A retryable write - specifically, an insert with w:majority, and likely other retryable writes - is interrupted on the server and retried after stepDown if it was written on the primary before the stepDown but not acknowledged by a majority before the stepDown. However, the retried write is not interrupted on the server. We have reproduced the original primary stepping up, but we expect that a node that was a secondary (and had already replicated the write in question) could step up to primary and exhibit the same behavior. We have not tried to reproduce that, but it should be done as part of this ticket.

An interesting question is whether or not this is a day-0 behavior.