Use Uninterruptible Power Supplies (UPSs) with Compaq ProLiant Clusters to minimize system downtime in the event of power loss. Cable the UPSs to enable the cluster to be shutdown cleanly before the UPS battery backup is exhausted. Proper cabling of UPSs is described in the ProLiant NonStop Clusters for SCO UnixWare 7 Reference Guide.
UPS-initiated shutdown can minimize data loss as well as improve cluster reboot speed when power returns. A monitoring process running within the cluster provides UPS-initiated shutdown. A simple configuration file controls this monitoring process.
Configure UPS-initiated shutdown by modifying the OS_SHUTDOWN_DELAY
, UPS_SERIAL_PORT
, and UPS_LOG_FILE
parameters
in the /opt/compaq/etc/nscupsd.cfg
configuration file.
OS_SHUTDOWN_DELAY
OS_SHUTDOWN_DELAY=15
indicates that a cluster-wide shutdown is initiated when the UPS has 15 minutes of battery backup power remaining. Measure the time required for a clean shutdown of
the cluster under peak operating conditions to ensure that the shutdown time is adequate.
UPS_LOG_FILE
var/spool/compaq/nscupsd.log
and should not be modified.
UPS_SERIAL_PORT
Colon-separated serial ports create a pair of UPSs in which both of the UPSs must signal that they are low on power before a cluster shuts down. Such a group of UPSs is referred to as a logical UPS. A logical UPS identifies UPSs that are grouped together to provide redundant power to a cluster. The drain of a single UPS within a logical UPS should not result in the loss of any key cluster resources.
Semicolon-separated serial ports identify a list of UPSs or a list of logical UPSs. A low-power indication by any of those UPSs results in a cluster-wide shutdown. This parameter is used when a cluster spans multiple power domains and the loss of any one domain results in a cluster-wide shutdown to protect the cluster.
NOTE: A serial connection to a UPS is used in determining shutdown if the node with the serial port is an active member in the cluster.
A two-node cluster or single-rack cluster with two UPSs, the cluster
should shut down when both UPSs are low on power.
To have both UPSs signal that they are low before shutting down the cluster, combine both UPSs into a single logical UPS. This results in a UPS_SERIAL_PORT
configuration of:
UPS_SERIAL_PORT="/dev/tty00.1:/dev/tty00.2"
/dev/tty00.1
is the device identifier for the node 1 serial port tied to UPS1 and
/dev/tty00.2
is the device identifier for the node 2 serial port tied to UPS2.
In a multiple-rack cluster that uses fully redundant
power in all key cluster components, the cluster should shut down only if both
UPSs in a single rack lose power. For both UPSs within a rack to signal
that they are low before shutting down the cluster, both UPSs within a rack are
joined into a single logical UPS. This results in an UPS_SERIAL_PORT
configuration of:
example: UPS_SERIAL_PORT="/dev/tty00.1:/dev/tty00.2;/dev/tty01.1:/dev/tty01.2;"
After UPS-initiated shutdown has been configured as described in the preceding sections, restart the UPS daemon by entering the following two commands:
/etc/init.d/nscups stop
/etc/init.d/nscups start
If multiple UPSs are configured and one UPS loses power, the warning message will incorrectly state the system will shut down when UPS1 and UPS1 lose power instead of stating the system will shut down when UPS1 and UPS2 lose power. The system will operate correctly, the message is in error.