This separate loop (code link) over the data is not needed, it can be done inline with another one of the many loops in this code path to improve cache locality and avoid needlessly thrashing the cache on large multi-inserts.

int64_t totalLength = 0;
for (size_t i = 0; i < nRecords; i++)
    totalLength += records[i].data.size(); 

Two options:

• Keep the interface the same, and compute the totalLength in both of the other loops (one for capped collections, one for regular). Pro: keeps responsibility of this function the same AND each record's size is already used in both of the other loops.
• Compute this at a higher layer in one of the other loops and pass into this function.