Pinpoint

What it is

Pinpoint is an open-source APM (application performance management) tool built for tracing requests across distributed systems, with particular strength in Java and JVM-based stacks (Spring, Java EE, etc.), though it also supports PHP, Python, and a growing list of other agents. It works via bytecode instrumentation — no code changes required — to capture call stacks, response times, and inter-service dependency maps. Its signature feature is a topology map showing service-to-service call graphs with real-time throughput, paired with a code-level call stack view down to SQL query text and exception traces. This is transaction tracing, not metrics/logs aggregation — think Jaeger/Zipkin territory but with much deeper Java-specific instrumentation out of the box.

Who builds it and why

Pinpoint originated at NAVER, South Korea's dominant search engine and internet services company, to solve their own internal problem: tracing requests across thousands of JVM services at scale without instrumenting application code manually. It's been open source since inception. The project is still primarily maintained by NAVER engineers, with 167 outside contributors over its lifetime — a healthy but not massive external contributor base relative to its star count, which suggests it's used widely but contributed to by a smaller core of committed maintainers rather than a broad community-driven governance model. Not CNCF-hosted; maturity status not publicly available, meaning it sits outside the CNCF graduation pipeline entirely and isn't backed by that governance structure.

Production readiness signal

13,838 GitHub stars and a commit as recent as mid-2026 indicate active maintenance, not an abandoned project. 167 contributors is a reasonable contributor base for a specialized APM tool. Apache-2.0 licensing removes legal friction for commercial use. Latest release v3.1.0 suggests a mature, incrementally-versioned project rather than one still finding its architecture. Java as the core implementation language is consistent with its target audience — expect the agent runtime and instrumentation layer to feel native to JVM operators, but also expect the usual bytecode-instrumentation caveats: agent version compatibility with your JVM/framework versions matters, and upgrades require care.

Who should use this

Teams running large JVM-heavy microservice fleets (Spring Boot, Java EE) who need deep call-stack-level tracing without instrumenting every service manually. If you have hundreds of Java services and need to see exactly which downstream call, SQL query, or exception is causing latency, Pinpoint's out-of-the-box bytecode instrumentation saves real engineering time versus manually wiring OpenTelemetry SDKs everywhere. Good fit if you're comfortable running and scaling a dedicated collector + HBase storage backend, and if Java is your dominant language.

Who should NOT use this

If your stack is polyglot with Java as a minority language, Pinpoint's agent support outside JVM (PHP, Python, others) is thinner and less mature than its Java instrumentation — you'll get a second-class experience. If you're already standardized on OpenTelemetry for vendor-neutral instrumentation, adopting Pinpoint means a parallel, Pinpoint-specific agent model that doesn't interoperate cleanly with OTel pipelines. Small teams or small service counts shouldn't take this on — Pinpoint requires operating HBase as a storage backend, which is real operational overhead for tracing data you could get more simply from Tempo or Jaeger with a lighter storage engine. If you need unified metrics/logs/traces in one pane of glass, Pinpoint is traces-only; you'll still need a separate metrics and logging stack.

Alternatives

• Jaeger — CNCF-graduated, OpenTelemetry-native, broader polyglot language support, lighter storage options (Cassandra, Elasticsearch, or in-memory).
• SkyWalking — Also CNCF, similar Java-first APM heritage (originated in China), includes built-in metrics and alerting alongside tracing.
• Datadog APM — Commercial, zero self-hosted infrastructure, better polyglot parity, but ongoing per-host licensing cost instead of self-managed HBase.

Pricing

Fully open source under Apache-2.0. No paid tier, no vendor. You own the entire operational cost: running the collector, web UI, and HBase storage cluster.