A delay (>= 20 s) is needed between power-off and power-on events for machines.
There is empirical evidence that some cores take much longer that average to power off. If power is applied again "too soon", they do not boot correctly. This problem can be avoided by waiting before turning power back on after a machine has been powered off.
This problem has only manifested on large machines, probably because "power off" is not an atomic event. spalloc-server has to issue power off commands sequentially to each frame in the machine.
A delay (>= 20 s) is needed between power-off and power-on events for machines.
There is empirical evidence that some cores take much longer that average to power off. If power is applied again "too soon", they do not boot correctly. This problem can be avoided by waiting before turning power back on after a machine has been powered off.
This problem has only manifested on large machines, probably because "power off" is not an atomic event. spalloc-server has to issue power off commands sequentially to each frame in the machine.