-
Type:
Improvement
-
Resolution: Unresolved
-
Priority:
Unknown
-
None
-
Component/s: Change Streams
-
Needed
Summary
- Customer problem: change-stream clients of replica sets don’t go through mongos (by definition) and get none of Phase 3’s protection – a stream could fail due to missing coverage, while other members in the replica-set can provide the requested oplog entires. This is essentially a routing problem.
- Why it matters: this extends Phase 3’s protection to direct replica-set clients.
- Phase outcome: a driver spec, implementation in various drivers, and an adoption plan.
- Priority: opportunistic. This ticket must not block Phases 1–3.
Motivation
Who is the affected end user?
Users of MongoDB change streams.
How does this affect the end user?
It breaks active change-streams despite having members in the replica-set that can serve the requested oplog history.
How likely is it that this problem or use case will occur?
Very likely – this is a steady-state issue that can primarily be attributed in differences in oplog retention among members of a replica-set.
If the problem does occur, what are the consequences and how severe are they?
This shows up in users' application as broken change streams.
Is this issue urgent?
Not urgent at the time of filing this ticket.
Is this ticket required by a downstream team?
Atlas Stream Processing
Is this ticket only for tests?
No
Acceptance Criteria
- A driver spec proposal is drafted, covering all of: the topology signal, eligibility predicate, selection and retry behavior, no-eligible-member error semantics, mixed-version behavior, and fallback.
- A per-driver assessment lists which official drivers can consume the signal without pipeline inspection, and the constraints for those that can’t.
- A written implementation and adoption plan exists, separate from the server plan.
- The spec’s signal and error semantics are consistent with the finalized Phase 3 / Phase 2.5 contract.