CSE Roadmap

Specialist tier · about 2 hours · 5 parts

Part 4 of 5 · 2 min read

Why no topology is universally correct

Each topology solves a specific availability problem by accepting a specific complexity cost; picking one means accepting its particular cost.

Why no topology is universally correct

Think of choosing a vehicle for a trip. A small efficient car, a larger vehicle with more capacity, and an off-road vehicle each serve a real trip well. None of them serves every possible trip best.

In plain words: single-leader replication's simplicity becomes a real liability precisely when writers are geographically distributed or when the leader's availability cannot be guaranteed. Multi-leader and leaderless replication's improved availability becomes a real liability precisely when an application cannot tolerate conflicting writes or stale reads. There is no topology that maximizes both simplicity and availability at once, because the added coordination that availability requires is exactly what simplicity gives up.

A worked example. A banking system handling account balances typically prioritizes avoiding conflicting writes over maximum write availability, favoring single-leader replication or a leaderless system configured with strict quorum sizes. A system logging high-volume sensor readings from many distributed devices, where losing strict ordering briefly is tolerable, may prioritize availability instead.

The key lines. No new SQL; this comparison is the summary judgment a system architect reaches after weighing the earlier four pages' specific trade-offs.

Remember this

There is no universally best replication topology. Single-leader trades availability for simplicity; multi-leader and leaderless trade simplicity for availability; the correct choice is whichever cost the specific system can actually afford.

Quick notes

  • Single-leader replication offers write order simplicity at the cost of a single point of failure for writes.
  • Multi-leader replication improves write availability and latency for distributed writers at the cost of conflict resolution.
  • Leaderless replication maximizes write availability at the cost of needing quorum-based reasoning about correctness.
  • The right topology depends on which cost, lost simplicity or lost absolute write availability, the specific system can tolerate.
  • No topology is correct for every system; the choice follows from the system's actual requirements.