Why Observability Is the Next Frontier for Serverless Computing

Serverless is rapidly becoming the default programming paradigm for cloud applications, with many Alibaba Group services and external enterprises migrating workloads to Function Compute. While it reduces operational overhead and costs, developers face new observability challenges.

1. Serverless as the Future Cloud Paradigm

Adoption is growing across Alibaba (Taobao, Fliggy, Xianyu, AMap, Yuque) and external companies (Weibo, Century Lianhua, Shimo Docs, TPLink, Lanmo Cloud Classroom). Benefits include up to 60% IT cost reduction and faster feature delivery.

2. Observability Challenges

Serverless architectures introduce several pain points:

Component Distribution: Applications span multiple cloud services, making end‑to‑end latency tracing complex.

Scheduling Black Box: Automatic scaling and cold starts are opaque, leaving developers uncertain about latency sources.

Execution Environment Black Box: No direct access to the underlying VM prevents traditional debugging.

Non‑standard Products: Lack of probe installation and incompatibility with open‑source monitoring tools hinder investigation.

Function Compute (FC) has responded with a series of observability enhancements.

3. Observability 1.0 – Basic Logs and Metrics

FC initially offered function logs (via SLS) and basic metrics (invocation count, errors, latency, memory) that serve as a simple “stethoscope” for applications.

4. Observability 2.0 – Cloud‑Native Capabilities

To meet growing demands, FC introduced:

Integration with OpenTracing for distributed tracing, exposing cold‑start times and end‑to‑end request latency.

Built‑in ARMS APM via a single environment variable, providing CPU, memory, JVM, profiling, and SQL metrics without code changes.

FCInsights request‑level metrics that capture per‑request latency, memory usage, error type, cold‑start flag, and trace IDs.

A unified Monitoring Center that aggregates metrics, logs, and traces, supporting multi‑dimensional views (region, service, function, qualifier, request) and full‑stack root‑cause analysis.

Extended programming model with RuntimeLifeCycle hooks, enabling third‑party APM integration (e.g., Grafana, New Relic, Tingyun) by emitting custom metrics before instance pause or recycle.

These features make FC the first FaaS provider to fully embrace open‑source observability standards and support seamless migration between container and serverless environments.

5. Summary and Future Roadmap

FC’s observability has progressed from a closed, basic setup (1.0) to an open, cloud‑native ecosystem (2.0), moving toward “white‑box” serverless. Future plans include adding alerting, instance‑level metrics, Prometheus and OpenTelemetry integration, richer metric exposure, and continued support for open‑source dashboards.