Details
-
Improvement
-
Resolution: Done
-
Critical - P2
-
None
-
None
-
None
-
Replication
Description
Based on w= GLE, we should have a minimum number of secondaries reading the oplog directly from the primary.
The primary should compute the "running maximum" w (maybe the highest w in the last minute) and there should be that many secondaries reading directly, so that w= is always as low latency as possible AND if one goes down, it won't increase latency suddenly.