Given a collection with a unique index; bulk inserts seem to silently drop documents; See the attached python script to reproduce the problem. It inserts documents with duplicate unique keys - inserts are performed using different bulk sizes (from bulk size 1 to one bulk for the whole test data). Among the 18 documents in the test data there are 10 unique keys, however, for none singleton inserts less than 10 items end up in the collection (the number of items in the collection is roughly number of bulks +/- 1).