From BF-46053.

The reshard_collection_2d2r_with_users_with_heavy_memory_usage_locust workload starts with an initial collection of 500k documents. Then CRUD traffic and heavy memory traffic is performed in the background while the resharding is performed.

The problem that occurs in the BFGs is that the original 500k documents has defined keys for the old and new shard key values, but the newly inserted rows (from CRUD traffic) have a null "new" shard key. This dilutes the collection is duplicate documents and leads to cardinality errors while calculating split points.

In the latest BFG, 700k null "new key" documents are inserted by the end of the workload (compared to 500k unique documents).

The proposal to fix the issue is to ensure that all new rows inserted with CRUD traffic use a unique value for the new shard key field.