Overview
Composite backends combine several underlying stores so that load can be distributed or redundancy increased for critical data. They improve throughput and resilience in environments with high availability or scalability requirements.
Sharded In-Memory Store
Sharding distributes data across multiple backends to improve throughput and balance load. In the sharded in-memory store, each key is hashed and a modulus operation selects one of the configured backends to hold or serve the associated value. Each backend owns a deterministic subset of keys, which avoids hotspots and allows capacity to scale horizontally.
Sharded Blob Store
Sharding distributes data across multiple backends to improve scalability and balance load. In the sharded blob store, each blob's key is hashed and a modulus operation selects one of the configured backends to hold or retrieve it. Each backend therefore owns a deterministic portion of the blob namespace, and reads and writes for a given blob always resolve to the same backend.
SQL Read Replicas
SQL read replicas are a well-known database architecture pattern for separating read and write traffic. A primary database handles all write operations, ensuring consistent updates; one or more read replicas, synchronised from the primary, handle read traffic. This separation reduces load on the primary and allows the read path to scale horizontally.