Part 5 of 5 · 4 min read
Exercises
Easy
Easy 1: Name the three topologies (15 minutes)
List the three replication topologies covered across this level, in order from fewest to most nodes able to accept writes.
Done when: single-leader, multi-leader, and leaderless are correctly listed in that order.
Hint
Single-leader, multi-leader, leaderless, in order from fewest to most nodes able to accept writes.
Easy 2: State single-leader's main cost (15 minutes)
State the main cost single-leader replication accepts in exchange for its simplicity.
Done when: the leader becoming a single point of failure, or a bottleneck, for writes is correctly named.
Hint
The single designated leader becomes a single point of failure, or a bottleneck, for all writes.
Easy 3: State leaderless replication's main cost (15 minutes)
State the main cost leaderless replication accepts in exchange for its write availability.
Done when: needing quorum-based reasoning, and the risk of reads missing the most recent write, is correctly named.
Hint
It needs quorum-based reasoning about correctness, and reads can occasionally miss the most recent write.
Medium
Medium 1: Match a scenario to a topology (25 minutes)
A company runs an internal inventory system used only by employees in one warehouse, where brief write downtime during a failover is fully acceptable. Recommend a topology.
Done when: single-leader replication is recommended, correctly reasoning that write availability during a rare failover matters less than the system's simplicity here.
Hint
Single-leader replication, since the single-location use case tolerates brief failover downtime and does not need the complexity of multiple writable leaders.
Medium 2: Match a different scenario to a topology (25 minutes)
A company runs a collaborative document editor used by small teams spread across three distant offices, where each office needs fast local writes. Recommend a topology.
Done when: multi-leader replication is recommended, correctly reasoning that each office benefits from writing to a nearby leader rather than waiting on one distant leader, accepting the resulting conflict-resolution cost.
Hint
Multi-leader replication, since each distant office benefits from a nearby leader rather than waiting on one far-away leader for every write.
Medium 3: Explain why a single topology cannot serve both scenarios well (30 minutes)
Using the two scenarios above, explain why single-leader replication would serve the second scenario poorly, and why multi-leader replication would be unnecessary complexity for the first.
Done when: the explanation correctly notes that single-leader would force the second scenario's distant offices to wait on one leader, while multi-leader would add unneeded conflict-resolution complexity to the first scenario's single-location use case.
Hint
Single-leader would make the second scenario's offices wait on one distant leader, hurting latency; multi-leader would add unneeded conflict-resolution overhead to the first scenario's single-location system.
Hard
Hard 1: Recommend a topology for a strict-consistency financial system (30 minutes)
A system processing account transfers must never accept two conflicting writes to the same account balance. Recommend a topology and justify it against the alternatives.
Done when: single-leader replication, or a leaderless system configured with strict quorum sizes guaranteeing overlap, is recommended, with the justification correctly explaining that multi-leader's conflict-prone nature is specifically unsuitable for account balances.
Hint
Single-leader, or a strictly-configured leaderless system, since account balances cannot tolerate multi-leader's conflicting-write risk.
Hard 2: Recommend a topology for a globally distributed, high-write-volume logging system (35 minutes)
A system ingests sensor readings from devices on every continent, at high volume, where losing strict write ordering briefly is tolerable but losing write availability during a regional outage is not. Recommend a topology and justify it.
Done when: leaderless replication, or multi-leader replication, is recommended, with the justification correctly explaining that this system's tolerance for brief inconsistency, paired with its need for continuous write availability across regions, favors availability-maximizing topologies over single-leader's simplicity.
Hint
Leaderless or multi-leader replication, since continuous write availability across regions matters more here than strict write ordering.
Hard 3: Explain the general pattern connecting topology choice to the five pages in this level (40 minutes)
Explain, in general terms, how every topology choice in this level reduces to the same underlying trade-off between write availability and coordination complexity, referencing at least two of the earlier pages in this level.
Done when: the explanation correctly identifies the shared trade-off, availability against coordination complexity, and correctly connects it to at least two earlier pages, such as replication lag's consistency guarantees or quorum sizing's overlap guarantee, showing both as instances of the same underlying pattern.
Hint
Every choice trades some amount of coordination complexity for some amount of write availability; replication lag's consistency guarantees and quorum sizing's overlap guarantee are both specific mechanisms for managing that same underlying trade-off.