The changes in SERVER-68380 made it such that MongoDB uses WT_CURSOR::search less often (i.e. only for collection accesses) and as of SERVER-86216, almost never uses WT_CURSOR::search_near.

As far as I can tell, we only increment the operation read latency stats for calls to search() and search_near().

The MongoDB usage pattern for indexes is to use WT_CURSOR::bound() followed by a call to next() or prev(). We are currently losing information about the latency of these operations.

I suspect that incrementing a metric on each call to next() will be too expensive, so perhaps we can consider more sustainable tracing inside MongoDB instead.