I would like to implement hash-based sharding in my own application on top of MongoDB. For that purpose, I would like to pull a stable pseudorandom subset of documents into each of my servers, and I would like to do so without enlarging the documents by adding additional fields, and without using JavaScript in the aggregation pipeline (for performance reasons).

The idea: add a hash function, such as md5, to aggregation pipelines. The function would accept an object/array containing the data to be hashed, and would return the hash, ideally as a number.