We spawn the replicated size and count flusher thread when a node steps up to be the primary. At the time, the node may not be accepting writes yet. A WT checkpoint could also happen concurrent, triggering a flush of the oplog entries from internal collections. If the flusher attempts to perform a write before the node becomes writable, it will get the error "code":10107,"codeName":"NotWritablePrimary","errmsg":"Cannot register write intent if we are not primary."}. Currently this error is treated as "the node is no longer a primary" and the flusher thread will be stopped. So at startup, this causes the flusher thread to be DOA.