Using the Data Migration Facility (DMF) 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 DMF (see “Software Requirements” in Chapter 1).
You should configure DMF according to DMF Administrator's Guide for SGI InfiniteStorage.
This chapter discusses the following:
The following command adds the DMF user script to an existing service. The script used is /usr/lib/clumanager/services/svclib_dmf:
# sgicm-config-cluster-cmd --service=service1 \ --userscript=/usr/lib/clumanager/services/svclib_dmf |
You could also add the script by modifying the service in the GUI. For more information, see “Step 8: Configure the Service” in Chapter 4.
To run DMF, you must configure the parameters shown in Table 9-1. A required parameter must be defined by all users of DMF. An optional parameter is needed only by users of certain MSPs or the library server. DMF cannot start unless the required filesystems and directories defined by these parameters are first mounted and available on shared disks.
Table 9-1. DMF Administrative Filesystem and Directory Parameters
In addition, the working directory used by the dmaudit(8) command must also be available when DMF starts. To configure the directory, run the dmaudit command and select the <workdir> item in the <config> menu.
You can configure the DMF administrative filesystems (also known as support filesystems) as local XVM filesystems. You must define them as instructed in “Configuring DMF for Local XVM Filesystems”. SGI Cluster Manager ensures that the DMF plug-in script is called after the necessary filesystems are mounted.
 | Note: You should only configure DMF administrative filesystems as CXFS filesystems if they are also using DMAPI. |
To provide the best chance for database recovery, you should place the JOURNAL_DIR directory on a separate filesystem and a different physical device from the HOME_DIR directory.
If you use CXFS filesystems, you must define them as instructed in “Configuring DMF for CXFS Filesystems that Use DMAPI”.
To configure the DMF administrative filesystems as local XVM filesystems, do the following:
Ensure that the DMF configuration is identical on all members.
Create the DMF administrative filesystems on shared disks as local XVM filesystems (xvm type). See “Step 10: Add the Disk and Filesystem Information to the Service (Optional)” in Chapter 4.
Configure the SGI Cluster Manager local XVM volumes using the local XVM plug-in. See Chapter 11, “Local XVM Plug-In”.
SGI recommends that you configure DMF administrative filesystems as local XVM filesystems, as discussed in “Configuring DMF for Local XVM Filesystems”. DMF cannot start until the DMF administrative filesystems are available. If they are CXFS filesystems, CXFS must recover them before they are accessible.
To configure DMF filesystems that use DMAPI as CXFS filesystems, do the following:
Ensure that the DMF configuration is identical on all members.
Create the DMF administrative filesystems as CXFS filesystems (cxfs type). See “Step 10: Add the Disk and Filesystem Information to the Service (Optional)” in Chapter 4.
Note: Although the optional MOVE_FS and CACHE_DIR DMF administrative filesystems require the dmapi mount option, they should not be CXFS filesystems because they are only required on the node running dmdaemon. Configure the SGI Cluster Manager CXFS filesystems using the CXFS plug-in. For DMF-managed filesystems, configure relocate-mds=true (on) because DMF must run on the CXFS metadata server for that filesystem. See Chapter 8, “CXFS Plug-In”.
The /etc/dmf/sgicm_dmf.config file lets you configure other information required by SGI Cluster Manager. The sgicm_dmf.config file exists on all members in the cluster and should be edited as necessary on each member.
| Note: You must maintain the sgicm_dmf.config file on each member; a change on one member is unknown to the other members. |
You can specify the following directives in the sgicm_dmf.config file:
The following directive lets you control how the DMF behaves when a monitor failure occurs:
email-only-on-monitor-failure true|false
By setting email-only-on-monitor-failure to true, DMF will not return an error but will send a monitor failure mail message to the email address specified in email-addresses . The default is false. If email-only-on-monitor-failure is false(explicitly or by default), no email is sent; the monitor script will return an error and the service will be failed over.
The following directive specifies the email addresses to which a monitor failure message will be sent:
email-addresses email1[,email2[,email3...]]
The default address is root. If you specify multiple email addresses, you must separate them with commas. (An improperly formatted directive will be ignored.)
For example, the following sends email messages about monitor failures to chris and pat and does not return an error:
% cat /etc/dmf/sgicm_dmf.config email-only-on-monitor-failure true email-addresses chris@mycompany.com,pat@mycompany.com |
If the /etc/dmf/sgicm_dmf.config does not exist, no email is sent; the monitor script will return an error and the service will be failed over.
For the order in which DMF is started/stopped, see Chapter 6, “Creating a New Highly Available Application”.
When the DMF service is to be managed by SGI Cluster Manager, it is important that only the Cluster Manager starts DMF. Perform these commands on each member of the cluster to ensure that only the Cluster Manager can start DMF:
# touch /etc/dmf_failsafe # chkconfig dmf off |
To use the Tape Management Facility (TMF) with DMF in a SGI Cluster Manager environment, you must configure the appropriate TMF device groups in the /etc/tmf/sgicm_tmf.config file according to the instructions in Chapter 10, “Tape Management Facility (TMF) Failover Script”.
If TMF is configured as a mount service in the /etc/dmf/dmf.conf file, the DMF plug-in will automatically call the /usr/lib/clumanager/service/helper_tmf TMF failover script and pass along the appropriate TMF device group names.
The service timeout value should be at least 100 seconds if DMF is being used with TMF-managed tape devices. The following command will set the service timeout to 100 seconds for the SGI Cluster Manager service service1:
# sgicm-config-cluster-cmd --service service1 --servicetimeout=100 |
To do this with the GUI, see “Step 8: Configure the Service” in Chapter 4.