Service architectures
Tair (Redis OSS-compatible) offers two instance architectures — standard and cluster — each with an optional read/write splitting mode. Choose 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
|
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 support dynamically enabling read/write splitting.
Choose standard if your dataset fits on a single node and your query rate is predictable. Choose cluster if you need horizontal scalability — either because data volume exceeds single-node capacity or because your workload requires parallelism across shards.
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 automatically.
When to use 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 master node's capacity, enable read/write splitting to scale read performance. This mode adds multiple proxy nodes and read-only nodes alongside the base master-replica pair.
Component roles:
-
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 and offload the master node.
When to use this configuration:
-
Read traffic is high and the workload is read-heavy.
-
You need persistent storage with the ability to scale read throughput.
Cluster architecture
In a cluster instance, data is distributed across multiple shards. Each shard uses a multi-node master-replica architecture, providing both horizontal scalability and high availability.
When to use cluster architecture:
-
Data volume exceeds single-node capacity.
-
Your workload requires high QPS.
-
Throughput-intensive operations benefit from parallel processing across shards.
Cluster architecture with read/write splitting
When read traffic exceeds the capacity of individual shard master nodes, enable read/write splitting for the cluster. 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 master node's capacity within individual shards.
-
Overall read throughput needs to scale beyond what master nodes alone can provide.
Editions, series types, and engine versions
Architecture is one of several dimensions that define a Tair instance:
|
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, see: