Tair (Redis OSS-compatible) provides two instance architectures -- standard and cluster -- each with an optional read/write splitting mode. Choose an architecture based on your data volume, throughput requirements, and read/write ratio.
Key concepts
A Tair instance is built from the following components:
| Component | Description |
|---|---|
| Node | The smallest unit of a Tair deployment. Each node runs a Redis-compatible process. |
| Shard | A grouping of nodes that stores a subset of data. In a cluster instance, data is partitioned across multiple shards. |
| Master node | The node that handles write operations within a shard or standard instance. |
| Replica node | A copy of the master node that provides failover capability. If the master node fails, workloads switch over to the replica node. |
| Read-only node | An additional node that serves read traffic only. Available when read/write splitting is enabled. |
| Proxy node | A routing layer that distributes read and write requests to the appropriate nodes. Used in cluster proxy mode and read/write splitting configurations. |
Choose an architecture
The following table summarizes the differences between standard and cluster architectures.
| Dimension | Standard architecture | Cluster architecture |
|---|---|---|
| Internal structure | One master node and one or more replica nodes | Multiple shards, each with its own master node and one or more replica nodes |
| Data partitioning | No. All data resides on a single shard. | Yes. Data is distributed across shards. |
| Best for | Small data volumes with stable query rates | Large data volumes, high QPS, or throughput-intensive workloads |
| Read/write splitting | Supported. Add read-only nodes and proxy nodes. | Supported. Add read-only nodes per shard. |
Both architectures allow you to dynamically enable read/write splitting.
Standard architecture
A standard instance uses a master-replica architecture. The master node handles all read and write operations, while the replica node maintains a real-time copy of the data. If the master node fails, workloads switch over to the replica node to maintain high availability.
When to use the standard architecture:
Your data fits on a single instance.
Your query rate is stable and does not exceed single-node capacity.
You need persistent storage with high availability.
Standard architecture with read/write splitting
When read traffic exceeds the capacity of the master node, enable read/write splitting to scale read performance. A standard read/write splitting instance adds multiple proxy nodes and read-only nodes to the base master-replica pair.
Proxy nodes route write requests to the master node and distribute read requests across the master node and read-only nodes.
Read-only nodes serve read traffic, offloading the master node.
When to use this configuration:
High QPS with a read-heavy workload.
Persistent storage on instances.
Cluster architecture
In a cluster instance, data is partitioned across multiple shards. Each shard uses a multi-node master-replica architecture, which provides both horizontal scalability and high availability.
When to use the cluster architecture:
Large data volumes that exceed single-node capacity.
High QPS requirements.
Throughput-intensive workloads that benefit from parallel processing across shards.
Cluster architecture with read/write splitting
For cluster instances where read traffic exceeds the performance limit of the master node in individual shards, enable read/write splitting. Each shard switches from a standard master-replica configuration to a read/write splitting architecture, with dedicated read-only nodes to handle read traffic.
When to use this configuration:
Read traffic exceeds the performance limit of the master node within individual shards.
Overall read performance needs to scale beyond what master nodes alone can handle.
Editions, series types, and engine versions
Architecture is one of several dimensions that define a Tair instance. The following dimensions also apply:
| Dimension | Examples |
|---|---|
| Edition | Redis Open-Source Edition, Tair (Enterprise Edition) |
| Series type | DRAM-based instances |
| Engine version | Redis 5.0, Redis 7.0 |
The architecture descriptions on this page apply across all editions, series types, and engine versions. For detailed specifications of each architecture, see the following topics: