SERVER-128705 added a guard to db_raii.cpp around a check for interrupt. This was needed because the RSTL based killConflictingOperations code had a check that interrupted any contexts that ever held the global lock, this could cause a ctx that 

1. Took a write / local write intent
2. Did its work
3. Released it
4. Began a lock free read

To get interrupted from its read during stepdown, which caused some issues around elections. If we were able to remove that condition from the kill code we can revisit if it is safe to have the checkForInterrupt call be un-guarded