-
Type:
Task
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
-
Replication
-
None
-
None
-
None
-
None
-
None
-
None
-
None
A groupless primary (feature flag on, latest FCV, attached storage) mints the groupId at election and announces it in the new-primary noop
Details
- Mint: UUID::gen(); foundingOpTime is the new-primary noop's opTime. Both written in the same WUOW as the (noop).
- Extend the noop payload with groupId and foundingOpTime. Old binaries ignore them (appliers skip noop payloads).
- A primary that already holds a groupId re-announces it. Only a groupless primary mints.
- Add an isNewPrimaryNoop-style accessor for the fields.
- Audit consumers that match the noop by exact object equality and replace them with isNewPrimaryNoop.
- Provider gating: minting and announcement run only on attached storage. Gate on !rss::ReplicatedStorageService::get(opCtx).getPersistenceProvider().shouldUseReplicatedRecordIds() (plus flag on + latest FCV). On disaggregated storage rid replication is provider-driven and unconditional, so no mint ever happens, no groupId table is created, and the noop carries no fields.
- Create the collection with createRecordIdsGroupIdCollection before opening the noop's WUOW (untimestamped catalog op, own WUOW), then setRecordIdsGroupId inside the noop's WUOW so the mint commits at the noop's timestamp, atomically with the fence. The write is insert-only and fasserts if a document already exists since a primary already holding a groupId re-announces it without rewriting.
Acceptance criteria
- Noop carries (groupId, foundingOpTime) exactly when the primary holds one.
- Unit test: mint-on-first-election, re-announce on subsequent elections, no mint when the flag is off.
- Unit test: no mint, no table create and no noop fields on disaggregated storage.
- depends on
-
SERVER-138433 Effective membership accessor + disagg sentinel
-
- In Code Review
-
-
SERVER-138105 Create the RecordIds groupId consistency marker collection
-
- Closed
-
- is depended on by
-
SERVER-138108 Expose RecordIds groupId membership in hello and replSetGetStatus
-
- Open
-
-
SERVER-138110 Clear groupId on mismatched new-primary noop (C1), with batcher fence
-
- Open
-
-
SERVER-138124 Membership-based recordId emission and honoring in the oplog path
-
- Open
-
-
SERVER-138231 Logical initial sync joins the RecordIds group
-
- Open
-