Welcome to the documentation for Prodfiler, the world's first publicly available distributed lightweight continuous whole-system profiler.

In these pages, we strive to provide you with all the information you need to use Prodfiler. If you find there is anything missing, or have a question not addressed here, please open an issue and let us know -- we will try to fix it ASAP.

To ask any questions related to Prodfiler please head over to our community forum.

• Getting Started
• Filtering
• Deploying on Kubernetes, ECS and Nomad
• Feature Reference
• FAQ

Prodfiler is the world's first distributed lightweight continuous whole-system profiler. This is a bit of a mouthful; in practice this means that Prodfiler can answer the question:

Across a fleet of hundreds or thousands of machines, which lines of code are actually eating how much of the overall CPU time? How does this change over time?

Prodfiler is inspired by software available inside some Hyperscalers, most notably Google and Facebook. Google has an internal system called 'Google-Wide Profiling', which has led to very significant savings, a culture of performance awareness/improvement, and to long-term developments such as SwissTables and even the various hardware offload designs. Likewise, Facebook runs a system, which is described in this talk from the SPARK AI Summit.

Both Google and Facebook achieve this capability by recompiling all dependencies from scratch and generally do not make this capability available to the outside world.

Prodfiler is designed to bring the same capability to everybody running a Linux Kernel with version 4.15 or higher.

Continuous in-production profiling can provide the following benefits:

• Make your services faster
• Reduce your cloud bill (since performance improvements mean fewer instances needed)
• Identify and debug performance regressions
• Help identify parts of your software that scale poorly
• Find out about system daemons and other components that are consuming CPU unexpectedly

As a side effect, you can reduce your CO2 footprint, since savings in CPU consumption translate pretty directly into savings in power consumption.

Prodfiler deployment is easy: You sign up and create a project; once that is done, you receive a command line with which you can launch the agent (either manually on a single machine, or via HELM charts on a K8s cluster, or on a group of ECS nodes, etc.).

Please check the getting started guide.

You can find a detailed overview of all features in our Prodfiler feature reference.

We have a detailed FAQ which can answer many of the frequently-asked questions.

If you encounter any problems please contact us via our community forum.

• Linux x86_64 with kernel version >= 4.15