Introduction to Linux Clustering
1. Hot failover
In a hot failover, the application is written specially for clustering and is able to
continue running on another node, without any interruption to client services.
Hot-failover is not often found in commonly used applications, and is usually found
in specially written programs for banking or telco situations.
2. Warm failover
Warm failover is what solutions like Linux-HA and Redhat Cluster Suite provide –
the application doesn't have an instant recovery feature, but the cluster suite quickly
restarts the application without minimal client disruption on another piece of running
hardware.
Using a cluster management solution, the application does not need to be written to
support clustering, so you can provide redundancy for any service you desire.
This can sometimes cause a small outage, as some applications can't tolerate the
change of the server in the background. In other causes, the application is able to
continue on, with no interruption to the client users with the exception of a bit of a
delay (eg: NFS is good at not being affected by a server change behind the scenes).
3. Cold failover
Cold fail-over is commonly used as a solution for redundancy where a cluster was not
able to be setup. In a cold failover, the dead computer need to be powered down, and
a spare computer started up. This is usually a manual process.
4.2.2 FENCING
When a node crashes or becomes unresponsive, it MUST be quickly powered off or blocked
from the storage device (fencing).
Fencing is required, because if the cluster assumes the node has crashed, and reallocates it's
services and IO, if the server was to wake up, it could cause havoc and possibly disk
corruption.
Therefore a failover device must be available so that the cluster can do STONITH – Shoot
The Other Node In The Head – by doing an instant power off of the node.
Various devices exist – smart power switches are usually used and scripts exist in Redhat
Cluster Suite that can connect to a number of commonly available devices to shutdown
© Copyright 2008 Jethro Carr Page 9/33
Komentáře k této Příručce