How QQ Space Photo Album Handled a 4‑Fold Traffic Surge on New Year’s Day

Preface

On the first day of the 2017 New Year holiday, many users began sharing their 18‑year‑old photos on QQ Space, creating a nostalgic surge that flooded the album service with unprecedented traffic.

Business Data Review

Key metrics showed a 4× peak in image download volume, a 4× increase in upload volume, and a 12× spike in image‑with‑text posts, with over 70% of the traffic targeting rarely accessed “cold” images.

Business Architecture Analysis

The album system consists of an upload pipeline (proxy → logic for sharding, permissions, caching → storage integration → persistent storage) and a download pipeline where the image adaptation module selects the optimal image size based on request context and returns the appropriate URL.

Daily Operations Work

The SNG operations team routinely performs capacity management, including link tracing, traffic capture, and full‑link data aggregation to identify bottlenecks.

Capture link modules via packet tracing.

Determine call chains from device‑reported data.

Obtain call‑chain data from naming services.

Summarize full‑link data to map dependencies.

Capacity Emergency Measures

Standard testing cannot reveal extreme cold‑data access patterns like the “18‑year‑old photo” event, so a set of emergency mechanisms was introduced:

Monitoring and elastic capacity: IaaS‑level metrics (CPU, traffic) trigger automatic scaling when anomalies appear.

WeChat SNG’s “Weaving Cloud” concept organizes services into packages, configurations, permissions, and test tools, enabling rapid provisioning of thousands of machines.

Automation pipeline: standardized, configurable, and automated deployment reduces reliance on manual scripts and documentation.

Flexible Business Architecture

To cope with the sudden load while preserving core user experience, several flexible strategies were applied:

Optimized image indexing to reduce storage pressure, increasing batch size by 3× and cutting interaction count by two‑thirds.

Added local disk cache to the upload path, allowing images to be cached locally before being flushed to backend storage.

Adjusted image adaptation to serve smaller versions for large‑image requests, lowering bandwidth consumption.

Skipped album validity checks during upload to reduce index lookups.

Enabled overload protection: when a node’s CPU exceeds 80%, excess requests are dropped to prevent cascade failures.

Temporarily disabled non‑core features (e.g., face‑centered cropping, delete‑mark checks) to free resources.

Cross‑region traffic scheduling: users were redirected to less‑loaded regions.

Summary

The “18‑year‑old photo” flashback event demonstrated that rapid, massive user activity leaves very little reaction time for operations; only a disciplined, standardized operations practice and a robust, automated infrastructure can handle such spikes gracefully.

Afterword

Future plans include intelligent capacity‑based scheduling, a resource‑hosting platform, and an automated rehearsal system to further improve resilience.