How Tencent News Cut PUSH Platform Code by 87% and Boosted Performance 3.5×
The article details how Tencent News' PUSH platform was re‑architected—consolidating modules, unifying the tech stack to Go, building an in‑house message channel, and introducing batch IO and priority scheduling—resulting in a 70% cost cut, 3.5‑fold throughput increase, and dramatically lower latency.
Introduction
Tencent News PUSH platform distributes premium news, serving as a key user activation channel. The team achieved major technical gains: reducing code from 680k lines to 86k, rewriting most C++ modules in Go, and addressing over‑micro‑service issues.
Old Architecture Issues
Excessively long module chain (up to 18 modules, 17 RPC hops) causing latency.
Service dependencies bottlenecked by slow audience‑package fetching.
Poor fault tolerance and lack of automatic failover.
No priority distinction between manual and automated pushes.
Mixed C++ and Go stacks hindered code reuse.
Low testing efficiency and long lead‑time for changes.
Optimization Solutions
Full‑link Business Closure
Built an in‑house message channel, merged 15 modules into 6, cut code to 86k lines, and unified interfaces, raising registration success from 90% to 99.9%.
Unified Technology Stack
Rewrote all non‑recommendation modules in Go, eliminating C++ components.
Link Consolidation
Combined trigger and scheduling modules, reduced RPC hops from 17 to 2, and streamlined data flow.
Number‑Package Service
Implemented a custom paging service for audience packages, removing the external bottleneck.
Offline Pre‑Filtering
Split audience packages by system/brand, moving filtering offline to cut online latency.
Batch IO Aggregation
Wrapped single IO calls into asynchronous queues, enabling batch processing and higher throughput.
Priority Scheduling
Introduced task priority queues and user‑level ordering to ensure hot news reaches users first.
Automatic Fault Recovery
Added backup nodes for sharding and LRU cache monitoring to automatically reroute traffic on node failure.
Automated Testing
Created regression and diff testing pipelines that replay recorded traffic with captured dependency data.
Results
Operational cost reduced by ~70%.
Push throughput increased 3.5×.
Hot‑push latency (P90) dropped 90%.
Click PV for hot pushes rose 10% and overall UV improved.
Zero incidents after February 2025.
Tencent Cloud Developer
Official Tencent Cloud community account that brings together developers, shares practical tech insights, and fosters an influential tech exchange community.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.