Serverless Architecture for Homework Photo Processing at Hangzhou Mingshitang Using Alibaba Cloud Function Compute

Hangzhou Mingshitang, an online education technology company, faced rapidly increasing homework review demand, especially during school holidays, where daily image processing peaked at over one million and required real‑time handling.

The original architecture deployed on Kubernetes suffered from two main issues: limited Kafka consumer concurrency due to partition count and difficulty scaling processing threads, leading to task backlog during peak periods.

To address these challenges, the team migrated the workload to Alibaba Cloud Function Compute (FC), redesigning the processing pipeline to upload homework photos to OSS, store metadata in a database, and send a Kafka message that triggers FC via a Kafka connector.

Key optimizations included breaking the processing logic into finer‑grained functions, replacing Java microservices with TypeScript/Node.js to eliminate long cold‑start times, and enabling single‑instance multi‑concurrency, achieving over 70% resource utilization.

After migration, request latency dropped from 100‑200 ms on Kubernetes to around 50 ms on FC even under peak load, and cold‑start latency stayed under 100 ms, demonstrating high elastic efficiency.

The new serverless solution also simplified operations: built‑in monitoring and logging improved troubleshooting, and the pay‑per‑use model reduced infrastructure costs.

Remaining pain points include fragmented integration between Kafka and FC consoles and the need to smooth CI/CD integration, which the team addressed by adopting Serverless Devs tools.

Looking forward, Mingshitang plans to further explore serverless adoption across more high‑burst services in collaboration with Alibaba Cloud, continuing to leverage technology to enhance education.