5.3 Partition Routing Components

This is a historical learning note and may contain outdated or incomplete understanding.

A partitioned system needs to map a logical partition to its current node. The routing responsibility can live in several places.

1. Client-Side Routing

The client caches partition-to-node metadata and sends requests directly to the owner.

Pros: one network hop after metadata is known; no central data-plane proxy.

Cons: every client library must handle metadata refresh, topology changes, retries, and compatibility.

2. Proxy Routing

Clients send requests to a routing tier, which resolves ownership and forwards them.

Pros: simple clients and centralized policy.

Cons: another network hop and a proxy tier that must scale and remain highly available.

3. Server-Side Routing

A client contacts any cluster node. That node either forwards the request to the owner or returns a redirect/topology hint.

Pros: simple discovery and topology logic stays inside the cluster.

Cons: misrouted requests add latency and internal traffic until clients learn the topology.

Real systems often combine these approaches: clients cache metadata, any server can return corrected ownership information, and proxies may exist at organizational boundaries.