All Three or Bust

A service uses N = 3, W = 3, R = 1 with strict quorums and no sloppy quorum. During a rolling restart, exactly one replica of each shard is offline for about a minute at a time. What should the team expect?

A

Reads fail intermittently, because R = 1 cannot satisfy R + W > N while a replica is restarting

B

Writes fail for the whole restart window while reads keep succeeding from any one live replica

C

Both reads and writes fail, because a quorum always needs a majority of the replicas to be live

D

Writes succeed and hinted handoff replays them to the restarting replica once it comes back