The difference between kReplicatedFastCountWatermarksWritten and kReplicatedFastCountTailerWatermarksSeen tells us how many watermarks in the oplog have not been consumed by the oplog tailer. If this number is large, then it is likely that the oplog tailer is behind. We can derive this metric manually, but it would be great to have an explicit derived metric for the difference.

Add a derived metric to the t2 repo for the difference between these two terms. Also consider adding a Grafana plot for this value because it helps indicate the health of our flusher/tailer API.