What Tencent Cloud’s Outage Reveals About IaaS vs PaaS Reliability

1. Observed Failure Symptoms

The outage manifested as a collapse of the API system, causing interruption of many PaaS cloud products such as console, cloud functions, micro‑services, OCR, captcha, etc., while data‑plane services like running VMs, VPC, and cloud disks remained unaffected. Independent API‑based object storage and CDN streaming were also not impacted.

2. Actual Scope Was Limited

IaaS products (cloud hosts, containers, disks, VPC) were not affected because they do not rely on the failed API.

Although control‑plane functions for IaaS were disrupted, the outage occurred between 15:20 and 16:00 (extending to 17:00 for a Shanghai node), a period when customers rarely perform large‑scale scaling.

CDN could bypass most authentication failures, and large video customers with pre‑authorized quotas were unaffected.

The most visible impact was on the console and API system, causing user alarm and false‑positive monitoring alerts.

3. Evidence Supporting the PaaS Classification

The author’s upcoming book defines IaaS by specifications and capacity limits, while PaaS is measured by software‑recognizable user‑action counts. When an API system crashes, IaaS only loses control capabilities, whereas every step of a PaaS workflow depends on the API, leading to starkly different failure manifestations.

4. Customers Should Adopt Multi‑Cloud Redundancy

Since no cloud product is fault‑free, technical teams must design redundancy and rapid‑switch plans before failures occur. IaaS can use availability zones for isolation, but PaaS lacks such concepts, forcing customers to rely on multi‑cloud strategies. Monitoring information is richer for IaaS, while PaaS exposes only simple API endpoints, making reliability assessment difficult.

5. Sensational News Adds No Value

Hype‑filled articles about cloud outages provide little technical insight; they often repeat empty excitement without clear description of phenomena, leading readers to misjudge the underlying issues.

6. Old Joke: Service Status Pages

Criticism of missing health‑status pages overlooks that each product line already offers its own API status endpoint. Adding a unified status page would increase complexity without clear benefit, as many customers do not rely on such a page.

7. Old Joke: System Disk Data Loss

Historical incidents of system‑disk data loss are cited without relevance to current customers; the real concern is the lack of transparent incident details that allow technical verification.