3.5.1 Matrix Server Administration Guide
Chapter 12: Configure Virtual Hosts 218
Copyright © 1999-2007 PolyServe, Inc. All rights reserved.
You can use the following Advanced settings to affect how ClusterPulse
selects the network interface for failover.
• The Event Severity setting allows you to specify whether ClusterPulse
should consider the existence of monitor events (such as a script
failure or timeout) when it chooses a network interface for failover. If
the events are considered, the network interface for the affected server
becomes less desirable.
• For service monitors, you can assign a priority to each monitor (the
Service Priority setting). If ClusterPulse cannot locate an interface
where all services are “up” on the underlying server, it selects an
interface where the highest priority service is “up” on the underlying
server.
Specify Failback Behavior of the Virtual Host
When you configure a virtual host, you can specify the failback action
that the virtual host will take following a failover to a backup server.
There are two options:
AUTOFAILBACK and NOFAILBACK.
AUTOFAILBACK is the default for virtual hosts. (Some PolyServe products
may provide entities that use
NOFAILBACK as the default.)
AUTOFAILBACK
AUTOFAILBACK is intended to return the virtual host to its original
configuration, or as close to it as possible. After the virtual host fails over
to a backup server, ClusterPulse watches the health of the servers higher
in the list for that virtual host. When the health of one of these servers is
equal to or greater than the backup server where the virtual host currently
resides, the virtual host automatically attempts to fail over to that server.
NOFAILBACK
NOFAILBACK is intended to minimize failovers. The virtual host remains
active on the backup server until a healthier server (it’s health is greater
than the backup server) becomes available, at which point the virtual host
will fail over to that server.
For example, consider a two-node matrix that is configured with its
primary on node 1 and backup on node 2, and that uses the
NOFAILBACK
option. Three service monitors are configured on the virtual host. When a