Overview

Prototype correct behavior across step-up, step-down, restart, and rollback for the new write path.

Scope of Work

• Step-up: drain unapplied oplog, initialize the detector floor to lastApplied, then accept writes.
• Step-down: stop accepting writes, stop/drain the primary applier, abort staged transactions.
• Restart: replay through existing startup recovery.
• Rollback of an unapplied oplog tail on a former primary.
• One jstest per scenario.

Acceptance Criteria

• Each scenario works in the prototype or the gap is documented.
• Findings recorded on this ticket, including known holes.

Technical Notes

• Prototype quality; not intended for commit to master.

Design Note Alignment

• Staged writes are never durable, so the buffer is empty after restart by construction; restart needs no staged-write recovery.
• Prepared transactions with staged writes are forbidden in the prototype; prepare/2PC is out of scope.
• Design reference: src/mongo/db/txns/staged_writes_design.md (PR #66680).