Bulk insertions on a node are divided into smaller sub-batches, each containing up to 64 documents by default (as specified by internalInsertMaxBatchSize). Non-transactional bulk insertions don't atomically synchronize with DDL operations; instead, they only synchronize at the individual sub-batch level. If a ddl operation occurs between two sub-batch insertions, the second batch will operate on the old incarnation of the database/collection.
For some ddls, such as rename or drop, the insertion will recreate the old namespace and insert data there.

Specifically for rename this can result in two collections containing partial data instead of a single collection with all the data.