What should be done?

$collStats created the whole parallel universe for itself not only in the Stage interface but also in the aggregate command implementation. Could we fold that special case in the general design?

At the moment, the complexity is around $collStats does not always need to query database, and if does it uses its specific query to fetch data. There are other stages that does not query database such as $documents and $listSessions.

The proposed design separates Stages to ProducerStage and ProcessorStage (see #3102), each ProducerStage may decide how to produce document iterator.

Where?

https://github.com/FerretDB/FerretDB/blob/main/internal/handlers/pg/msg_aggregate.go
Also find issue link in the code base

Definition of Done

• remove Type() interface
• remove $collStats specific code from msg_aggregate.go, it should be in collStats.go.
• calling database is optional by design (see proposed design Reorganise aggregation by deciding how initial document is fetched #3102)
• integration/compatibility tests added/updated;
• spot refactorings done;