Memorystore

Memorystore is Google Cloud's managed Valkey and Redis service, an in-memory key-value database for cloud apps, billed by the hour on the capacity provisioned.

Memorystore takes the running of Valkey and Redis off an engineering team's hands, so an application gets a fast in-memory store without anyone patching servers or wiring up failover. Google builds and operates it as part of Google Cloud, and it is reached from the same account and console as the rest of a team's cloud projects. The service comes as four engine offerings: Memorystore for Valkey, Memorystore for Redis Cluster, Memorystore for Redis and Memorystore for Memcached. Each one runs the engine it is named after. Memorystore for Valkey runs in Cluster Mode Enabled, which shards the keyspace across nodes behind a discovery endpoint, or in Cluster Mode Disabled. Replicas stream from each primary and take over automatically during maintenance or when a shard fails. RDB persistence saves snapshots to durable storage on a schedule, while AOF persistence logs every write command. Cross-region replication copies an instance to secondary instances in other regions. Valkey instances also handle JSON documents, bloom filters and vector search beside hashes, sorted sets and streams. Memorystore for Redis Cluster brings the sharded design to Redis, and Memorystore for Redis runs standalone instances in a Basic or a Standard tier. Memorystore for Memcached is being retired, with shutdown set for January 31, 2029. The service suits engineering teams whose applications already run on Google Cloud and who want caching or a session store without operating the machines behind it. Memorystore for Valkey accepts the Valkey command set apart from a published list of blocked commands, which matters to teams bringing existing code. Developers building generative AI features use its vector search and the LangChain integrations to store embeddings and retrieve context for a model. Security teams govern access through ACL policies, IAM authentication without static passwords, TLS in transit and customer-managed encryption keys. Applications use Memorystore to cache hot data in front of a primary database and to hold user sessions. Services publish events over pub/sub channels, or queue work in streams that consumer groups share out among several workers. Scripts and functions run atomically on the server when one change has to touch several keys. Clients on Compute Engine virtual machines and other Google Cloud workloads connect privately through Private Service Connect, while IAM decides which principals may sign in. For a retrieval-augmented AI application, a single Valkey instance can hold the vector index next to the ordinary cached keys the rest of the app reads.