Watchdog Timeouts Seen With Heavy Workloads and Maximum Memory Configurations (CR 7083001)

With certain unusual heavy workloads, especially where a highly processor-intensive workload is bound to CPU 0, the host might appear to suddenly reset back to OBP without any sign of there having been a crash or a panic. The Oracle ILOM event log contains a host watchdog expired entry. The problem is more likely to occur on systems with full memory configurations.

If you see this sort of sudden reset, display the SP event log using this command from the Oracle ILOM CLI:

-> show /SP/logs/event/list

If you see an entry labeled Host watchdog expired, you are experiencing this issue.

Workaround: Contact your authorized service provider to see if a fix is available.

There are two ways you can work around this issue:

• You can extend the watchdog period by adding this entry to the Oracle Solaris /etc/system file:

  set watchdog_timeout = 60000

  This extends the watchdog timeout period to 1 minute (60000 milliseconds).

• In extreme cases, you can disable the watchdog timeout altogether by adding this entry to the /etc/system file:

  set watchdog_enabled = 0

Whenever you modify the /etc/system file you must reboot the system for the changes to take effect.

If you do not want to reboot the system immediately after editing /etc/system, you can apply an additional temporary workaround that takes effect immediately. To apply this temporary workaround, as root type.

# psrset -c -F 0

This command creates a temporary processor set containing only CPU 0, preventing application workloads from using this processor and preventing this issue from occurring.

This temporary processor set is removed on the next operating system reboot, at which point the /etc/system file workaround takes effect.