Start Cluster Operations

Reboot the master node:

[root@scyld ~]# reboot

After rebooting, run:

[root@scyld ~]# uname -r

and confirm that the master node is running a Scyld ClusterWare kernel.

Normally, Scyld ClusterWare services automatically start whenever the master node reboots. However, Scyld ClusterWare requires that you have read and accepted an End User License Agreement (EULA). You should have done this when executing the install-scyld script. Additionally, any error in the /etc/beowulf/config configuration file will result in a clusterware service startup error.

Test for a functional ClusterWare by executing the simple command to view the cluster status: bpstat. A successful first output line should begin with Node(s). If that does not appear, then attempt to start the ClusterWare service manually and look for an error message:

[root@scyld ~]# systemctl start clusterware

Once the clusterware service is up and running, the master node can PXE boot as many compute nodes into the cluster as were defined by the “nodes” directive in the /etc/beowulf/config file. You can monitor the cluster status with the graphical beostatus, with the text-based beostatus -c, or with a simple bpstat -U.

Note: Depending upon BIOS settings, the compute nodes’ DHCP requests may timeout because the master node hadn’t been ready to respond, and compute nodes would then revert to a BIOS prompt waiting for human input. If bpstat continues to show that compute nodes are down, then ‘reset’ or powercycle each compute node, either manually or using an already configured ipmitool.