Currently in multiApply(), which takes a vector of oplog entries and distributes them across a set of writer threads to apply in parallel, any errors reported by the writer threads are stored in statusVector and returned at the end of the function. It may be possible to exit this function early after all the writer threads have finished running here on any writer errors instead of running the post oplog application steps. This should be fine because we will abort the process during steady state replication, and the initial synced will reset the database state before retrying.