Tair Global Distributed Cache - geo-replication, active geo-redundancy, disaster recovery - Tair (Redis® OSS-Compatible)

Tair (Redis OSS-compatible) provides Global Distributed Cache, a geo-replication system that synchronizes data across off-site instances for active geo-redundancy and disaster recovery. A Global Distributed Cache instance consists of up to three child instances that synchronize data in real time, reducing the physical distance between data and users to lower access latency.

Benefits and scenarios

For geographically distributed businesses, cross-region access increases latency and degrades user experience. The Global Distributed Cache feature of Tair (Enterprise Edition) resolves these issues. The Global Distributed Cache feature offers the following benefits:

Benefit	Description
High reliability	Supports resumable transfers from breakpoints. Tolerates synchronization interruptions lasting days, overcoming native Redis limitations for incremental sync across data centers or regions. Automatically handles anomalous activity, such as primary/secondary failovers and backup-based reconstruction of child instances.
High performance	High throughput: A synchronization channel in a standard architecture instance achieves up to 50,000 QPS in one direction. Cluster architecture throughput scales linearly with the number of shards or nodes. Low latency: Intra-continent synchronization latency is less than 1 second on stable networks. Cross-continent latency averages 1 to 5 seconds, depending on link throughput and round-trip time (RT).
High correctness	Synchronizes binary logging data to peer nodes in generation order. Loopback control prevents circular synchronization of binary logging data. Exactly-once processing ensures synchronized binary logging data executes only once.

Scenarios

Common in multimedia, gaming, and e-commerce for cross-region data synchronization and global deployments.

Scenario	Description
Active geo-redundancy	Multiple sites in different regions serve traffic simultaneously, enabling nearest-access routing. This high availability (HA) design keeps all sites active at all times.
Data disaster recovery	Two-way synchronization between child instances supports intra-city disaster recovery, three data centers across two regions, and three-region disaster recovery.
Load distribution	Distribute traffic across multiple child instances to exceed the load limit of a single instance during high-traffic events such as large-scale promotions.
Data synchronization	Two-way data synchronization between child instances supports use cases such as data analytics and testing.

Architecture

A Global Distributed Cache instance of Tair (Enterprise Edition) is a logical collection of child instances connected by synchronization channels. All data in each child instance synchronizes in real time at the instance level. The following figure shows the architecture.

Components:

Component	Description
Child instance	An independent instance within a Global Distributed Cache instance. Each child instance supports read and write operations with its own endpoint. Child instances maintain eventual consistency through real-time, two-way synchronization. Note Child instances must be memory-optimized Tair (Enterprise Edition) instances. The Redis Open-Source Edition does not support this feature.
Synchronization channel	A one-way link for real-time data synchronization between child instances. Bidirectional replication uses two synchronization channels in opposite directions. The Global Distributed Cache feature of Tair (Enterprise Edition) adds server-id and opid to native Redis AOF logs. The synchronization channel reads this binary logging data for synchronization.
Channel Manager	Manages the lifecycle of synchronization channels and handles anomalous activity on child instances, such as primary/secondary failovers and backup-based reconstruction.

Note

Cross-border synchronization between the Chinese mainland and other regions is not supported. All child instances in a Global Distributed Cache instance must be located entirely within the Chinese mainland or entirely in other regions. Limits of Global Distributed Cache.

Billing

This feature is free of charge. You pay only for the memory-optimized Tair (Enterprise Edition) child instances. Billing items.

Get started

Important

The cloud-native Global Distributed Cache feature is available on a whitelist basis. Alibaba Cloud assesses the network capabilities of your intended regions before granting access. To apply, submit a ticket.

Create the first child instance. You can convert an existing memory-optimized Tair (Enterprise Edition) instance or create a new one.

The Global Distributed Cache instance is automatically created with the first child instance.
Add a second or third child instance to the Global Distributed Cache instance.
In your application code, route requests to the nearest child instance endpoint for each region.

FAQ

Q: Can I upgrade the minor versions of all child instances in a Global Distributed Cache instance at the same time?

A: Stagger upgrades by more than 30 minutes between child instances to avoid simultaneous service interruptions.
Q: Can I change the architecture of a child instance, such as from standard to cluster?

A: No, you cannot. You must plan your architecture in advance.
Q: Can I change the instance type of a child instance, such as upgrading a standard architecture instance from 8 GB to 16 GB?

A: Yes. All child instances must use the same instance type. Apply the same upgrade or downgrade to every child instance to avoid performance or capacity issues.
Q: Does a Global Distributed Cache instance support multi-write?

A: Yes. You can write to different child instances, but avoid modifying the same key on multiple child instances simultaneously or in close succession to prevent data inconsistency. Data consistency limits.