Feature

Update the BMC without touching your nodes

in this fork since v2.7.0

Upstream power-cycles all four compute modules to patch the management computer beside them. This fork leaves their rails alone, so updating the board does not take the cluster down.

0modules power-cycled by a BMC update

0cluster nodes lost

48 sfor the board itself to come back

measured on a nine-node cluster during a live upgrade, 2026-09-08.

See it in the demo →The Power Control page, with every module powered and counted.Do it on your board →Coming from stock firmware, step by step.The evidence →This row and twelve others, each with its method.Updates that undo themselves →What happens if the image you are updating to is bad.

The four compute modules on the Power Control page, powered and counted in uptime, across a BMC upgrade.

The argument

Upstream cycles the modules because its update path resets the board's USB multiplexer and power rails together. Separating them is most of the change. The cost of not separating them is that patching a management computer becomes a cluster outage, which is the opposite of what a BMC is for.

The full argument, with every measurement →