How Xianyu’s IFTTT Engine Boosts Real‑Time Two‑Way User Interaction

Alibaba’s Xianyu platform introduced an IFTTT‑based system to address two main issues: insufficiently rich user relationship layers and one‑way, non‑real‑time interactions between buyers and sellers.

IFTTT Concept

IFTTT (If This Then That) consists of three parts—Trigger, Action, and Recipe—where a satisfied condition (Trigger) initiates a corresponding action.

The power of IFTTT lies in chaining many simple flows into a complex, cross‑platform state machine.

Xianyu IFTTT Features

Multi‑dimensional user relationship perception: richer relationship metadata enables better user profiling and targeting.

Real‑time bidirectional interaction: supports both buyer‑to‑seller and seller‑to‑buyer relationships with low latency.

Technical Solution

The system follows the IFTTT specification and is modeled into three layers: Channel, Trigger, and Action.

Channel Layer

Manages and stores user relationship data, defining metadata such as relationship type, source account, and target account. It serves as the foundation for both Trigger and Action processing.

Trigger Layer

Custom business events (e.g., new item follow, price drop, lottery draw) that, when fired, compute a user list based on the configured relationship type.

Action Layer

Processes the computed user list, delivering push notifications, benefits, or other custom logic. Actions are plug‑in components that support A/B testing and rapid experimentation.

Fast Scene Integration

Instead of invasive AOP hooks, the solution captures massive network request logs via SLS, feeds them into a Blink streaming job, and applies dynamic rules from Diamond to filter and format data before sending it to the Channel layer, achieving low‑latency, decoupled integration.

User List Calculation

Implemented with a Chain‑of‑Responsibility pattern, allowing each Trigger scenario to plug in its own filtering logic without affecting others.

PushAction Details

Sensitive user filtering

Fatigue level verification (user‑level, business‑level, target‑level)

A/B testing of recipient groups

Message assembly

Logging each Action node to SLS for troubleshooting

Collecting send and click metrics for business decisions

User Relationship Storage

The relationship data reaches terabyte scale with peak TPS/QPS over ten thousand. Alibaba’s internally developed Lindorm (a high‑performance KV store based on HBase) was chosen, delivering read QPS of 70 000 and write TPS exceeding 100 000.

Results

Since launch, Xianyu IFTTT supports multiple scenarios (e.g., follow‑new, price‑drop, rental listings), handling billions of relationship events daily, millions of Triggers, and hundreds of millions of Actions, with click‑through rates more than double those of offline pushes.

Future work includes abstracting higher‑dimensional recipes to turn Xianyu IFTTT into a full‑featured workflow orchestration platform.