Using the CXFS clustered filesystem product with SGI Cluster Manager for Linux requires the value-add SGI product on the SGI Cluster Manager 4.2 for Linux - Storage Software Plug-ins CD and the supported level of CXFS (see “Software Requirements” in Chapter 1).
You should configure the CXFS cluster, nodes, and filesystems according to CXFS Administration Guide for SGI InfiniteStorage.
You must give careful consideration when choosing nodes used for the SGI Cluster Manager cluster and nodes capable of being the CXFS metadata server. Certain services, such as DMF, require that the CXFS metadata server and the SGI Cluster Manager service be provided by the same node (see “Configuring DMF for CXFS Filesystems that Use DMAPI” in Chapter 9). Others, such as NFS and Samba, do not have this requirement, but it may be desirable to enforce it in order to ensure that these services provide the best performance possible. (Some NFS and Samba workloads can cause significant performance problems when the NFS or Samba service is provided from a node that is not also the CXFS metadata server.) Unless a service must run on the CXFS metadata server, you should configure SGI Cluster Manager so that it does not relocate the CXFS metadata server when it fails over a service.
This chapter discusses the following:
To use CXFS relocation and fail over the service from one member to another, you must set the relocation_ok parameter to 1 (enable) on all potential CXFS metadata servers. For more details, see the CXFS Administration Guide for SGI InfiniteStorage.
For I/O fencing, SGI Cluster Manager members should use l2network power controllers in order to prevent conflicts with CXFS I/O fencing methods. For more information, see “L2 Connections” in Chapter 2.
To include CXFS filesystems in the SGI Cluster Manager configuration, add filesystems as devices used by a service, as shown in Figure 8-1.
Enter the following:
 | Note: The Force Unmount item in the GUI and CLI is ignored for CXFS filesystems. |
In the CLI, do the following:
sgicm-config-cluster-cmd --service=servicename \
--add_device \
--name=/dev/cxvm/volumename
sgicm-config-cluster-cmd --service=servicename \
--device=/dev/cxvm/volumename \
--mount \
--mountpoint=mountpoint \
--fstype=cxfs \
--options=relocate-mds=true|false |
You can specify multiple CXFS filesystems by adding multiple devices to the service.
| Note: You will not define a --userscript value when defining a service to failover CXFS filesystems. User scripts are used for failing over user-written applications. For more information on user scripts, see Chapter 6, “Creating a New Highly Available Application”. |
For more information, see “Step 10: Add the Disk and Filesystem Information to the Service (Optional)” in Chapter 4.
The members in the failover domain for the service that has CXFS filesystems should be same as the list of potential metadata servers for the CXFS filesystem. For example: machines node1 and node2 can be metadata servers for CXFS filesystem /cxfs_san1. The SGI Cluster Manager service nfs1 that uses /cxfs_san1 should have a failover domain of node1 and node2.
You should start CXFS cluster services and CXFS services before starting SGI Cluster Manager daemons. SGI Cluster Manager will wait for the CXFS filesystem to be mounted by CXFS before starting NFS, Samba, and other applications running on the CXFS filesystem. Therefore, service timeouts for all SGI Cluster Manager services that include CXFS filesystems should be carefully adjusted accordingly.
For the order in which CXFS is started/stopped, see Chapter 6, “Creating a New Highly Available Application”.