Booting I/O Nodes

It is important that the raids are unmounted before halting or rebooting the I/O nodes. It takes about 40 minutes to fsck a raid. A few minutes spent taking down the I/O nodes correctly can save many minutes when it comes back up.

The raids are attached to the following nodes:

• c-3.SU-22
• c-5.SU-22
• c-7.SU-22
• c-11.SU-22
• c-13.SU-22
• c-15.SU-22

I (wmdavid) have found "boot-node halt" command from the sss1 to be unreliable. Sometimes the power is cycled before the the raid is unmounted. I think the best method right now would be to login to each I/O node as root and execute the "halt" command. After all six I/O nodes are down the "boot-node" command can be executed to power cycle the the nodes. This method is only feasible with 6 I/O nodes. As the I/O nodes grow, a more scalable booting procedure must be created. The procedure should have bi-directional communications so the boot script does not cycle the power until it _knows_ that the raids (and other disks) are unmounted.

In some error cases, it may be faster to restart fyod on the I/O nodes rather than reboot them. Login as root on each I/O node and execute the following commands:

/cplant/etc/fyod stop
/cplant/etc/fyod start