Chain Replication
You'll learn how chain replication delivers strong consistency and high throughput by ordering nodes in a linear chain.
The Core Idea
Chain replication organizes replica nodes in a strict linear order: a head, zero or more middle nodes, and a tail. All writes enter at the head and flow sequentially through each node until they reach the tail. Only the tail sends acknowledgments back to clients. Reads can be served directly from the tail.
This design is elegant: because every write must pass through the entire chain in order, all nodes see updates in exactly the same sequence. The tail always has the most recent committed state, giving you strong consistency without complex quorum logic.
How It Works
- Write arrives → Client sends write to the head
- Head processes → Applies update, forwards to next node
- Chain propagation → Each node applies update, forwards downstream
- Tail acknowledges → Only when tail applies the update does it respond to the client
- Read requests → Served directly from the tail (which has all committed writes)
Think of it like a factory assembly line: each station completes its task and passes the product forward. The last station (tail) is the only one that declares "done!" and ships to the customer.
Trade-offs
Strengths:
- Strong consistency guaranteed
- Good write throughput (pipeline parallelism)
- Tail always has complete, committed state for reads
Limitations:
- Write latency grows with chain length (must traverse all nodes)
- Head or tail failure requires reconfiguration
- Less flexible than quorum-based approaches
Key Takeaway: Chain replication provides strong consistency by serializing all writes through an ordered chain of nodes—simple, predictable, and efficient for workloads where linearizability matters more than minimum latency.