Infrastructure Health and Smoke Tests
Relevant source files
The following files were used as context for generating this wiki page:
Infrastructure health and smoke tests verify the integrity of the Flyhard runtime environment. They ensure that all dependencies, hardware accelerators, and graphics subsystems function correctly before starting neural model training or simulation. The system validates the GPU environment, library versions, and the ability to render simulation frames across multiple backends.
Sources: README.md:16-25, scripts/runtime_health.py
Runtime Environment Validation
The runtime validation process checks for the presence and correct versioning of critical software packages and hardware drivers. It specifically verifies the GPU availability via PyTorch and checks the installation status of specialized libraries like CARLA, MuJoCo, and FlyGym.
Dependency and Version Tracking
A build receipt tracks the exact versions of all packages in the runtime. The system maintains SHA-256 hashes for every source file and environment configuration to ensure reproducibility.
The flow shows the sequential verification of hardware and software components leading to a build receipt.
Sources: reports/2026-09-09-runtime/build-receipt.json:1-15, scripts/runtime_health.py
Key Package Requirements
| Package | Version | Purpose |
|---|---|---|
torch | 2.8.0+cu128 | Neural connectome policy execution |
carla | 0.9.16 | Vehicle simulation and world environment |
mujoco | 3.9.0 | Body and contact physics |
flygym | 2.1.0 | NeuroMechFly anatomy and simulation helpers |
pyvista | 0.49.0 | 3D visualization and mesh processing |
Sources: reports/2026-09-09-runtime/build-receipt.json:4-12
Graphics and Rendering Smoke Tests
Smoke tests verify that the offscreen rendering pipelines are functional. This is critical for capturing simulation footage and neural activity visualizations. The system tests two primary rendering paths: MuJoCo for body physics and VTK/PyVista for neural connectome geometry.
Rendering Verification Flow
The graphics smoke test executes minimal rendering tasks to produce artifact images, which are then hashed and verified.
This diagram illustrates the parallel testing of physics and geometry rendering backends.
Sources: scripts/runtime_graphics_smoke.py, reports/2026-09-09-runtime/artifact-manifest.json:40-57
Artifact Generation
The smoke tests produce specific files to confirm successful execution. These files are logged in the artifact-manifest.json with their byte sizes and SHA-256 signatures.
- MuJoCo Artifacts: Produces
mujoco.jsonandmujoco.pngto verify physics-based image capture. - VTK Artifacts: Produces
vtk.jsonandvtk.pngto verify connectivity geometry rendering.
Sources: reports/2026-09-09-runtime/artifact-manifest.json:40-57
Build and Deployment Status
The infrastructure includes automated build tracking through GitHub Actions. Build metadata records the status of compressed layers and the results of health checks.
Build Metadata Summary
| Metric | Value / Source |
|---|---|
| Build Status | success |
| Tests Passed | 8 |
| Compressed Size | 19.56 GB |
| Base Image | runpod/pytorch@sha256:4d1721... |
| Cached Layers | CARLA, apt graphics, Python env |
Sources: reports/2026-09-09-runtime/build.json:8-18, reports/2026-09-09-runtime/build-receipt.json:14
Runtime Health Execution
The health check identifies the underlying platform and ensures that non-finite physical states are not present during a baseline simulation run. It asserts that simulation timesteps remain consistent (e.g., 1e-4 seconds).
Sources: scripts/body_baseline.py:65-71, scripts/runtime_health.py
Infrastructure tests ensure the Flyhard environment is stable and reproducible. By validating GPU accessibility, package versions, and rendering pipelines, the system prevents failures in more complex simulation and training stages.