Kuaishou’s High‑Availability Strategy for Complex B‑End Systems

In a recent technical conference, Kuaishou senior e‑commerce architect Zhao Qingjie shared the high‑availability practices for the platform’s B‑end services, which serve merchants, suppliers and operation teams rather than end‑consumers.

B‑End vs C‑End: Key Differences

C‑end focuses on performance and user experience for massive concurrent traffic.

B‑end prioritises business complexity, data consistency, permission management and stability.

Typical B‑end characteristics include multi‑role, long‑chain, strong‑consistency requirements, and higher user‑sensitivity to failures.

Challenges Faced by Kuaishou B‑End

Historical architectural debt: rapid feature rollout led to tightly coupled services, over‑generalised modules and an unbounded API layer.

Fault recovery difficulty: low traffic masks issues, but once a problem occurs the impact and user complaints are severe.

Team collaboration friction: multiple teams maintain shared interfaces, raising coordination cost and lacking stability awareness.

Blind spots in fault governance: many long‑tail, chronic failures are hard to detect with traditional monitoring.

Multi‑Dimensional Architecture Governance

Governance and refactoring across service, business, storage and disaster‑recovery layers.

Build a quality‑assurance system tailored to B‑end characteristics, including fault‑prevention, detection, localisation and recovery mechanisms.

Introduce supporting organisational mechanisms, monitoring tools and collaboration processes to embed stability culture in teams.

The governance is a full‑stack, deep‑level "clean‑up" that addresses legacy problems while laying a foundation for future evolution.

First‑Generation Architecture Pitfalls

Unclear service responsibilities due to ad‑hoc feature additions.

Severe API coupling; the API layer became a shared pool without clear boundaries.

Arbitrary data access; multiple services directly share the same database, leading to duplicated logic.

Coarse deployment granularity, causing high release risk and poor isolation.

Strong contagion effect: a failure in one service can cascade to many dependent modules.

These issues turned the B‑end system into a high‑incident zone, dragging down overall platform stability metrics.

Quality‑Assurance and Monitoring体系

The team built a closed‑loop from prevention to discovery, localisation and recovery, and introduced concrete measures such as:

Structured logs to facilitate issue tracing.

Unified e‑commerce error codes.

Full‑domain monitoring dashboards.

Regular disaster‑recovery drills and multi‑region failover designs.

Governance Outcomes

From 2023 to 2024, the incident rate dropped by 82%, demonstrating the effectiveness of the multi‑dimensional governance.

Appropriate architecture governance must start from the current system state, business characteristics, and root causes of problems.

The sharing emphasizes that high availability is not merely an operations task but a collective responsibility that requires systematic design, monitoring, and cultural adoption across teams.