Add Raspberry Pi 5 performance testing environment setup

[Copilot]

Bootstraps a Raspberry Pi 5 as a self-hosted GitHub Actions performance node: Docker-containerised benchmarks, a custom HAT for GP2040-CE input-latency measurement, and workflows that publish results back to GitHub.

Setup Scripts (scripts/)

• setup.sh — orchestrator; accepts --skip-docker, --skip-runner, --skip-hat, --runner-url, --runner-token
• setup-system.sh — OS packages, CPU governor → performance, swap off, sysctl network tuning, NTP, hostname
• setup-docker.sh — Docker Engine via official Debian repo, daemon.json with overlay2 + BuildKit, adds user to docker group
• setup-github-runner.sh — github-runner system user, ARM64 runner binary, systemd unit with labels self-hosted,Linux,ARM64,raspberry-pi-5,performancenode
• setup-hat.sh — Python venv (RPi.GPIO, gpiozero, lgpio), udev rules for GPIO and GP2040-CE USB (VID 0x2E8A), dtparam overlays in /boot/firmware/config.txt

Docker Benchmark Containers (docker/)

Multi-stage debian:bookworm-slim image with separate targets for wrk, iperf3, stress-ng, and sysbench. Each entrypoint emits a structured JSON result file. Compose profiles (http, network, stress, cpu, memory, io) enable targeted local runs.

GP2040-CE Latency HAT (hat/)

latency_test.py drives GPIO 17 to trigger a button press on the connected GP2040-CE device, waits for the round-trip signal on GPIO 4, and collects configurable samples (default 1 000 + 50 warmup). Outputs min/max/mean/median/stdev/p95/p99 as JSON.

# Core measurement loop
GPIO.output(trigger_pin, GPIO.HIGH)
t0 = time.perf_counter()
while (time.perf_counter() - t0) < timeout_sec:
    if GPIO.input(signal_pin) == GPIO.HIGH:
        latency_ms = (time.perf_counter() - t0) * 1000.0
        break

Pin config lives in hat/config/hat-config.json (BCM 4 = signal in, 17 = trigger out, 27 = status LED).

GitHub Actions Workflows (.github/workflows/)

Workflow	Schedule	Runner label
performance-test.yml	Daily 02:00 UTC	performancenode
latency-test.yml	Daily 03:00 UTC	performancenode

Both support workflow_dispatch for manual runs with parameter overrides. Results uploaded as artifacts (benchmarks 90 days, latency 365 days) with a markdown stats table written to the run summary. Explicit permissions: contents: read / actions: write blocks on all jobs.

Original prompt

This will houses the setup scripts for a Raspberry Pi 5 to run as a dedicated performance testing environment for applications inside docker containers, as well as having a custom HAT that will allow for latency testing of the GP2040-CE project. All of these performance tests will be run as GitHub Actions on the local Pi, reporting their results on GitHub.