Sloppy Quorum and Hinted Handoff
You'll learn how Dynamo maintains availability during node failures by relaxing quorum rules and tracking temporary writes.
The Problem with Strict Quorums
In a strict quorum system, if you need to write to 3 specific replica nodes but one is down, the write fails. This hurts availability—exactly what Dynamo was designed to avoid.
What Is a Sloppy Quorum?
A sloppy quorum relaxes the rule: instead of requiring writes to reach specific replica nodes, Dynamo accepts writes to any available nodes in the cluster. If Node A is down, Node D (not normally a replica for this key) can temporarily accept the write.
Analogy: Imagine a package delivery where your neighbor signs for your parcel when you're not home, rather than the delivery failing entirely.
Hinted Handoff: The Cleanup Step
When Node D accepts a write meant for Node A, it stores a hint alongside the data: "This belongs to Node A." This is the hinted handoff.
Once Node A recovers, Node D automatically transfers the data back to its rightful owner and deletes its temporary copy. The system "heals" itself.
Why This Matters
- Always-writable: Writes succeed even during failures
- Availability over consistency: Temporary inconsistency is acceptable
- Self-healing: No manual intervention needed after recovery
- Trade-off: Data may temporarily live on "wrong" nodes
The Availability Promise
With sloppy quorum, Dynamo can accept writes as long as any W nodes in the entire cluster are available—not just the designated replicas. This is crucial for Amazon's "shopping cart must always work" requirement.
Key Takeaway: Sloppy quorum and hinted handoff sacrifice strict data placement for maximum availability, allowing writes to succeed during failures by temporarily storing data on substitute nodes with recovery hints.