-
Type:
Task
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
-
Storage Execution
-
None
-
None
-
None
-
None
-
None
-
None
-
None
FCV downgrade is the clean disable for set-wide Replicated RecordIds: every node drops its groupId collection, fenced at the downgrade entry.
Details
- The trigger is the transition from upgraded to downgrading, the first observable entry of the downgrade, not its completion: the primary stops emitting rids and clears its own groupId at that point, and secondaries clear when applying the entry, so the whole downgrade window is fenced.
- The clear DROPS the collection rather than deleting the document: the table is a local collection special-cased to logged=false, and last-LTS binaries expect local tables logged=true and (validateTableLogging would flag it after binary downgrade). Dropping leaves nothing for the old binary to misinterpret.
- Binary downgrade requires FCV downgrade first, so no old binary ever runs with a groupId on disk. Downgrade targets from FCV 9.0: 8.0 (lastLTS) and 8.3 (lastContinuous).
Acceptance criteria
- Downgrade drops the table on primary and secondaries, including concurrent or failed setFCV.
- Binary downgrade after FCV downgrade finds no table on disk and validateTableLogging is clean.
- Re-upgrade re-founds and requires resync, covered by the dedicated test.
- depends on
-
SERVER-138108 Expose RecordIds groupId membership in hello and replSetGetStatus
-
- Open
-
- is depended on by
-
SERVER-138124 Membership-based recordId emission and honoring in the oplog path
-
- Open
-