Introduction to Linux Clustering
When a cluster node fails, the services that are running on it need to be migrated to
other machines. The cluster management software will look at a list of other nodes
called the “failover domain”, and will select one from the list to run the service.
The list can also be prioritised if desired – both the order of what servers should be
used in a failure, as well as whether or not the service should be moved back to a
higher priority node when one becomes avaliable.
In the event of the service running out of online hosts in the failover domain, the
service will be stopped until a node comes back online.
● Fencing devices
A fencing device is used to power off or reset unresponsive/crashed cluster nodes.
This is typically something like a network controlled power strip, or a out-of-band
management card in the server.
5.1.2 LUCI AND RICCI
Luci does not have to be run on the cluster itself, although that is the recommended method
as you can cluster luci and thus be able to always administrate the cluster.
The ricci daemon runs on all the nodes and allows Luci to communicate with the nodes to
configure them.
Luci is smart enough to install the packages it requires via yum on the nodes when you add
them to the cluster which makes setup easier.
5.1.3 SYSTEM-CONFIG-CLUSTER
Luci seems to be replacing system-config-cluster as the favourite program to use, but at this
stage system-config-cluster is a capable GTK GUI application for cluster configuration.
You can run it on any cluster node, once you save your changes you can then click a button
to send out the new configuration to all the cluster nodes.
5.1.4 LOAD BALANCING
Redhat Cluster Suite is focused on providing HA servers and doesn't provide any special
features for doing load balancing.
However, you can setup load balancing by the following method:
© Copyright 2008 Jethro Carr Page 13/33
Komentarze do niniejszej Instrukcji