2 / 6 Quorums, Replication & Read Repair

The Keys Nobody Reads

An audit-log keyspace is written once per record and only about 1% of records are ever read back. One replica was offline for six hours, well past the window in which hints were retained, so it missed a large batch of writes. Reads that do happen are being repaired correctly. What is the right fix for the rest?

A

Lower W so the write path stops depending on the replica that fell behind during the outage

B

Run anti-entropy repair over the key ranges so cold keys are resynced without being read

C

Raise R toward N so that future reads pull the missing records from a fully caught-up replica

D

Leave it alone, since read repair converges every key given enough client traffic over time