14.2.1 Notes on Clusters Between Control Domains

Installation of PRIMECLUSTER

After installing PRIMECLUSTER, set the OBP auto-boot? variable to "false" and specify "set halt_on_panic=1" in /etc/system in order to prevent the control domain forcibly stopped from rebooting automatically.

For details, refer to "4. Installation of PRIMECLUSTER (Control Domain)" in "14.1.1.3 Cluster System Between Control Domains".

Setting of Guest Domain

PRIMECLUSTER on the control domain controls only startup and shutdown of guest domains. It does not control the operating systems and applications on guest domains. Therefore, set up your guest domains so that operating systems and applications are automatically booted in extension of each guest domain's startup. For details on how to automatically start operating systems in guest domains, refer to "9. Guest Domain Setup (Guest Domain)" in "14.1.1.3 Cluster System Between Control Domains".

Registering Cmdline resources for monitoring the guest domain status

The import of ZFS storage pool may fail on the switch destination guest domain if all of the following conditions are met:

Each guest domain has different host identifier values (output of hostid).
The ZFS storage pool is located on a disk (other than the system disk) shared between guest domains.

In the environment where all of the above conditions are met, the following settings are necessary to prevent this event.

Perform the procedure below when the guest domain is in "bound" state.

Set the same host identifier to each guest domain that shares a disk.
Execute the following command on the control domain of each guest domain that shares the disk.
The host identifier must be other than 00000000, and it must be the same value on each guest domain that shares the disk.
```
# ldm set-domain hostid=<hhhhhhhh> <guest_domain_name>
```
(<hhhhhhhh>: Specify the host identifier in 8-digit hexadecimal number.)
<Execution example>
```
# ldm set-domain hostid=1234abcd ldom01
```
Make sure the same host identifier is set to each guest domain.
Execute the following command on the control domain and confirm that the same host identifier is set to each guest domain that shares the disk.
```
# ldm list-bindings <guest_domain_name> | awk '/HOSTID/,/0x/'
```
<Execution example>
```
# ldm list-bindings ldom01 | awk '/HOSTID/,/0x/'
HOSTID
    0x1234abcd
```

Controlling Guest Domains

Never use the ldm(1M) command immediately to start or stop a guest domain or to bind its resources.
Guest domains have to be started and stopped from PRIMECLUSTER.
Otherwise, you risk loss or damage of data.

Stopping the System While Guest Domains Are Operational

Control Domains
Run the shutdown(1M) or the init(1M) command to stop the system. Using reboot(1M), uadmin(1M) or similar commands may possibly cause data loss as the control domain is rebooted without suspended guest domains.
Guest Domains
To stop guest domains, run the shutdown(1M) command with the "i0" option.
After that, put the cluster applications of the suspended guest domains in the control domain offline.

When Control Domain Is Stopped

Guest domains on a forcibly stopped node or partition, which have been monitored by the cluster system, are sometimes not stopped although the control domain has been forced to stop. Data loss may occur when restoring the control domain while any of the guest domains are not stopped.
In order to stop absolutely all input and output from guest domains, make sure to turn off the electrical power of nodes or partitions by either of the following methods before restoring the stopped control domain. After that, reboot the disconnected nodes or partitions. Also note that no kernel crash dump is collected for operating systems in guest domains.

On the ILOM remote control, select "Immediate Power Off".
Stop partitions with the poweroff command of XSCF.
Press and hold the power button of the rack for a while.

See

For details on the remote control, refer to the "Integrated Lights Out Manager User's Guide".
For details on XSCF, refer to "Fujitsu SPARC M12 and Fujitsu M10/SPARC M10 Domain Configuration Guide".

Manual Switching of Cluster Applications

When guest domains are switched to a standby system, PRIMECLUSTER uses the ldm(1M) command to stop those guest domains.
Therefore, whenever PRIMECLUSTER suspends a guest domain, the OS and all applications in that guest domain stop abruptly.
In order to avoid such abrupt stops when switching cluster applications manually, log in to the guest domain OS and stop the OS by running the shutdown(1M) or a similar command before switching the cluster application.

Shutting Down the Control Domains

Before shutting down a control domain, stop RMS on the control domain to set Cmdline resources which control guest domains offline. For how to stop RMS, refer to "7.2.1.2 Stopping RMS".

14.2.1 Notes on Clusters Between Control Domains

14.2.1.1 During Installation

14.2.1.2 During Operation