⚖️ Cassandra Administration – Consistency Levels & Replication
20 questions · 15 min timer · Explanations included · Free
Cassandra doesn’t decide consistency for you. You decide it per query, and the decision is a trade between how fresh the data is, how fast the query returns, and how much of the cluster has to be alive for it to work at all.
The rule underneath every question here: R + W > RF guarantees you read your last write. With a replication factor of 3, QUORUM reads and QUORUM writes give you 2 + 2 > 3, so it holds. ONE and ONE gives you 1 + 1, and it doesn’t.
What this quiz covers
- ONE, TWO, QUORUM, ALL — throughput and latency against staleness risk
- LOCAL_ONE and LOCAL_QUORUM — why multi-datacentre deployments almost always use the LOCAL variants, and what cross-DC latency does when you don’t
- EACH_QUORUM — acknowledgement from every datacentre, and the availability cost
- ALL — strict consistency, and why one dead replica takes the whole operation down
- Availability during partitions — which levels keep serving and which stop
Common mistakes
- Calling Cassandra “eventually consistent” without qualification — consistency is tunable, and that’s the actual answer
- Using QUORUM in a multi-DC cluster where LOCAL_QUORUM was intended, and paying cross-region latency on every request
- Choosing ALL for correctness, then discovering it has no fault tolerance at all
Twenty questions below, each with an explanation.
