Today the MCO bootstraps with a bootstrap MCC/MCS to generate and serve master configs. When the in-cluster MCC comes up, it then tries to regen the same MachineConfig via the in-cluster MCs at the time.

This often causes a drift and for the install to fail. See https://github.com/openshift/machine-config-operator/issues/2114 and https://docs.google.com/document/d/1fgP6Kv1D-75e1Ot0Kg-W2qPyxWDp2_CALltlBLuseec/edit#heading=h.ny6l9ud82fxx for more context. For the most recent occurrence of this, see: https://github.com/openshift/machine-config-operator/pull/3513

Early on this helped us see differences between bootstrap and in-cluster behaviour more easily, but we do have the bootstrap machineconfig on-disk on the masters. In theory, we should just be able to use that directly and attempt to consolidate the changes.

In the case of a drift, instead of failing, we can consider doing an immediate update instead to the latest version.