3.5.1 Matrix Server Administration Guide

Chapter 12: Configure Virtual Hosts 217
Copyright © 1999-2007 PolyServe, Inc. All rights reserved.
Customize Service and Device Monitors for Failover
By default, when a service or device monitor probe fails, indicating that
the watched service is down or the monitored device cannot be accessed,
ClusterPulse will fail over the associated virtual host to another server
where the monitored service or device is up.
You can customize this behavior using the Advanced monitor settings.
Specify Failover/Failback Behavior
The Probe Severity setting allows you to specify whether a failure of the
service or device monitor probe should cause the virtual host to fail over.
For example, you could configure a custom device monitor to watch a
router. The device monitor probe might occasionally time out because of
heavy network traffic to the router; however the router is still
functioning. In this case, you can configure the device monitor so that the
virtual host will not fail over if only the device monitor has failed.
The Probe Severity setting also affects the failback behavior of the virtual
host.
There are three settings for Probe Severity:
NOFAILOVER. The virtual host does not fail over when a monitor
probe fails.
AUTORECOVER. The virtual host fails over when a monitor probe
fails. When the service or device is recovered on the original node,
failback occurs according to the virtual host’s failback policy.
NOAUTORECOVER. The virtual host fails over when a monitor probe
fails and the monitor is disabled on the original node, preventing
automatic failback. When the monitor is reenabled, failback occurs
according to the virtual host’s failback policy
Select a Backup Interface for Failover
When ClusterPulse needs to fail over a virtual host, it determines the best
network interface to receive the virtual host. It bases its decision on the
status of the available backup interfaces and their underlying servers.
Typically, it looks for an “up” network interface where all services are
“up” on the underlying server.