Build fluency in the vocabulary of a replication design where any replica can accept a write directly.
0 / 5 completed
1 / 5
At standup, a dev mentions a distributed database design where any replica can accept a read or a write directly, rather than every write being routed through one elected leader. What is this replication style called?
Leaderless replication, in the style popularized by Amazon's Dynamo, lets any replica accept a read or a write directly, rather than routing every write through one elected leader. Single-leader replication routes every write through one node, which becomes a bottleneck and a single point of failure for writes specifically. This leaderless design is what lets a Dynamo-style database keep accepting writes even if some replicas, or even the usual coordinator, are temporarily unreachable.
2 / 5
During a design review, the team wants a write that was intended for an unreachable replica to be temporarily stored by another available node, then forwarded once the original replica rejoins. Which capability supports this?
Hinted handoff temporarily stores a write intended for an unreachable replica on another available node, then forwards it once the original replica rejoins the cluster. Discarding the write outright whenever its intended replica is unreachable would lose data that the client believed had already succeeded. This handoff mechanism is what lets a leaderless system keep accepting writes during a temporary replica outage without silently dropping them.
3 / 5
In a code review, a dev notices a read operation that finds two replicas returning conflicting values for the same key pushes the newer value back out to the stale replica before returning a result to the client. What does this represent?
Read repair reconciles a stale replica's value during a read by pushing the newer, correct value back out to whichever replica was found to be behind, using each value's version to determine which is more recent. Returning whichever conflicting value happens to be read first ignores the divergence entirely and leaves the stale replica wrong for future reads too. This repair-on-read mechanism is one of the ways a leaderless system gradually heals a replica that's fallen behind.
4 / 5
An incident report shows a client received a stale value on read even though a very recent write had already succeeded, because the read only consulted a single replica instead of enough replicas to satisfy the system's quorum guarantee. What practice would prevent this?
Requiring a read to consult enough replicas that its read quorum and the prior write's quorum are guaranteed to overlap ensures at least one consulted replica has the most recent value. Allowing a read to consult only a single replica, with no such quorum overlap guaranteed, risks exactly the stale read this incident describes. This quorum-overlap requirement, commonly expressed as read-quorum plus write-quorum exceeding the total replica count, is the core consistency guarantee a leaderless system relies on.
5 / 5
During a PR review, a teammate asks why the team chose leaderless replication for a write-heavy, globally distributed workload instead of single-leader replication routing every write through one elected node. What is the reasoning?
Single-leader replication makes that one node a bottleneck and a single point of failure for every write, which matters a great deal for a write-heavy, globally distributed workload spread across regions. Leaderless replication lets any available replica accept a write even during a partial outage or a network partition, keeping writes flowing where a single-leader design would stall. The tradeoff is the added complexity of reconciling conflicting writes and tuning quorum settings to get the consistency guarantee the workload actually needs.