How Leading Chinese Companies Scale Elasticsearch for Billions of Orders

Many Chinese companies—including JD.com, Didi, Toutiao, Ele.me, 360 Security, Xiaomi, and Vivo—use Elasticsearch for search and large‑scale near‑real‑time analytics with Kibana, Logstash, and Beats.

1. JD.com Order Center Elasticsearch Evolution

The JD.com order center stores order data in MySQL but uses Elasticsearch to offload heavy read traffic and support complex queries. The ES cluster now holds over 1 billion documents with daily query volume of 500 million, employing a real‑time active‑standby architecture for high read/write stability.

The cluster uses a VIP load balancer, a gateway layer of ES client nodes acting as smart request distributors, and a data‑node layer for storage and processing. A primary shard with two replicas (one primary, two replicas) balances requests via round‑robin, and scaling is achieved by adding replica sets.

Shard count is tuned to balance single‑ID lookup throughput and aggregation pagination performance, based on extensive load testing.

Older orders are archived to a historical database after a configurable retention period.

2. Ctrip Elasticsearch Use Cases

1. Hotel Order Elasticsearch Implementation – Real‑time indexing of sharded databases into a dedicated web service improves query convenience while maintaining performance.

2. Flight Ticket Elasticsearch Cluster Operations – Data flows from Kafka through ETL pipelines, storing cold data in HDFS and hot/ warm data in databases and caches.

3. Large‑Scale Elasticsearch Cluster Management – The largest log cluster runs 120 data nodes on 70 servers, handling 60 billion daily index records (25 TB new index files, 50 TB with replicas), peak indexing rates of millions per second, and a total data volume of ~1 PB across 3 441 indices and 17 000 shards.

3. Qunar Order Center Solution

Qunar’s hotel orders grew from 30 k to 1 M daily, exceeding MySQL sharding limits. Elasticsearch was introduced to store searchable fields while the DB kept detailed order data, enabling complex queries across multiple dimensions.

Current setup: 8 primary shards per index, 1.4 billion documents (≈2 × 10⁸) occupying 64 GB, with cluster disks of 240 GB.

4. 58.com Information Security Department

The Elastic Stack is deployed for security data ingestion, storage selection, performance tuning of master and data nodes, high‑throughput low‑latency search, and Kibana visualizations for operations and product teams.

5. Didi Elasticsearch Multi‑Cluster Architecture

Didi built an Elasticsearch platform with over 3 500 instances, >5 PB data, and peak write throughput exceeding 20 million TPS. Use cases include map search, customer service analytics, and log services.

Sink Service – Kafka streams (business logs, MySQL binlog, custom reports) are consumed and written to Elasticsearch, providing write protection and multi‑cluster disaster recovery.

Gateway Service – All queries pass through a gateway that offers HTTP REST, TCP, and SQL interfaces, with features such as access control, rate limiting, DSL throttling, and multi‑cluster failover.

6. Practical Order Search Solution

Elasticsearch’s support for structured queries and real‑time updates addresses pain points of traditional order reporting. The architecture separates searchable fields in ES from detailed order data in the DB, exposing a unified service API for front‑ and back‑end applications and reporting tools.