-
Type:
Task
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: None
-
None
-
Replication
-
Repl 2026-11-09, Repl 2026-11-23
-
None
-
None
-
None
-
None
-
None
-
None
-
None
In WRITING-40450 we came up with proposed changes in order to avoid a replaced/restarted node from going back in time and affecting the majority commit point of the replset.
Changes taken from the decision doc:
- Server changes so that a node in STARTUP2 cannot vote.
- Server changes so that initial sync (LIS or FCBIS) is not considered completed until it is caught up to the primary’s view of max(lastCommitted, my lastWritten) (from the beginning of initial sync).
- Server changes so that a node seeded with a snapshot (any node already in the replset) starts with a catchup phase (in STARTUP2 state) to get to the current primary’s view of max(lastCommitted, my lastWritten) (the first observed majority commit point as of restart).
We should model these changes in TLA+ and verify that they are viable.
Once the changes are modeled in TLA+ we should write a 1-pager document presenting the results and review it with the Replication team to get consensus around these changes before implementing them.
The implementation of these changes will be done in follow-up tickets.
This is a follow-up from WRITING-39616.
- is depended on by
-
SERVER-135299 Node should not vote during STARTUP2
-
- Open
-
-
SERVER-136918 Initial sync should not be considered done until it is caught up to the known majority commit point
-
- Open
-