alternative to BMC to manage metal

[aarnaud]

Hi, I open this discussion to brainstorm about alternatives.

First of all, thanks for this project.

I know the main design is to use BMC solution to control the machine but from my point of view is also too limiting because not all hardware has BMC, like rackable workstation that is cheaper for some usage that doesn't require strong hardware.

• Some machine can be managed by Intel AMT or AMD Dash, it's not BMC but you can control power state, but not all hardware have it (ex: arm)
• Un-allocated machine that reboot loop until affected status if we can't control power state seem not ideal
• To reset a machine, it need to be rebooted to force to be wiped, if not there stay up and can't be reused.

Do you think we can use, if no BMC configured, this logic bellow:

• Use agent with a loop check to trigger a reboot when the machine is allocated, we can also add a max uptime in this state to ensure we reboot every X min/hour to use up-to-date agent. That limiting the number of reboots, that may also reduce hardware fatigue.
• On talos side, have a loop check that ensures the machine is still allocated inside CAPI if not, trigger a reboot or/and a talos reset. Or CAPI can use the reset feature with Talos certificate generated for the cluster
• If WakeOnLan works we can replace reboot by shutdown.

Obviously, I can contribute to develop this logic.

What to you think ?

[aarnaud]

Ref: #1285 #852 #1294

[aarnaud]

I push a draft PR here #1312