluster member (depending on the type of power switch in use) and resume services previously running on the failed member.
There are several parameters governing the amount of time prior to initiating failover, including the following:
Table A-5. Failover Interval Parameters in cluquorumd
| Name | Default (sec.) | Description |
|---|---|---|
| pingInterval | 2 | The frequency at which cluquorumd updates its on-disk state information and reads in the state of the other cluster member. |
| sameTimeNetdown | 7 | The number of intervals that must elapse before concluding a cluster member has failed when the cluhbd heartbeat daemon is unable to communicate with the other cluster member |
| sameTimeNetup | 12 | The number of intervals that must elapse before concluding a cluster member to have failed, when the cluhbd heartbeat daemon is able to communicate with the other cluster member. The value of this parameter must be greater than the sameTimeNetdown parameter. |
For example, suppose one cluster member has a hardware fault and ceases to function. In this case, both cluquorumd and cluhbd will agree that the other cluster member has failed. After a period of (pingInterval * sameTimeNetdown), (which by default is 14 seconds total), the failover will commence.
In order to tune this failover interval, the cludb utility is used. For example, to modify the failover interval down to a duration of 10 seconds, rather than the default of 14 seconds, the value of the sameTimeNetdown parameter should be set to 5 as follows:
cludb -p cluquorumd%sameTimeNetdown 5 |
 | Note |
|---|---|
Caution should be used when tuning these parameters. If a specified failover interval is too small, there is risk of incorrectly concluding that a member has failed during spikes in activity. Also note that if the power switch type "watchdog" is in use, it is necessary that the watchdog expiration interval be less than the failover interval. It is recommended that the watchdog interval be set to approximately 2/3 the duration of the failover interval. |