How Huolala Scaled Elasticsearch to 40B Records with Serverless Cloud Architecture

Under the wave of digitalization and intelligence, efficient retrieval and real‑time analysis of massive data have become a core competitive edge for many industries. Huolala, a global smart logistics and intra‑city freight platform covering 14+ markets and serving tens of millions of users, continuously innovates its technology foundation.

On September 26 at the Yunqi Conference AI Search and Vector Engine forum, Chen Minhua, Huolala’s Elasticsearch technical lead, shared deep practices of Elasticsearch in a global high‑concurrency scenario and the significant gains after migrating to Alibaba Cloud Elasticsearch Serverless, offering a replicable technical reference for the industry.

Huolala Business and Elasticsearch Overview

Huolala’s business spans over 14 core markets and more than 360 cities in mainland China, connecting over 1.2 million active drivers and 14 million active users. Leveraging multi‑language development (Java, PHP, Golang, Python, C++) and multi‑cloud infrastructure (Alibaba Cloud, etc.), Huolala supports diverse scenarios such as freight, corporate logistics, moving, cold chain, LTL, vehicle rental, and after‑market services. Elasticsearch is deeply applied in the following areas:

Huolala ES Use Cases

Core Business: High‑efficiency operation, data query and intelligent scenarios. In high‑concurrency modules like the order‑grabbing hall, the system handles massive requests stably.

Log Collection: Using an ELK cluster for log collection, processing and visualization, greatly improving operational efficiency.

Multidimensional Query: In the customer‑service system, supports cross‑condition multidimensional queries for fast data analysis and output.

AI Scenarios: Provides robust search and data processing for intelligent客服, image recognition, and knowledge‑base retrieval, driving comprehensive AI upgrades.

Huolala Elasticsearch Cluster Characteristics

Large Scale: Over 15 000 CPU cores, more than 4 billion records, total data exceeding 4 PB.

High Concurrency: Peak QPS surpasses 10 million, supporting 300+ concurrent business applications.

Strong Real‑time: Average response time 24 ms, maintaining sub‑10 ms speed in core high‑concurrency scenarios.

To ensure global continuity and high‑performance data retrieval, Huolala adopts a multi‑AZ architecture, improving resource utilization, system stability, and providing a solid foundation for scaling and flexible deployment.

Multi‑AZ Architecture

Multi‑Availability‑Zone Deployment: Distributes nodes across different geographic zones to avoid single‑point failures, enhance fault tolerance, and support load balancing.

Hot‑Warm‑Cold Separation: Classifies data by access frequency into hot, warm, and cold tiers, storing them on high‑performance or low‑cost nodes to optimize resource usage, reduce storage cost, and improve scalability.

Cluster Management Platform

Monitoring Platform: Real‑time request, resource, inspection alarm, and abnormal log monitoring for visualized operation and instant alerts.

Emergency Platform: Supports automatic SQL killing, one‑click cluster scaling, and emergency disk expansion.

Drill Platform: Conducts AZ failure drills, fault injection tests, and emergency rehearsals to improve disaster‑recovery response.

Change Platform: Provides resource request, configuration change, and task management for controlled and safe deployments.

Governance Platform: Optimizes slow SQL, index governance, and resource water‑level governance to enhance cluster performance and resource efficiency.

Benefits of Migrating to Alibaba Cloud Elasticsearch Serverless

Traditional scaling often leads to resource waste, low efficiency, and high risk. Serverless architecture dramatically reduces cost while delivering high availability, elasticity, and low‑maintenance traffic handling.

Serverless Strategies for Different Traffic Growth Types

Daily Peaks and Valleys: Traditional: purchase resources based on peak, causing waste. Serverless: retain minimal resources for daily load, elastically scale during peaks, avoiding idle waste.

Expected Traffic Growth: Traditional: manual scaling during low‑peak periods, cumbersome and time‑consuming. Serverless: automatically increase quota before events, achieving minute‑level scaling without service disruption.

Unexpected Traffic Surge: Traditional: emergency throttling and temporary scaling, risking user experience. Serverless: automatic scaling for small surges and automatic throttling for large spikes, ensuring stable operation.

After migrating the ELK log cluster to Alibaba Cloud Elasticsearch Serverless, Huolala achieved a 50 % improvement in operational efficiency and a 60 % reduction in resource costs, with elastic write resources scaling from 120 CU to 150 CU during peaks and shrinking to 12‑18 CU during troughs.

Future Outlook

Huolala will continue deepening AI and Elasticsearch integration: AI‑driven intelligent scheduling and precise matching on the business side to boost transportation and operation efficiency; AI‑powered inspection, prediction, and automated emergency response on the operations side for higher system stability and security; and intelligent query conversion and technical assistant capabilities on the R&D side to improve developer experience and accelerate product iteration, driving a shift from data‑driven to intelligence‑driven operations.