Amazon's Dynamo kept the cart always writable and tuned consistency last.

During peak Amazon traffic, tens of millions of customers use tens of thousands of servers, and the shopping cart has to accept anything they add. A database that refuses writes while it repairs a node would lose sales, so the requirement is availability, not strict consistency.

Dynamo stores data as key-value blobs, partitions and replicates them with consistent hashing, and versions every write so concurrent edits can be reconciled later with vector clocks. Reads and writes only need a quorum, and a 'sloppy' quorum lets a write land on a temporary node while the true owner is unavailable, then get copied back on recovery.

The result is a system that keeps accepting writes, returns a coherent value once the versions are reconciled, and reserves strict consistency for the small set of services that truly need it.