CheckReplDBHash currently does does a w:<all_nodes> write to ensure all previous commands have replicated and that no commands are executed after running {{dbhash]}.

But this does not guarantee that writes with w:0 close to the end of a test are treated correctly; they may still get inserted into the oplog after the CheckReplDBHash write.

To fix this issue, it should be sufficient to do an fsync on the primary, given that the test fixture has exactly 2 nodes and no failover.