This page defines key terms for Tair (Redis OSS-compatible) to help you build a mental model of the product before exploring other documentation.
Tair resources are organized in a hierarchy: an instance is the top-level unit, each instance contains one or more data shards, and each shard consists of one or more nodes. Understanding this hierarchy helps you interpret configuration options and capacity limits across the documentation.
| Term | Description |
|---|---|
| Instance ID | The basic unit of Tair (Redis OSS-compatible). Each instance maps to a single user space with its own limits on connections, bandwidth, and CPU power. The exact limits depend on the instance type. Find your instance IDs in the Tair console. |
| Data sharding | A method of partitioning data across multiple shards to improve performance and scalability. The architecture you choose determines how shards are used:<br><br>- The standard architecture uses a single shard in a master-replica setup. All data is stored in one shard (the master node), with up to 9 replica nodes.<br>- The cluster architecture distributes data across 2 to 256 shards. The entire database space is divided into 16,384 slots, and each shard handles a subset of those slots. For example, in a 3-shard cluster: shard 1 handles slots [0, 5460], shard 2 handles slots [5461, 10922], and shard 3 handles slots [10923, 16383]. Each shard supports one master node and up to four replica nodes, all with identical specifications. |
| Instance type | Determines the node configuration and availability behavior of an instance:<br><br>- High Availability: Uses a master-replica architecture with two or more nodes. The master node handles all read and write requests. The replica node is reserved for high availability (HA) and does not serve external traffic. If the master node fails, the system promotes the replica with the most complete and up-to-date data within 30 seconds. A read replica handles read requests and provides disaster recovery capabilities; read replicas are available only in the read/write splitting architecture.<br>- Standalone: Contains a single data node with no replica. Data is not replicated in real time. Use this type for cache-only workloads that do not require high data reliability. Standalone instances are available at a relatively low price. |
| Primary (secondary) zone node | Applies to instances deployed across multiple availability zones. In normal operation, the master node (the primary zone node) runs in the primary zone, and replica nodes (secondary zone nodes) run in the secondary zone. For more information, see Regions and zones.<br><br>Zone distribution rules:<br>- 1 master + 2 replicas: 1 master and 1 replica in the primary zone; 1 replica in the secondary zone.<br>- 1 master + 3 or more replicas: At least 1 master and 1 replica must be in the primary zone. You can deploy the remaining replica nodes in the primary zone or the secondary zone based on your requirements. |
| Deployment mode | Controls which management architecture backs the instance:<br><br>- Cloud-native: A new-generation architecture with better scalability, elasticity, and more flexible specification configuration.<br>- Classic: The conventional management architecture.<br><br>For a detailed comparison, see Comparison between cloud-native instances and classic instances. |
| Storage medium | Tair supports the following storage mediums. Choose based on your performance and cost requirements:<br><br>- Redis Open-Source Edition — Stores data in memory. Best for standard open source Redis workloads.<br>- [DRAM-based instance](https://www.alibabacloud.com/help/en/redis/product-overview/dram-based-instances) — Stores data in memory using a multi-threaded model, delivering approximately 3x the performance of a Redis Open-Source Edition instance at the same specifications. Supports semi-synchronous replication, point-in-time recovery (PITR), and Global Distributed Cache. Best for ultra-high performance or active geo-redundancy.<br>- [Persistent memory-optimized instance](https://www.alibabacloud.com/help/en/redis/product-overview/persistent-memory-optimized-instances-1) — Stores data in persistent memory with command-level persistence. Best for workloads that require both high performance and strong data consistency.<br>- [ESSD/SSD-based instance](https://www.alibabacloud.com/help/en/redis/product-overview/essd-based-instances-1) — Stores data in Enterprise SSDs (ESSDs) or standard SSDs, delivering approximately 60% of Redis Open-Source Edition performance at a cost as low as 15% of a Redis Open-Source Edition instance. Provides command-level persistence and massive storage capacity. Best for workloads that require moderate performance and high cost-effectiveness. |
| Version compatibility | Tair is compatible with Redis 7.0, Redis 6.0, Redis 5.0, and Redis 4.0. |
| Eviction policy | Tair uses the same eviction policy as open source Redis. For details, see Key eviction. |
| DB | Short for database. Tair supports 256 databases (DB 0 to DB 255). Data is written to DB 0 by default. The total number of databases cannot be modified. |