There is a specific race that is possible (found by Antithesis) where after a specific interleaving the _bgSync object would have not been created yet when the mongod went to stepup from primary, which means that when we are in stepup the state would be the default which is Starting instead of the expected Stopped. This means that startup will cause the bgsync thread to fetch oplog entries, which means the primaries tries to apply its own writes. This crashes due to stable timestamp assertions, and the node can restart and recover automatically.