Apache SkyWalking

What it is

SkyWalking is an APM (Application Performance Monitoring) and observability platform purpose-built for distributed systems — microservices, cloud-native architectures, and service mesh deployments. It covers distributed tracing, service topology mapping, metrics aggregation, and log correlation in one system. Unlike agent-only tracers, SkyWalking ships its own backend (OAP — Observability Analysis Platform), storage integrations, and UI, so it functions as a full stack rather than just an instrumentation library. It supports multiple language agents (Java, .NET, Node.js, Python, Go, PHP, etc.) plus native OpenTelemetry and Zipkin protocol ingestion, which matters if your environment is polyglot.

Who builds it and why

SkyWalking is an Apache Software Foundation top-level project, originally created by Wu Sheng and open-sourced in 2015 before entering the Apache Incubator. It's maintained by a large, mostly China-based contributor base (many from companies running SkyWalking at scale internally — Tencent, Huawei-adjacent ecosystems, and others), alongside global contributors. 527 contributors and continuous commit activity through 2026 indicate it's not a single-vendor side project — it's a community-governed ASF project with its own release cadence and PMC oversight. The motivation is straightforward: give engineering teams tracing and topology visibility for microservices without vendor lock-in to a commercial APM.

Production readiness signal

24,854 GitHub stars, 527 contributors, and a recent v10.4.0 release with active commits into mid-2026 point to a mature, actively maintained project — this isn't dormant tooling. Apache-2.0 licensing removes legal friction for enterprise adoption. It's been production-hardened at scale by large Chinese internet companies for years, and it's a known quantity in Java-heavy shops specifically (the OAP backend itself is written in Java, and Java agent instrumentation is the most mature path). CNCF maturity level isn't publicly listed here — SkyWalking is an Apache project, not a CNCF one, so don't conflate the two governance models when evaluating trust signals.

Who should use this

Teams running Java-centric or mixed-language microservice architectures who want tracing, metrics, and topology in a single self-hosted system, without paying per-host or per-GB fees to a commercial APM vendor. It's a strong fit if you're already invested in JVM-based services, need service mesh (Istio/Envoy) observability, or want a backend that speaks both its native protocol and OpenTelemetry so you can migrate incrementally. Platform teams with the operational capacity to run and tune a Java backend (OAP) plus a storage layer (Elasticsearch, BanyanDB, or similar) will get the most value.

Who should NOT use this

If you're a small team without dedicated platform/SRE capacity, running your own OAP cluster plus storage backend is real operational overhead — don't take this on if you just need basic tracing for a handful of services; use a managed SaaS APM instead. If your stack is pure OpenTelemetry-native with no interest in a bundled backend, running the OTel Collector plus a dedicated tracing backend (Tempo, Jaeger) gives you more modularity. Shops without Java expertise in-house will find debugging and tuning the OAP backend itself painful, since it's a nontrivial Java application in its own right. And if you need enterprise support contracts, SLAs, or a managed control plane, SkyWalking gives you none of that natively — you're on your own or reliant on third-party vendors.

Alternatives

• Jaeger — CNCF-graduated distributed tracing focused purely on traces, lighter-weight, no bundled metrics/topology UI.
• Grafana Tempo + Mimir + Loki (LGTM stack) — modular OSS observability stack with strong OpenTelemetry alignment, more moving parts to integrate yourself.
• Datadog APM / New Relic — commercial SaaS APMs with less operational burden but recurring per-host/per-GB costs and vendor lock-in.

Pricing

Fully open source under Apache-2.0. No licensing cost, no paid tier, no vendor gate. Your cost is entirely operational: compute for OAP, storage backend costs (Elasticsearch/BanyanDB), and engineering time to run and upgrade it.