Booting Diskful Compute Nodes

In addition to booting diskless clients, ClusterWare can also integrate with so-called "diskful" compute nodes that boot from full installations on local disk drives. (See Using Kickstart and Appendix: Using Debian Preseed for examples.)

The administrator can add a locally installed node to the cluster using the same mechanisms as done with a diskless node. For example, if the new node's network interface is physically connected to the private cluster network shared by existing head node(s) and compute nodes, and if that interface is configured to use a dynamic IP address that will be assigned by DHCP at boot time, then on the head node execute:

scyld-nodectl create mac=00:11:22:33:44:55

and ClusterWare will assign the next available IP address to that MAC address.

Or if the new node has a static IP address, first ensure that static address is in the defined ClusterWare DHCP range, then additionally specify that static address:

scyld-nodectl create mac=00:11:22:33:44:55 index=100 ip=10.10.42.100

For the new node to fully integrate as a ClusterWare compute node, the cluster administrator as user root should install on the new node the clusterware-node package and dependencies. Since the node will not be using the initramfs provided by a ClusterWare boot configuration for diskless nodes, then after installing clusterware-node the administrator needs to update the configuration file /opt/scyld/clusterware-node/etc/node.sh on the node, which instructs the node how to communicate with a head node. This file must define the base_url of a head node and optionally the iface network interface name that assigned the MAC address used in the scyld-nodectl create that added the node to the database. For example, for interface eth0 that will connect to head node head-01, after editing:

[root@newnode]$ cat /opt/scyld/clusterware-node/etc/node.sh
# Specify a base_url for any head node
base_url=http://head-01/api/v1

# Specify the network interface used to reach the head node(s)
iface=eth0

# Optionally force (sslverify=yes) or disable (sslverify=no) SSL
# certificate checking. Leaving this option blank allows for HTTPS
# without certificate checking.
#sslverify=

On the node manually execute:

/usr/bin/update-node-status --hardware --upload

which sends the initial node hardware information to the head node, then reboot the node, which will now be fully integrated into the cluster.

The newly booted compute node can be controlled through the customary scyld-nodectl reboot, shutdown, and exec commands. To support --hard mode, the administrator must configure the node's power_uri field to provide appropriate ipmitool authentication and an IP address of the node's Baseboard Management Controller (BMC), e.g., ipmi:///admin:password@172.45.88.1. See Compute Nodes IPMI access and Database Objects Fields and Attributes for details.