Introduction to Linux Clustering
In the event of any node failing, Redhat
Cluster Suite will move all the services to
the alternate node and switch DRBD from
secondary to primary on the new master
node.
● A network failure runs the risk of a split-
brain situation – if neither node can see
the other, they will both try and become
master. To fix this, run ping to a remote
server to provide a tie-breaking third vote
(Quorum heuristics).
● If you have a SAN or NAS, instead of
storing the data locally and replicating
with DRBD, the attached storage device
could be used instead.
The SAN could be setup with a non-clustered or a clustered filesystem, the difference
being that a clustered filesystem is required if you require both servers to be able to
run services at once
If using a SAN, the cluster is scalable to more than two nodes, but the SAN could
become a single point of failure for the cluster and are a lot more expensive than
software solutions like DRBD.
Suitable Environments:
● Making any mission critical server HA.
● Any business or organisation that can not tolerate hardware downtime of their
production sever.
● Ecommerce servers that need to provide mail/websites/databases.
● Small hosting organisations. (larger ones should use designs like the one below)
8.2 Five node HA cluster with DRBD
One of the problems with DRBD is that it only works for two-node clusters. It is possible to
add a third node if the commercial version is purchased, but no DRBD solution exists which
can work more than three nodes at max.
When building clusters, it is more economical to have a single multi-node cluster rather than
many two-node clusters, as only one computer needs to be set aside for spare resources.
© Copyright 2008 Jethro Carr Page 26/33
host2
host1
switch
floating IP
smart
power
switch
sda
sda
DRBD
internet
remote host
for
quorum ping
Komentarze do niniejszej Instrukcji