Course contentsShow
← System Design
Lesson 533 of 1,91915. Consistency ModelsFree lesson

Convergence Guarantees

Learn what 'eventually' means: bounded time vs unbounded, and how systems provide convergence guarantees.

Convergence Guarantees

What you'll learn: How "eventually consistent" systems define what "eventually" actually means—and why that matters.

When we say a system is "eventually consistent," we're making a promise: given enough time without new writes, all replicas will converge to the same value. But what does "enough time" mean? This is where convergence guarantees come in.

Bounded vs Unbounded Convergence

Unbounded convergence means the system will converge, but provides no upper limit on how long it takes. It could be milliseconds, seconds, or theoretically forever if writes never stop. Most basic eventually consistent systems operate this way—they guarantee convergence only when the system stabilizes.

Bounded convergence provides a concrete promise: "All replicas will converge within X milliseconds after the last write." This is much stronger and more useful for application logic, but harder to implement in distributed systems with variable network delays.

Real-World Analogies

Think of gossip spreading through a town. Unbounded: "Everyone will eventually hear the news." Bounded: "Everyone will hear within 24 hours."

Or consider package tracking. Unbounded: "Your tracking info will eventually update." Bounded: "Status updates appear within 5 minutes of scan events."

Why It Matters

Without clear convergence guarantees, applications can't make reliable decisions. If you're displaying data from replicas, unbounded convergence means users might see stale data for unpredictable periods. Bounded guarantees let you add UI hints like "Data is at most 2 seconds old."

Systems like Cassandra offer tunable consistency—you can trade stronger convergence guarantees (more replicas must acknowledge) for higher latency.

Key Takeaway: Convergence guarantees transform "eventually" from a vague promise into something measurable—either with time bounds or conditions under which convergence occurs.