How We Scaled an Online Education Platform 20× During COVID: A Technical Playbook

Facing Business Pain Points

The pandemic forced schools to postpone opening, prompting the Ministry of Education to launch a "stop classes, keep learning" policy. To support K12 students, the company quickly opened free online courses nationwide, causing a sudden surge in traffic that required a 20‑times QPS increase.

Three Decisions for Rapid Scaling

1. Online Vertical Upgrade – The core database was initially a modest high‑availability UDB instance supporting up to 3,000 QPS. Anticipating the surge, the team performed a vertical upgrade of memory and disk, boosting performance with only a brief, seconds‑long pause.

2. High‑Performance Read/Write Separation – Because read requests far outnumbered writes (≈50:1), the architecture was changed to a primary‑plus‑multiple‑replica cluster with a read/write‑separating proxy. The proxy forwards write and transactional reads to the primary, while ordinary reads are distributed among replicas, maintaining full MySQL compatibility without code changes.

3. High‑Availability New Architecture: Expanding Connections – When connection counts hit the 6,000‑connection limit of the existing high‑availability UDB, the team adopted a drift‑VIP + dual‑master design, allowing a seamless two‑minute upgrade without data migration and improving network latency and compatibility for MySQL and PostgreSQL.

Operation Situation

From February 1 to 10, daily registrations rose from 8,300 to 30,000, peaking at 58,000. The platform sustained 200,000+ QPS on the read/write‑separating proxy and, after scaling to a 1‑primary‑6‑replica cluster with dual‑proxy, handled the continued growth smoothly, serving over one million students nationwide.

Slow Query Optimization

Using an ORM layer concealed low‑level MySQL details, leading to numerous slow queries under high load. The UDB DBA team collaborated remotely to identify and index problematic tables, eliminating most slow queries and stabilizing performance.

Final Thoughts

This rapid‑response scaling demonstrates effective cooperation between cloud users and providers. The extensive use of UDB’s elastic, fully‑managed services proved essential, and the experience informed future product improvements such as the upcoming fast‑recovery UDB offering.