How Tencent Meeting Scaled Rapidly with Cloud‑Native TKE: Architecture & Practices

Background

Due to the COVID‑19 pandemic, online office applications surged. Tencent Meeting, an enterprise‑grade video‑conferencing product, experienced explosive user growth and needed fast, reliable version upgrades. In 40 days it delivered 14 major releases, illustrating a rapid‑iteration model.

Why Cloud‑Native TKE?

Tencent Cloud Container Service (TKE) offers a stable, secure, high‑performance Kubernetes platform with extensions such as cloud disks, load balancers, and custom operators. It solves environment‑consistency challenges, simplifies large‑scale cluster management, and reduces cost.

Solution Overview

Using TKE, the team built a management platform that adds dynamic routing, fixed networking, elastic scaling, and controllable upgrades. The platform runs on Tencent Cloud CVM, CBS, CLB, VPC and extends TKE via operators for routing (L5‑controller), network (IPAMD), resource management, and auto‑scaling.

Key Technical Areas

1. Quality Assurance

Ensuring service stability through dynamic routing, anomaly detection, parallel scaling, and business migration.

1.1 Dynamic Routing

Implemented via a custom L5‑controller that watches service changes, triggers callbacks, and updates routes. A primary fast path guarantees second‑level propagation, while a secondary consistency check reconciles routes every minute.

1.2 Health Checks

livenessProbe

readinessProbe

postStart

preStop

terminationGracePeriodSeconds

These probes isolate unhealthy pods and protect traffic.

1.3 Parallel Scaling

Extended HPA to run detection at high concurrency and allow custom calculation periods down to seconds, enabling immediate scaling under burst load.

1.4 Business Migration

Helm deployment

Namespace packaging and copy

Namespace reuse with custom adjustments

2. Efficiency

Automation of cluster creation, environment initialization, and console integration; CI/CD pipelines across multiple networks; centralized CMDB‑based authorization; and enhanced elastic scaling.

2.1 Process Automation

Cluster provisioning via API

Environment setup scripts

Console integration for batch operations

2.2 CI/CD

All internal services, including Tencent Meeting, use a unified CI/CD workflow that supports cross‑environment deployments.

2.3 Authorization Management

CMDB controller synchronizes pod‑product relationships, enabling automatic credential provisioning.

2.4 Elastic Scaling

Multi‑metric support (Metrics Server, Prometheus, custom)

Parallel workers for real‑time scaling

Customizable thresholds and periods

3. Controlled Upgrade

Four extensions ensure reliable upgrades: fixed networking, lightweight upgrade, batch rollout, and capacity guarantees.

3.1 Fixed Networking

Pods retain IP after recreation, supporting IP‑based authentication and whitelist scenarios. Implemented via IPAMD Controller, TKE‑ENI‑AGENT, and CNI.

3.2 StatefulsetPlus Operator

Custom workload extending StatefulSet with fixed IP, gray‑scale batch updates, and in‑place upgrades.

3.3 Batch Upgrade

Operators allow pre‑defined batches, manual or automatic progression, rollback, and per‑batch health checks, enabling controlled, region‑aware rollouts.

Conclusion

By leveraging cloud‑native technologies on TKE, Tencent Meeting achieved rapid, reliable version iteration and massive scaling, demonstrating the critical role of container platforms in supporting billions of users.