Scaling data / Advanced / 8 min
Replication and quorums
Pick N, W and R. Watch stale reads appear and vanish, and what a failed replica does to availability.
Start the simulation → Change the load, break a part, watch what happens. Free, no sign-up.
In short
- What it is
- Each piece of data is copied to N replicas; a write needs W of them to confirm and a read asks R of them.
- Why it matters
- Copies survive machine and region failures and let more machines serve reads.
- The trade-off
- If W + R > N every read sees the latest write but is slower; smaller W and R are faster but reads can be stale (eventual consistency).
- Say it in an interview
- "With N = 3, W = 2, R = 2 the read and write sets always overlap, so reads are never stale and one node can be down."
Check yourself
N = 3. Which W and R guarantee a read sees the latest successful write?
Answer: W = 2, R = 2. You need R + W > N, so every read set overlaps every write set in at least one replica. 2 + 2 = 4 > 3.
N = 3, W = 3. One replica fails. What happens to writes?
Answer: They are rejected until the replica returns. Writes need acknowledgements from all three replicas, so losing one makes writes unavailable.
What does read repair do?
Answer: When a read sees an old copy, it writes the newer value back to that replica. Reads compare versions from several replicas and fix the out-of-date ones on the spot.