Node Migration Commands
Whenever you reassign nodes to different masters in /etc/beowulf/config, the node migration can be put into effect with the killall -HUP command and the bpctl -R command.
- killall -HUP
The killall -HUP command is used to send SIGHUP (Signal: Hangup) to the beoserv daemon, causing beoserv to re-read the cluster config file. The command needs to be run on all masters to which you have synchronized the config file.
After copying an updated /etc/beowulf/config to each affected master, or alternatively after replicating (with care!) the changed entries into each master's config file, you can manually log into each affected master and run the command locally:
Or you can set up remote login from the original master and execute the command via ssh to the IP address of the new master, e.g.,:[root@cluster ~] # killall -HUP beoserv
Sending SIGHUP affects nodes only when they reboot in the future, allowing you to change settings for some nodes without affecting other nodes. Use of SIGHUP is an alternative to restarting Beowulf services, which reboots all compute nodes in the cluster.[root@cluster ~] # ssh 10.1.1.2 killall -HUP beoserv
- bpctl -R
Use the bpctl -R command to tell any range of compute nodes to reboot, for instance:
The specified nodes will do a full cold reboot through the BIOS. Any changes you have made to the configuration parameters in /etc/beowulf/config will take effect as the nodes reboot.[root@cluster ~] # bpctl -S 20-39 -R
Note that /usr/sbin commands such as bpctl do not work "out-of-the-box" for Red Hat. You should add /usr/sbin to your PATH so that these commands will work. Otherwise, you will need to include the directory path on the command line when issuing the command, for example:
[root@cluster ~] # /usr/sbin/bpctl -S 20-39 -R
The bpctl command and available command line options are discussed in the Scyld ClusterWare Reference Guide and also in the bpctl man page.