For a host to become a member of a PSE cluster, the host name must be registered in a PSE cluster partition in the PSE cluster database, and a PSE cluster daemon, farmd, must run on that host. A host can be a member of several PSE clusters. If so, there are multiple instances of farmd running for each PSE cluster.
This section describes the steps involved to add and configure a new host into a Cluster.
# /usr/sbin/pseconfig add cluster-spec
where cluster-spec is psefarm for a basic PSE cluster, the clustername for a DNS-based PSE cluster, or the PSE cluster database file name, including directory path, for a file-based PSE cluster.
pseconfig does the following:
localhost dgram udp wait root /usr/sbin/farmd farmd -farm localhost:5001
When pseconfig completes, the farmd daemon should be started automatically by the inetd daemon as soon as a connection request arrives to the clustername IP port. To immediately starts the farmd daemon, run the lspart -farm cluster-spec -jobslots command. This command connects to the local farmd daemon and causes inetd to start it.
You can temporarily deconfigure a host from participating as a member of a Cluster by running the following command:
# /usr/sbin/pseconfig delete cluster-spec
where cluster-spec is psefarm for a basic PSE cluster, the clustername for a DNS-based PSE cluster, or the PSE cluster database file name, including directory path, for a file-based PSE cluster.
The pseconfig command performs the following:
To reconfigure the host to be a member of a PSE cluster, issue the following command:
# pseconfig add cluster-spec
Note that deconfiguring a host from a PSE cluster forces the host to stop accepting PSE cluster application execution request on that host. However, PSE cluster applications currently running continue to run until the application itself terminates.
Also note that manually killing the farmd daemon by using the kill command terminates the currently running farmd daemon, but as soon as a new connection requests arrive, the inetd daemon starts a new farmd daemon.
To permanently remove a machine form a Cluster do the following:
This section describes the tasks required to maintain the PSE cluster daemon.
To immediately signal a farmd daemon to reread a modified PSE cluster database, run the following command on any of the PSE cluster members:
# kill -HUP cat /var/run/farmd.clustername.pid
The signaled farmd domain propagates the SIGHUP signal to all other farmd daemons in the PSE cluster.
Job slots are the number of simultaneous PSE cluster applications a farmd daemon allows at any given time for a given PSE cluster. The maximum number of job slots per given PSE cluster is maintained in the /etc/rc.config file under the variable PSE_SCHEDULING_UNITS. By default the number is 10. You can change this default number by running the following command:
# pseconfig -j n
where n is the new number to take effect.
By modifying the maximum number of job slots you can vary the potential system load that are caused by running PSE cluster applications. Setting the maximum number of job slots to 0 prevents the farmd daemons from accepting future PSE cluster applications. Refer to the following section on how to temporarily achieve the same effect as setting the job slots to 0.
In some cases it is desirable to temporarily disable the farmd daemon from accepting a PSE cluster application execution request. For example, when a PSE cluster member is being used for some other purpose and should not be affected by the increase in system load from any PSE cluster applications. To disable a farmd daemon, issue the following command:
# kill -USR1 `cat /var/run/farmd.clustername.pid`
To reenable the PSE cluster daemon, use the following command:
# kill -USR2 `cat /var/run/farmd.clustername.pid`
When a host is a member of a PSE cluster, exercise caution when you want to shut down the host for any reason. Even though no users are logged in to the host, there might be a PSE cluster application currently running on the host. Shutting down the system may severely disrupt a currently running PSE cluster application, even if it was not started at this host.
Follow this sequence to shut down a PSE cluster member:
For example, use the following command to determine if any PSE cluster applications are running on your local host:
# pspart -localhost
If there is a PSE cluster application currently running, you should wait until the application terminates before shutting down the system.
After a host has been configured to become a member of some PSE clusters, one farmd daemon per PSE cluster starts at system boot. The starting of each farmd daemon at system boot is done by the /sbin/rc3.d/S90farmd shell script. Similarly, the shutting down of each farmd daemon at system shutdown is done by the /sbin/rc2.d /K29farmd shell script.
Note that Digital UNIX provides a special convention for file naming in the run command directory structure (/sbin/rc0, /sbin/rc2, and /sbin/rc3). A prefix of either "K" or "S" to the command names in these directories determines whether the system starts or stops these command. For further information, refer to Digital UNIX (DEC OSF/1) guide to system administration.
The S90farmd and K29farmd shell scripts reads the /etc/rc.config file for any PSE cluster configuration information. The relevant variables in this file are:
| Variable | Explanation |
|---|---|
| PSE_CONF = {YES|NO} | Whether the host is configured to be a PSE cluster member |
| PSE_ SCHEDULING_UNITS | The maximum number of job slots per PSE cluster |
| PSE_CONFIGURED_DOMAINS=n | The number of PSE clusters for which the host is configured |
| PSE_FARM_DOMAIN_1 | clustername1, the first PSE cluster name |
| PSE_FARM_DOMAIN_n | clusternamen, the nth PSE cluster name |
The PSE software provides the psemon utility to monitor PSE activity. The psemon utility displays PSE cluster configuration and load information. Refer to the psemon(1) reference page for details.
