Foundations of distributed data · FND-03
Replication & consensus
- 3 h
- requires FND-01
From your world. Why Athena never 'loses' an S3 object; durability from replication and quorums vs. a single node
Explain to a data analyst who trusts S3 durability but has never heard of Raft:
"How do a bunch of unreliable machines agree on a single history of writes without ever electing two leaders — and why is that the same reason S3 never loses an acknowledged object?"
Rubric (self-check — aim for 5/6)
- States why replication is needed (single machines fail). States the new problem that it creates (which copy is the truth / split brain).
- Defines a quorum and the rule
W + R > N, and names the majority⌊N/2⌋ + 1. - Explains why majorities work: any two majorities overlap on at least one member.
- States the two election rules of Raft: one vote per term, and a majority to win. States that they give at most one leader per term.
- Connects to Kafka ISR /
acks=all/min.insync.replicasas a concrete quorum. - Maps to S3 durability (quorum-acknowledged replicated write) and explains why Athena never loses an acked object.