Issue

Fail-over is different for different types of gluster restart.

When glusterd is stop or restarted the time taken is:

real    0m0.039s
user    0m0.002s
sys 0m0.004s

For a reboot:

real    0m25.954s    <<<<<=====
user    0m0.000s
sys 0m0.007s

For a halt:

real    1m22.415s     <<<<<=====
user    0m0.002s
sys 0m0.009s

real    1m14.180s      <<<<<=====
user    0m0.002s
sys 0m0.007s

Therefore with the node completely down it takes around 1m14s to switch to the other node.
With a reboot the delay depends on the amount of time it took to reboot.

The time to switch is the same for a down NIC.
ifdown ens3:

real    1m8.210s     <<<<<=====
user    0m0.002s
sys 0m0.005s