Post 1 – http://www.guldmyr.com/blog/red-hat-clustering-and-storage-management-course-objectives/ Where I checked out udev, multipathing, iscsi, LVM and xfs.
This post is about getting using luci/ricci to get a Red Hat cluster working, but not on a RHEL machine because sadly I do not have one available for practice purposes. So CentOS64 it is. Using openstack for virtualization.
Topology: Four hosts on all three networks, -a, -b and internal. Three cluster nodes and one management node.
Get the basic cluster going:
- image four identical nodes
- ssh-key is distributed
- /etc/hosts file has all hosts, IPs and networks
- network interfaces are configured –
- set a gateway in /etc/sysconfig/network
- all traffic allowed from -a and -b networks
- at a minimum allow traffic from the network that the hostname corresponds to that you enter in luci
- dns (PEERDNS=no is good with several dhcp interfaces)
- timesync with ntpd
- luci installed on mgmt-node # ricci is a web gui
- ricci installed on all cluster nodes # this is the service talks with corosync
- password set for user ricci on cluster nodes
- create cluster in luci
- multicast perhaps doesn’t work so well in openstack ?
- on cluster nodes this runs “yum -y install cman rgmanager lvm2-cluster sg3_utils gfs2-utils” if shared storage is selected, probably less if not.
- fencing is really important, how to do it in openstack would require a bit of work though. Not as easy as with kvm/xvm to send a destroy domain message.
- Update and distribute cluster.conf
- Have a service run on a node on the cluster (doesn’t have to have a shared storage for this).
- rg_test test /etc/cluster/cluster.conf start service name-of-service
Share an iSCSI target between all nodes:
- Using management node to share the iSCSI LUN.
- tgtd, multipath
- clvmd running on all nodes
- lvmconf – make sure locking is set correctly
- create vg with clustering
- partprobe; multipath -r # do this often
- vgs/lvs and make sure all nodes see the clusterd lv
- minimum GFS filesystem is around 128M – you didn’t use all the vg right? =)
- for testing/small cluster lowering the journal size is goodness