With Primary Driven Index Builds, almost all WriteUnitOfWork instances are expected to be batched operations, such that on commit the batched operations get serialized into an applyOps oplog entry with many keys. When testing, however, it has been observed that during an index build's scan phase, following initial batches of ci operations in a single applyOps oplog entry, there are many single-key, top-level ci oplog entries, creating potential inefficiencies in the oplog. We should investigate why and when this is happening and fix any potential issues so that we can group batched operations into fewer oplog entries.

This has been observed when running jstests/noPassthrough/index_builds/primary_driven/resumable_load_phase.js, for example.