Planning the NFS server automation

Before you implement NFS server automation, you must fulfill specific prerequisites and infrastructure requirements.

Prerequisites

Virtual IP address: For each virtual IP address that is defined in the high availability policy, an equivalency of network interfaces will be defined. Only network interfaces with the same network interface name can be part of the equivalency. Ensure that the network interfaces you want to use have the same name on each system.

Infrastructure requirements

To allow an NFS server to failover between different systems the following requirements against the system infrastructure have to be fulfilled.

At least one shared disk is required for the NFS server high availability configuration

One or more disks contain the data that is exported by the NFS server. They contain configuration files that the NFS server requires to run properly and behave identically on all systems.

On a Linux® machine the NFS server uses files located in /var/lib/nfs to store information about any NFS clients which have mounted a file system that is exported by this NFS server and information about file locks.

On AIX® this information is contained in the files /etc/xtab, /etc/exports, /etc/rmtab and the directory /var/statmon.

This information must persist after a node failover to the NFS server, otherwise the NFS clients lose their connections to the NFS server. The NFS client information is stored in a small file system on a shared disk to keep this information in sync on all nodes. These disk(s) must be accessible by all nodes that are allowed to host the NFS server, for example if the nodes are connected to a storage device. Each shared disk has a unique device name and number. The unique device name and number must be the same on all nodes. Depending on your operating system, these requirements are:

Linux: To transparently fail over the NFS server from one host to another on a Linux machine, the shared disk(s) must be connected to each server with the same unique major and minor number. System Automation for Multiplatforms can only identify the shared disk if all nodes identify the shared disk with the same unique number. The shared disk must have the same device name and number on all nodes, for example /dev/sdc1. You can determine the major and minor number as follows:
```
#cd /dev
#ls –l sd*
The last command will return an output like this
<systemname>:/dev # ls -l sd*
brw-rw---- 1 root disk 8, 0 2011-01-19 14:25 sda
brw-rw---- 1 root disk 8, 1 2011-01-19 14:25 sda1
brw-rw---- 1 root disk 8, 2 2011-01-19 14:25 sda2
brw-rw---- 1 root disk 8, 3 2011-01-19 14:25 sda3
brw-rw---- 1 root disk 8, 4 2011-01-19 14:25 sda4
```
For sda, the major number is 8 and the minor number is 0.
For each file system that resides on such a shared disk, a mount point must be defined within /etc/fstab. For each mount point, an IBM.AgFileSystem resource must exist, either harvested by the StorageRM resource manager or user-defined. It is recommended to specify a label for the file system, because this label is taken as resource name for harvested IBM.AgFileSystem resources. For more information about the StorageRM, see Storage resource manager.
AIX: On an AIX machine the volume group MAJOR NUMBER must be identical to the volume group containing the shared data on all of the nodes. For details about the MAJOR NUMBER, see the lsvg command man pages. Make sure the setup is done accordingly and keep in mind that this is also important for later changes of the configuration of the servers. This applies for every shared disk hosting data to be exported by the NFS server.

NFS version 3 and 4 only are supported for the NFS high availability policy

NFS version 4 authentication using Kerberos is not supported.

A single node can run only one NFS server

If you want to run consolidation projects, consolidate the exports of several NFS servers to one NFS server instead of placing several NFS servers on a single machine. If there is more than one NFS server resource within a System Automation cluster, then these resources must have an AntiCollocated relationship to ensure that these NFS servers are never started on the same node at a time.