This section describes notes when operating PRIMECLUSTER system.
Do not stop RMS while RMS is being started
Heartbeats between nodes are interrupted and the node where RMS is stopped may be forcibly shut down.
Stop RMS after completing its startup processing (completing the state transition processing of a cluster application).
Use hvshut -a to stop RMS on all nodes simultaneously
When executing the hvshut -l command on all nodes simultaneously, RMS will not be stopped and occasionally the timeout and hvshut command times out or hangs up.
When stopping RMS on all nodes, execute the hvshut -a command on any one of the nodes that configures a cluster system.
When stopping RMS on each node, execute the hvshut -l command on the node which stops RMS.
If mistakenly executing the hvshut -l command on all nodes simultaneously and the hvshut command times out, stop or reboot all the nodes. In addition, if the hvshut command hangs up, stop RMS forcibly using the hvshut -f command, and then stop or reboot all the nodes.
Do not stop operating system services after stopping RMS
Even if RMS is stopped using the hvshut command, other PRIMECLUSTER services (CF, SF, CRM, and so on) run.
Therefore, if you stop or reboot operating system services to modify its information (such as network information), heartbeat monitoring by CF fails and unexpected switchover will be occurred.
When modifying operating system information, be sure to do it after stopping all PRIMECLUSTER services (unloading CF) or in a single-user mode.
Create cluster applications used in RMS before starting RMS
If starting RMS without creating cluster applications, an error message (CML,14) will be output and RMS will not start.
The overview and the methods for creating cluster applications, "Chapter 6 Building Cluster Applications."
If operating systems hang up or slow down on a node in a cluster, a healthy node may be forcibly stopped.
If operating systems hang up or slow down on a node in a cluster due to system load, and so on, CF or RMS detects LEFTCLUSTER and stop the Shutdown Facility stops the node forcibly.
The Shutdown Facility forcibly stops a node according to the survival priority. Therefore, when the hang-up and slowdown of operating systems on the failed node are recovered before a healthy node forcibly stops the failed node, the healthy node may be forcibly stopped first.
When a system volume on a disk device cannot be referred to because all paths failed in a SAN boot configuration, the PRIMECLUSTER failure detection function cannot be operated depending on the status of the system.
Because the node which cannot refer to the system volume is unstable, set the node to panic status with the following method.
Stop the relevant node using the sdtool command.
# sdtool -k <the relevant node>
Set the active node to panic status. For details on how to set it, see the instruction manual of a main device.
Do not use the ipadm command for starting and stopping CIP as well as for changing its configuration.
If you start or stop CIP as well as to change its configuration using the ipadm command, an error message will be output and the command will fail.
Instead of that, use the cipconfig command or the ciptool command.
Do not use the service command of SMF for checking the status of PRIMECLUSTER and operating its system.
PRIMECLUSTER services are managed by SMF. Take note of the following two points:
The state of PRIMECLUSTER services cannot be checked by the service command (such as (svcs(1)). Check it by the system operation screen (Cluster Admin) or a command of PRIMECLUSTER as before.
Do not change the status of PRIMECLUSTER services using the service operation command (such as svcadm(1M)). Use the system operation screen (Cluster Admin) or a command of PRIMECLUSTER as before.
