Case Study: KMS Game Company’s Cloud‑Native Architecture and Elastic Microservice Deployment on Tencent Cloud

KMS is a Japanese game company that operates game business, digital comics, advertising, and cloud‑solution services. After being a benchmark customer of Microsoft Azure, KMS migrated to Tencent Cloud in 2021 and deployed its first product there, achieving an overall cost reduction of about 50%.

Challenges and Pain Points

KMS’s architecture serves different terminals for each game (Android, iOS, Web, etc.).

At the same time it maintains a unified back‑office management platform and a common foundation.

The backend is divided into battle systems, management systems, user systems, and other modules.

Game traffic exhibits strong peaks and valleys, requiring rapid scaling during high‑traffic periods and scaling‑down during low‑traffic periods to control costs.

These characteristics lead to the core problem of elastic scaling: different microservices need different scaling frequencies and ranges, directly affecting resource consumption and cost.

Architecture Design

The overall design includes a CDN that distributes static files (game resources, installers, images) stored in COS. Separate backend instances are provisioned for each client type; for example, the battle system instance serving iOS users is dedicated to that client group. The management side, which handles user settings and other generic functions, can share a smaller‑scale instance because its QPS is relatively low.

Elastic Scaling

KMS uses Tencent Cloud Elastic Microservice, a serverless PaaS platform for microservice applications. Users only need to configure scaling policies:

Timed scaling : resources are reduced during low‑traffic periods (e.g., mornings) and pre‑scaled up before expected peaks (e.g., evenings).

Metric scaling : automatic scaling based on CPU or memory usage thresholds (e.g., scaling out when usage exceeds 60% and limiting instance count between 1‑50).

Because the service is fully serverless, scaling occurs within seconds. In KMS’s own tests, the elastic microservice delivered 15% higher performance than Azure App Service and reduced game response latency by 50%, while overall cloud costs dropped by more than 50%.

DevOps

KMS adopts a rapid, safe release process supported by a CI/CD pipeline built on GitHub Actions. Automated unit and integration tests run on each commit, followed by image building, uploading, and deployment to the elastic microservice platform.

Batch Release

The elastic microservice platform supports batch upgrades. The system determines the number of batches based on current instance count, and each batch can be executed automatically or manually. Critical releases are performed manually so that after each batch the functionality can be verified before proceeding. If any batch encounters issues, an immediate rollback is possible without affecting the live service.

Other Design Practices

KMS manages four game distribution platforms and four environments using the platform’s environment management features, greatly improving operational efficiency. Nightly, test‑environment resources are automatically zeroed out and re‑enabled in the morning via automated scripts, reducing idle costs. The elastic microservice also performs health checks and automatically restarts unhealthy instances, eliminating the need for manual restarts that previously took about an hour.

Summary

KMS has been running stably on Tencent Cloud for over two years, cutting costs by 50% through thoughtful architecture and cloud‑native services. The company looks forward to deeper collaboration with Tencent Cloud to share more best‑practice experiences.