To calculate accurate split points on the shard key during reshardCollection operation, we would like to perform a distributed index scan, given that an index exists on the shard key. This should be done using an MQL-based aggregation pipeline.

One approach to achieve this is by adapting the existing autoSplitVector code. This code is currently implemented outside the query subsystem and handles shard-local splits.

Alternatively, creating a new MQL aggregation pipeline from scratch could also meet our objectives. It is recommended to work with Query team to benefit from their expertise in writing an optimized and efficient pipeline.