website/docs/azure/overview.md
Azure Cosmos DB Garnet Cache is a fully managed, high-performance caching service built on the Garnet remote cache-store from Microsoft Research. It provides enterprise-grade reliability, security, and scalability without the operational overhead of managing your own cache infrastructure. With consistent low latency and high throughput even with many client connections, Azure Cosmos DB Garnet Cache accelerates data access and leads to cost savings for large apps and services.
The Azure Cosmos DB Garnet Cache is currently in an expanded Private Preview. Please register your subscription.
Azure Cosmos DB Garnet Cache is a cloud-native caching service that combines the performance advantages of Garnet with Azure's managed service capabilities. Unlike traditional single-threaded caches or caches that are hash-partitioned within each node, Garnet uses a shared-everything architecture within nodes where all threads directly access the single shared memory space. This architectural advantage allows Garnet to maintain sub-millisecond latencies even under heavy load with thousands of concurrent client connections and translates to cost savings for applications with many simultaneous users. It offers:
Azure Cosmos DB Garnet Cache supports distributed caching across multiple application instances and is designed for workloads where throughput efficiency and data durability matter. Here are scenarios where Garnet's unique advantages deliver the most value:
Live trading and financial tick systems, multiplayer and gaming services, and IoT platforms with thousands of simultaneous connections. Garnet's multi-threaded architecture maintains sub-millisecond latency and consistent throughput even under heavy concurrent load, eliminating the performance degradation single-threaded caches experience with many parallel requests.
Applications where losing cached data is not acceptable like user sessions, payment information, and transaction data. Garnet's non-blocking AOF persistence eliminates the traditional tradeoff between persistence and throughput. It lets you durably cache critical data without sacrificing performance.
Vector search for recommendation engines, semantic search, and AI applications. Store and query high-dimensional vectors using VectorSet data structures with DiskANN indexing for single-digit millisecond vector search at scale.
Applications where high throughput requirements drive infrastructure costs. Garnet's multi-threaded efficiency means you can handle the same throughput with smaller SKUs or fewer cache nodes than other solutions, directly reducing operational costs and complexity.
| Feature | Support |
|---|---|
| Latency | 3ms P99, < 1ms P50 |
| Size | 5TB+ with clustering |
| Scaling | Horizontal scaling with sharding and replication or scale up SKU size |
| Availability | 99.99%* |
| Data persistence | Append only file (AOF) checkpointing |
| Advanced data structures | Support for Hash, Set, Sorted Set in addition to String |
| Vector search | VectorSet support with DiskANN indexing |
| Scripting | Lua scripting |
| Pub/Sub | Publish/subscribe messaging |
| Authentication | Microsoft Entra ID RBAC |
| Network isolation | Virtual network support with no public internet access |
| Encryption | At rest and in transit with TLS |
| Monitoring | Azure Monitor Metrics |
| Updates | Automatic updates with zero downtime |
*This is an estimated value. Actual availability varies depending on configuration. See high availability for more information.
Just like self-hosted Garnet, Azure Cosmos DB Garnet Cache uses the Redis RESP protocol, making it compatible with existing Redis clients and tools. You can migrate from Redis or other cache solutions with minimal code changes. Azure Cosmos DB Garnet Cache supports a subset of the self-hosted Garnet commands including Strings, Hashes, Sets, Sorted Sets, Pub/Sub, Lua scripting, and more. See the full list of supported commands.
Azure Cosmos DB Garnet Cache offers two performance tiers to match your workload requirements. An overview of single-node specs is below for each tier based on the available SKUs. Each cluster can be scaled to have one or more nodes of the same SKU determined by the shard count and replication factor configured. This allows for custom cache sizes up to 5TB+ that match your specific workload needs.
Recommended for balanced workloads, general caching, development and testing. Single-node specs ranging from:
Recommended for in-memory databases, large datasets, gaming leaderboards, vector search workloads. Single-node specs ranging from:
Ready to get started? First, register your subscription for access to the preview. Then, check out our quick start guide to create your first Azure Cosmos DB Garnet Cache instance in minutes.
Register your subscription to provision Azure Cosmos DB Garnet Caches. Here are the steps to request access. Please note that registration is not automatic and can take up to one week to process.