The user accounts within the guest operating system.

The properties of the CPU and other hardware devices.

The MAC address of the network interfaces.

The version of the guest operating system.

The software and applications installed on the guest system.

Regular backups become generally unnecessary as any data loss or corruption is automatically recovered by the high availability cluster.

Regular backups remain vital as they allow restoration of data that was accidentally deleted or corrupted.

Regular backups may be reduced to the cluster configuration and cluster state data as this information is sufficient to recover any state of the cluster, including data, at any time.

Backups should be located only on the local storage of the cluster and should not be backed up in order to prevent bandwidth congestion.

Backup procedures may have to be adapted to the cluster to ensure complete and recoverable backups.

Each nanny process checks the state of one configured service on one real server, and tells LVS if the service on that real server is malfunctioning.

The nanny process powers off a nonfunctional backend server.

The nanny process restarts the daemon of a failed service on a real server node.

The nanny process changes the IP routing table in the Linux kernel according to the LVS configuration.

cibquery -o resources

cibadmin -R

cibadmin -Q -o resources

cibquery -L resources

cib_admin -L resources

Load balancers cannot be protected against outages which is why they should always be built on dedicated hardware.

No actions are necessary as a load balancer cannot become a single point of failure.

Two or more load balancers should be deployed as a cascade that passes packets from one to the other.

Two load balancers should be deployed as a failover cluster.

Load balancers are used for performance and scalability reasons only and cannot be used to archive high availability clusters.

/etc/ocfs2/cluster.conf

/etc/ocfs/cluster.conf

/etc/ocfs2.conf

/etc/cluster.conf

A clone resource for the OCFS2 file system.

A shared storage device accessible by all nodes using the OCFS2 filesystem.

The distributed lock manager (DLM).

The cluster logical volume manager (CLVM).

The O2CB service.

cat /proc/drdb/status

cat /proc/drdb

drbdadm dstate all

drbdadm diskstatus all

drbdadm state --disk all

A security resource in order to permit other cluster services to access the data on the DRBD device.

A health resource in order to check the integrity of the file system.

A file system resource to mount the device.

A replication resource that synchronizes the DRBD device's data to all nodes that may take over the device after a failover.

A constraint to ensure the device is mounted before the dependent services are started.

OCFS2 is an integral part of Pacemaker and relies on Pacemaker services for functionality.

To avoid the need for shared storage, OCFS2 can replicate the content of its filesystems via the network.

When using OCFS2 with additional cluster software, OCFS2 must be integrated into the overall cluster manager to ensure consistent cluster behaviour.

OCFS2 can be used without any additional software as it contains its own cluster manager, O2CB.

In addition to filesystems, OCFS2 can handle other cluster services such as IP addresses and server daemons.

The distinct IP addresses of all backend servers are called virtual IP addresses.

Every server running keepalived has exactly one virtual IP address and cannot have any additional virtual IP addresses.

The virtual IP address is the address to which incoming connections are made in order to reach the load balancer.

keepalived uses VRRP to ensure the availability of the virtual IP address.

Within the whole LVS cluster managed by keepalived, the virtual IP address never appears on any network interface.

Clusters with one or two nodes cannot have a quorum and require no quorum configuration.

In a two node cluster, quorum can be determined and handled just as in a cluster with a higher number of nodes.

In a two node cluster the quorum is always held by the node that first started the cluster software.

The default quorum policy in a two node cluster must be set to ignore in order to configure a working cluster.