Perforator

Documentation | Post on Medium | Post on Habr

Perforator is a production-ready, open-source Continuous Profiling app that can collect CPU profiles from your production without affecting its performance, made by Yandex and inspired by Google-Wide Profiling. Perforator is deployed on tens of thousands of servers in Yandex and already has helped many developers to fix performance issues in their services.

Main features

• Efficient and high-quality collection of kernel + userspace stacks via eBPF.
• Scalable storage for storing profiles and binaries.
• Support of unwinding without frame pointers and debug symbols on host.
• Convenient query language and UI to inspect CPU usage of applications via flamegraphs.
• Support for C++, C, Go, and Rust, with experimental support for Java and Python.
• Generation of sPGO profiles for building applications with Profile Guided Optimization (PGO) via AutoFDO.

Minimal system requirements

Perforator runs on x86 64-bit Linux platforms consuming 512Mb of RAM (more on very large hosts with many CPUs) and <1% of host CPUs.

Quick start

You can profile your laptop using local perforator record CLI command.

You can also deploy Perforator on playground/production Kubernetes cluster using our Helm chart.

How to build

• Instructions on how to build from source are located here.

• If you want to use prebuilt binaries, you can find them here.

How to Contribute

We are welcome to contributions! The contributor's guide provides more details on how to get started as a contributor.

License

This project is licensed under the Apache 2.0 License (Apache-2.0). Apache 2.0 License

Some parts are licensed under the GPL 2.0 license. GPL 2.0 These include:

• eBPF program source code (perforator/agent/collector/progs/unwinder)
• OpenJDK profiling support (portions of code within perforator/internal/linguist/jvm)