When running aggregation, the aggregation framework will pre-evaluate some of the arguments. During that pre-evaluation, it may encounter and throw an error. If optimizations are disabled via failpoints, however, the pre-evaluation phase does not happen and the error is not hit, instead the aggregation simply returns no documents.

As a result, when the aggregation fuzzer is run in optimization mode, it frequently hits this discrepancy (one mongod throws an error and another returns 0 documents).

One possible solution from charlie.swanson would be to add context to the error messages hit during pre-evaluation. The fuzzer could then check for that context and validate accordingly.

This ticket is to investigate what it would take to implement that solution (or another solution if appropriate), make any needed changes to the mongo repo and open a ticket to update the aggregation fuzzer.