Implementing and Testing a High‑Throughput WeChat Red‑Packet System: 1M Connections and Up to 60k QPS

Introduction: The author was inspired by a 2015 article about handling 10 billion WeChat red‑packet requests and decided to reproduce a similar load on a single machine.

Background: Definitions of QPS, PPS, shake‑red‑packet and send‑red‑packet are given, and the target metrics are outlined: support 1 million connections, peak QPS 60 k, shake‑red‑packet rate 83 /s, send‑red‑packet rate 200 /s.

System capacity calculations: Based on 638 servers and 5.4 hundred million users, a single server should handle about 90 k users and roughly 2.3 k–6 k QPS.

Implementation: The prototype uses Golang 1.8r3 on Ubuntu 12.04 with Dell R2950 servers (8 CPU, 16 GB RAM) and 17 Debian 5.0 VMs as clients (each 6 k connections). Connections are divided into multiple SETs, each managing a few thousand connections, reducing goroutine count and lock contention. A simple NTP‑based algorithm distributes client requests evenly.

Monitoring: Custom Python scripts with ethtool and counters record per‑second request numbers; logs are visualized with gnuplot. Example command used to count connections: Alias ss2=Ss –ant | grep 1025 | grep EST | awk –F: "{print $8}" | sort | uniq –c

Results: The system achieved 1 M connections, sustained 30 k QPS stably and 60 k QPS with some fluctuations due to goroutine scheduling, network latency and packet loss. Graphs show client and server QPS, red‑packet generation and consumption, and Go pprof data.

Conclusion: The prototype meets the design goals of supporting 1 M users and up to 60 k QPS, demonstrating that a single‑machine can simulate a large‑scale WeChat red‑packet service, and the architecture can be horizontally scaled.