Dyno A Fuel Orchestrator for Performance Profiling & Benchmarking

• Dyno A Fuel Orchestrator for Performance Profiling & Benchmarking
  • Introduction
  • Requirements
  • Installation
  • Usage
    • Target
    • Forc Path
    • Output Folder
    • Print Output
    • Flamegraph Support
    • Hyperfine Support
    • Database Support
  • CI/CD Mode

Introduction

The dyno application is a performance profiling and benchmarking tool for Fuel Orchestrator (forc).

It is designed to ensure that areas of the Fuel codebase are not regressing in performance, and can provide insight into the specifics of code performance. The tool is primarily designed to be used in a CI/CD pipeline, but can also be run in a headless state on a local machine.

The tool manages dispatching of forc and measures performance frames at a fixed interval. Each frame is comprised of common hardware resource usage consumed by the running forc executable at the time the frame was captured. The tool also takes cues from forc to include extra information in the benchmarking frames (i.e: when certain compilation phases begin and end, generated binary information, etc), allowing for a more narrow and granular degree of inspection.

The tool can optionally use perf or sample (Mac) for stats, inferno for flame graphs and hyperfine for additional sanity checks.

Requirements

Important

The profiler requires the following:

• Rust Language
• Sway Language
• Sway Compiler
• Linux for flamegraph support using perf (Optional)
• Hyperfine (Optional)
• Postgres for database support (Optional)

In order for perf (used for flamegraph) to work properly on linux, the following options must be set in /etc/sysctl.conf:

echo "kernel.perf_event_paranoid=-1" | sudo tee -a /etc/sysctl.conf
echo "kernel.kptr_restrict=0" | sudo tee -a /etc/sysctl.conf
sudo sysctl -p /etc/sysctl.conf

Installation

• Install Rust
• Instal Sway
• Install Hyperfine(Optional)
• Install Docker(Optional)
• Install Postgres(Optional)

Git clone the sway repository and build.

git clone https://github.com/FuelLabs/sway
cd sway
cargo build --release

Git clone the dyno repository.

git clone https://github.com/ourovoros-io/dyno.git
cd dyno

Run the tool by point to the forc binary and a target folder.

cargo r --release -- -t <target_folder> -f <forc_profile_binary>

Usage

====================================================================================================
                         Welcome to the Fuel Dyno v0.1.0
====================================================================================================
Fuel Orchestrator Performance Profiling & Benchmarking

Usage: dyno [OPTIONS] --target <TARGET> --forc-path <FORC_PATH>

Options:
  -t, --target <TARGET>
          A path to a target folder or file to compile
  -f, --forc-path <FORC_PATH>
          The path to the forc binary
  -o, --output-folder <OUTPUT_FOLDER>
          [default: ./benchmarks]
  -p, --print-output
          Enable printing output (Optional)
      --flamegraph
          Flamegraph support (Optional)
      --hyperfine
          Enable hyperfine analysis (Optional)
      --max-iterations <MAX_ITERATIONS>
          Maximum iterations for hyperfine (Optional) [default: 2]
  -d, --database
          Database support (Optional)
  -h, --help
          Print help
  -V, --version
          Print version

---

Target

The target folder of items or item to be compiled by the Sway compiler and profiled by dyno.

---

Forc Path

Forc-path option allows the usage of a specific version of the compiler that is modified.

---

Output Folder

Set the output folder for the benchmarks data as json.

By default the profiler will store the benchmarks under ./benchmarks.

The format of the filenames is forc-version_md5-hash-of-forc_date_time.

---

Print Output

Provides a full print of all the stats generated by the comparison of two benchmark runs.

---

Flamegraph Support

Tip

To get the most detailed output from the flamegraph in release mode we need to enable debug. We need to take under consideration that each flamegraph for each test target generated is about 3 MB.

[profile.release]
debug = true

Warning

Do not forget to remove this as it increases the size of the binary.

Example output

---

Hyperfine Support

Enables hyperfine feature.

---

Database Support

test_with_docker_setup

---

CI/CD Mode

Please follow the instruction in the DynoSite README.md.

DynoSite README.md