Why Tendis Offers High‑Performance, Redis‑Compatible Distributed KV Storage

Overview

Tendis is an open‑source distributed key‑value store developed by Tencent. It implements the Redis protocol and data structures while providing petabyte‑scale capacity, low cost, and strong persistence suitable for warm‑cold data workloads.

Cluster Architecture

The cluster follows a decentralized design: every data node stores the complete routing table, so a client can connect to any node. The Redis MOVE protocol is used to forward a request to the node that owns the target slot.

Each node is assigned a distinct slot range; slot ranges never overlap between master nodes. Masters can migrate slots among themselves, enabling online rebalancing. Master‑slave replication is performed via a binlog mechanism.

All nodes exchange state through a gossip protocol similar to Redis Cluster. Optional hashtags can be attached to keys to influence slot assignment, allowing fine‑grained data placement with minimal operational overhead.

Typical Use Cases

Applications that require Redis‑compatible interfaces but need larger capacity and higher read/write throughput for warm‑cold data.

Cost‑sensitive services that also demand strong durability guarantees.

Workloads where Redis’s fork‑based memory model becomes a bottleneck.

Key Features

Redis Compatibility

Supports the majority of native Redis commands and data structures, providing drop‑in compatibility for existing Redis clients.

Persistent Storage

Uses RocksDB as the underlying storage engine. Data are stored in a custom format inside RocksDB, enabling storage capacities at the petabyte level.

Decentralized Design

Gossip‑based communication and optional hashtags give flexible data distribution while keeping management complexity low.

Horizontal Scaling

Nodes can be added or removed without service interruption. Slots can be migrated between any two nodes, allowing transparent scaling up to roughly 1,000 nodes.

Automatic Failover

Failed masters are detected automatically; a slave is promoted to master to maintain availability.

Hybrid Hot‑Cold Storage

The architecture allows the native Redis engine and the Tendis storage engine to coexist. Cold data can be offloaded to the RocksDB‑backed layer, reducing storage cost by up to 80 % while preserving hot‑data latency.

Roadmap

Further improve Redis compatibility, including full Lua script support.

Optimize overall performance and reduce latency spikes originating from RocksDB.

Explore hardware‑accelerated solutions to increase throughput.

Develop heterogeneous data interchange capabilities to simplify migration from other databases.

Enhance PaaS‑style operational management and platform features.

Open‑Source Repository

https://github.com/Tencent/Tendis Domestic mirror: https://git.code.tencent.com/TencentOpenSource/Tendis